Research ArticleKD-ACP A Software Framework for Social Computing inEmergency Management
Bin Chen Laobing Zhang Gang Guo and Xiaogang Qiu
Research Center of Computational Experiments and Parallel System Technology College of Information System and ManagementNational University of Defense Technology Changsha 410073 China
Correspondence should be addressed to Bin Chen nudtcb9372gmailcom
Received 4 June 2014 Accepted 23 August 2014
Academic Editor Praveen Agarwal
Copyright copy 2015 Bin Chen et alThis is an open access article distributed under the Creative CommonsAttribution License whichpermits unrestricted use distribution and reproduction in any medium provided the original work is properly cited
This paper addresses the application of a computational theory and related techniques for studying emergencymanagement in socialcomputing We propose a novel software framework called KD-ACPThe framework provides a systematic and automatic platformfor scientists to study the emergency management problems in three aspects modelling the society in emergency scenario as theartificial society investigating the emergency management problems by the repeat computational experiments parallel executionbetween artificial society and the actual societymanaged by the decisions fromcomputational experimentsThe software frameworkis composed of a series of tools These tools are categorized into three parts corresponding to ldquoArdquo ldquoCrdquo and ldquoPrdquo respectively UsingH1N1 epidemic in Beijing city as the case study themodelling and data generating of Beijing city experiments with settings of H1N1and intervention measures and parallel execution by situation tool are implemented by KD-ACPThe results output by the softwareframework shows that the emergency response decisions can be tested to find a more optimal one through the computationalexperiments In the end the advantages of the KD-ACP and the future work are summarized in the conclusion
1 Introduction
Emergency management attracts the attention of scientistsfrom social computing because the whole process of emer-gent events is deeply coupled with human society and theemergency response decisions need an approach to testifytheir effect without the reappearance of emergent events inthe society As a new paradigm of computing and technologydevelopment social computing helps scientists to understandand analyze individual and organizational behavior andfacilitate emergencymanagement research and application inmany aspects [1]
Based on the fruitful development of computationalmethodology on emergency management research over thelast decade lots of work has been done to solve the problemsin society domain Both the conceptual frameworks in mul-tiple discipline and the technological platforms developedfor the domain requirements are more and more popularin the research on emergency management especially theagent-based modelling and simulation [2] The bottom-uptechnique describes the society in microview by modelling
individual behavior communications in agents and evolu-tion rules of agent organizations It is worth notifying thatthe modelling of agent does not emphasize the intelligence ofindividual Large scale communications and the emergencephenomena are the objects of agent-based modelling andsimulation The agent oriented platforms such as Biowar [3]GASM [4] and EpiSims [5] to study emergency problemshave been proposed in many fields Biowar developed byCarnegie Mellon University is used to study the bioattacksin city with the ability of scalable agent modelling GASM(Global-Scale Agent Model) by Epstein simulates a globalH1N1 epidemic with 65 billion people EpiSims from LosAlamos national lab is used to testify the intervention mea-sures in epidemics of smallpox from United States Depart-ment of Health and Human Services
With the help of agent-based modelling simulationtechnique and the concept of artificial society [6] a novel con-ceptual framework based on artificial systems is introducedin the social computing The conceptual framework calledACP (Artificial Society Computational Experiments ParallelExecution) approach is proposed byWang in 2004 [7ndash9] It is
Hindawi Publishing CorporationMathematical Problems in EngineeringVolume 2015 Article ID 915429 27 pageshttpdxdoiorg1011552015915429
2 Mathematical Problems in Engineering
a novel approach in social computing to solve the problemsin society domain ACP approach is categorized into threeaspects representing and modelling society with artificialsystems analysis and evaluation by computational experi-ments and control andmanagement of real society by parallelexecution Under the instruction of conceptual framework ofACP approach a wide spectrum of complex systems such astransportation medicine finance and environment can bestudied in the computational manner Actually many real-world applications using ACP approach have been developedto solve the real problems in domains For instance complexsocioeconomic system [10] and the research framework fore-commerce system [11] are the good applications of ACPapproach in the economic area The ACP-based frame-work for integrative medicine [12] is proposed to solve theproblems in medicine An overall framework of emergencyrescue decision support system of petrochemical plant [13]is proposed based on ACP theory to study environmentrisk accidents of petrochemical plant Parallel BRT operationmanagement system [14] based on ACP approach has beenconstructed to detect the quantity of passengers on stationsreal-time traffic flow on stations or at intersections andqueuing length of vehicles on the road A novel parallel sys-tem for Urban Rail Transportation (URT) [15] based on ACPapproach is proposed to address issues on safety efficiencyand reliability of the operation of URT An artificial powersystem [16] is set up on the models of power systems andcomplex power grids to provide a feasible approach for thecontrol and management of the modern power system
Although a lot of work has been done on the concepts andtheory framework to study problems by social computing thefollowing problems of computational experiments in emer-gencymanagement are still not solved from the perspective ofmodelling for social systems theory and software frameworkof platform implementation
(i) The modelling and simulation of emergency man-agement are not given the special considerationThe representation of society focuses on the genericmodelling of agent (represented by Repast [17 18])The description of environments is too simple tomeet the requirements from research on emergencyproblems such as the building size the place relatedagent contact frequency
(ii) The existing tools and platforms cannot support thedesign of experiment Computational experimentscannot be done systematically and automatically
(iii) The existing applications of ACP-based frameworksare still domain specific A generic workflow andintegrated toolkit are needed to implement ACPapproach especially in the application of emergencymanagement
Therefore it is necessary to develop an ACP-based softwareframework for the research on emergency management Theartificial system is the projection of real world in the emergentscenarioThemodelling of the system including the emergentevents modelling and intervention measures modelling thedesign of computational experiments considering the settings
of emergency parameters the settings of large samples exper-iments and the parallel execution with loose connection ofreal society should be considered inside the software frame-work
As a result the purpose of this paper is to proposea software framework called KD-ACP applying the ACP-based computational theory and corresponding methods instudying emergency management problems KD is short forthe Chinese phonetic alphabets of China National Univer-sity of Defense Technology KD-ACP means the softwareframework is developed by National University of DefenseTechnology to implement ACP approach The remainder ofthis paper is organized as follows Section 2 summarizes theexisting agent-based modelling and simulation platformsSection 3 introduces ACP approach first and proposes theKD-ACP platform Section 4 illustrates the modelling ofBeijing city with KD-ACP both the agent models and initialdata are considered Section 5 shows how to do experimentswith KD-ACP using the H1N1 case study in artificial BeijingIn the end the paper is concluded in Section 6
2 Related Works
There have beenmany efforts on social computing especiallyon the emergency management Agent-based modelling andsimulation are popular in the implementation of socialcomputation The related works are mainly categorized intoSWARM-like agent-based modelling and simulation plat-forms and agent-based platforms for emergency manage-ment
21 SWARM-Like Agent-Based Modelling and SimulationPlatforms SWARM [19] originally proposed by Santa Feinstitute is widely used in many research areas such asbiology ecology and society The tool provides a simulationenvironment for simulating agent with the support of a seriesof class libraries It is worth noting that SWARM is the pre-cursor ofmultiagent simulation tool it influences lots ofmul-tiagent simulation platforms such as NetLogo [20] RePast(REursive Porus Agent Simulation Toolkit) MASON [21]and SOARS (Spot Oriented Agent Role Simulator) [22]
NetLogo is a multiagent programming language andmodelling environment for simulating natural and socialphenomena It is particularly well suited for modelling com-plex systems evolving over timeThe language is easy to studyand the agent-based complex systems could be built rapidlyRePast is a software framework for agent-based simulationcreated at theUniversity of Chicago An extensible simulationpackage makes RePast become a generic multiagent simula-tion platform in social science research computing MASONdesigned by George Mason University is used to serve as thebasis for a wide range of multiagent simulation tasks rangingfrom swarm robotics to machine learning to social complex-ity environments The tool is a fast discrete-event multiagentsimulation toolkit in Java SOARS is designed by TokyoInstitute of Technology to describe agent activities under theroles of social and organizational structureDecomposition ofmultiagent interaction is the most important characteristicsin this framework
Mathematical Problems in Engineering 3
All the SWARM-like agent-based modelling and simu-lation platforms provide a portable lightweight and easilyextensible environment for simulating agents in arbitraryresearch areas However the heterogeneity in specific socialcomputing domain is not considered Furthermore most ofthe platforms cannotwell support large scale agent simulationbecause of the lightweight engine The engine cannot affordthe simulation of super cities like Beijing andNewYorkwhichhave millions of people
22 Agent-Based Platforms for Emergency ManagementBiowar proposed by Carnegie Mellon University simulatesthe impact of background diseases bioterrorism attackswithin a city 62 diseases are modeled in this platform tosimulate the outbreaks on the populationrsquos behavior GASM(Global-Scale Agent Model) is designed to study the spread-ing of H1N1 65 billion population is modeled with thesupport of official statistical data A global H1N1 spread fromTokyo is simulated in GASM EpiSimS proposed by LosAlamos lab simulates the spread of disease in regions suchas cities allowing for the assessment of disease preventionintervention and response strategies The daily movementsand interactions of synthetic individuals are representedexplicitly Burke and Epstein propose a computational modelof smallpox epidemic transmission and control [23] Theagents in thismodel interact locally with one another in socialunits such as homes workplaces schools and hospitals
However these platforms cannot provide a generic soft-ware framework to study emergency problems Biowar onlyfocuses on social networks individuals are all modeled as thenodes of social networks Agents in GASM and EpisimS areisomorphic without considering the heterogeneity in specificdomains
To sum up this section briefly reviews the existing mul-tiagent simulation platforms including SWARM-like agent-based modelling and simulation platforms and agent-basedplatforms for emergency management However they can-not satisfy the requirements of simulation performanceadaptability of software framework and heterogeneity ofindividuals in research of different emergency scenarios
3 KD-ACP
KD-ACP is an integrated software framework designed andimplemented based on the principle of the ACP approachshown in Figure 1
31The ACPApproach ACP approach is a social computing-based research paradigm It is composed of three componentsas its name artificial society for A computational experi-ments for C and parallel execution for P The basic idea ofACP approach is listed as follows
(i) Model the complex societies involving human behav-ior and social organizations as artificial societies usingmultiagent modelling techniques in a ldquobottom-uprdquofashion Artificial societies are regarded as a researchplatform to study emergency management
Actual societies Artificial societies
Managementand control
Experimentationand evaluation
Training andlearning
TestingTesting Operation
Figure 1 The parallel execution of ACP approach [8 33]
(ii) Utilize innovative computing technologies to evaluateand analyze various factors in emergency manage-ment quantitatively the computers are regarded asthe experimental social laboratories for investigatingemergency management problems
(iii) Provide an effective mechanism for the control andmanagement of complex actual social society throughcomparison evaluation and interaction with artifi-cial society
It is worth notifying that ldquoPrdquo here is not the ldquoparallelrdquo inldquoparallel simulationrdquo but the representation of ldquoparallel exe-cutionrdquo The idea of parallel execution is to build the parallelscenarios by paralleling the actual societies and artificialsocieties Consequently parallel control and management ofactual societies are implemented with the help of interactionsbetween parallel scenarios The goal of parallel executionis to find the best plans to adjust the methods of controland management based on the comparison and analysis ofdifferences between actual and artificial societies Artificialsocieties provide possible simulated results of evolutions byrepeated computational experiments The simulated resultsprovide evidences for the adjustment plans These plans areused in the control and management of actual societies suchas emergency management After the application of theseplans the observations from actual societies are collected forthe comparison with expectation The differences are used tofeedback to artificial societies The new turn of comparisonand analysis to find best adjustments of control and manage-ment is repeated
The mechanism of ldquoparallel executionrdquo has been provedto be effective for use in networked complex traffic systemsand is closely related to emerging technologies in cloud com-puting social computing and cyber-physical-social systems[24] In order to promote the development of parallel controland management in emergency management the artificialsociety is proposed in ACP approach which is the expansionof ldquoartificial traffic systemsrdquo
Instructed by the ACP approach KD-ACP is also com-posed by three components The details of the architectureand implementation of KD-ACP are discussed below
32 The Software Architecture of KD-ACP The architectureof KD-ACP is shown in Figure 2 the software framework iscomposed of a series of tools These tools are grouped intothree parts to support artificial society modelling computa-tional experiments and parallel execution
4 Mathematical Problems in Engineering
A part
C part
P part
Population and GeospatialEnvironment Generation Tool (PGET)
Generic ModelingEnvironment (GME)
Artificial society population andgeospatial environment database
Initialize
InitializeInitialize
Model DevelopmentTool (MDT)
FSM based models
Artificial SocietyEditor (ASE)
Scenario ofartificial society
Artificial societyruntime database
Agent and emergencymodel repository
Experiments DesignTool (EDT)
Experiment plans
InitializeSettings of
emergent eventsEmergent Events
Configuration Tool (EECT)Intervention Measures
Configuration Tool (IMCT) Settings of intervention measuresemergency response plans
Experiments ManagementTool (EMT)
Population data emergent eventsdata and intervention measures
Open source dataRegistration Tool (OsdRT)
Actu
al so
ciet
y Artificial SocietySituation Tool (ASST)
Optimal emergencyresponse plan
Internet
C++ code of modelsdll of models
Figure 2 The software architecture of KD-ACP
In the ldquoArdquo part Generic Modeling Environment (GME)[25] and Model Development Tool (MDT) are the kerneltools in the modelling of artificial society GME is anopen source modelling tool which supports domain-specificmodelling The domains of artificial society are created byGME in our work Models such as agent environmentemergent event and intervention measure are described inspecific domains first in GME With the help of modeltransformation thesemodels are all transformed to the FiniteState Machine (FSM) models Meanwhile code generationsare supported byMDT and thesemodels are all implementedin C++ Artificial Society Editor (ASE) is used to describethe concrete scenario of actual society which defines thescope of models set for artificial society Population andGeospatial Environment generation Tool (PGET) generatesthe population and geospatial environment data with thesupport of statistical data from actual society
In the ldquoCrdquo part Emergency Events Configuration Tool(EECT) initializes the models of emergent events whileInterventionMeasures Configuration Tool (IMCT) initializesthe models of intervention measures Experiments plans aregenerated byExperimentsDesignTool (EDT) Based on theseplans Experiments Management Tool (EMT) is used to runand manage the computational experiments to study theemergency problems
In the ldquoPrdquo part Artificial Society Situation Tool (ASST)seemed as the monitor of running artificial society Thestatistical data and situation are shown by ASST at runtimeIn the meantime the emergency response plans are made by
emergency decision organizations Parts of the influences ofemergency plans are reflected on Internet Open source dataRegistration Tool (OsdRT) is used to register the open sourcedata from Internet to artificial society
KD-ACP is developed using the BrowserServer archi-tecture the tools are integrated in the home page of KD-ACP as shown in Figure 3 Each tool is activated by the clickon the link For example Artificial Society Editor is startedwhen the link of ASE is clicked The working environmentand programming languages of tools in KD-ACP are listed inTable 1
Moreover the implementation of KD-ACP is mainlycomposed of modelling phase and computational experi-ments phase It will be discussed in the next section
33 The Modelling of Artificial Society in KD-ACP It is acritical problem to focus on the key parts of society in socialcomputing Based on the ACP approach the bottom-upmodelling is used to build the artificial society As a resultmodelling of artificial society is composed of three basic ele-ments agents environments and rules for interactionsHow-ever we still meet the problem that specific features shouldbe supported in artificial society For example emergentevents and intervention measures are required in artificialsociety for emergency management The modelling of onlybasic elements cannot cover the specific features in domainsTherefore domain-specific modelling [26] is introduced tosolve the problems in modelling artificial society
Mathematical Problems in Engineering 5
Model DevelopmentTool (MDT)
Artificial SocietyEditor (ASE)
Population and GeospatialEnvironment Generation Tool (PGET)
Emergent EventsConfiguration Tool (EECT)
Intervention MeasuresConfiguration Tool (IMCT)
Experiments DesignTool (EDT)
Experiments ManagementTool (EMT)
Artificial SocietySituation Tool (ASST)
Open Source DataRegistration Tool (OsdRT)
Generic ModelingEnvironment (GME) Artificial
society (A)Computationalexperiments (c)
MDT EECT IMCT
ASST OsdRT
ASE PGET EDT EMT
Parallel execution (P)
Figure 3 The implementation of KD-ACP
Table 1 The working environment and programming languages of tools
Tools Working environment Programminglanguages
Developmentplatform User type
Generic ModelingEnvironment (GME)
General computer (desktopapplication) NULL NULL Domain experts
Model Development Tool(MDT)
General computer (desktopapplication) C++ Visual Studio Model developers
Artificial Society Editor(ASE)
General computer (desktopapplication) C Visual Studio Domain experts
Population and GeospatialEnvironment generationTool
General computer (desktopapplication) C Visual Studio Domain experts
Emergency EventsConfiguration Tool (EECT)
General computer (withInternet Explorer clientside)
C ASPNET Domain experts
Intervention MeasuresConfiguration Tool (IMCT)
General computer (withInternet Explorer clientside)
C ASPNET Domain experts
Experiments ManagementTool (EMT)
General computer (withInternet Explorer clientside)
C ASPNET
Users of computationexperimentServer of EMT
Nodes in supercomputer(Console Program serverside)
C++ Visual Studio
Runtime Infrastructure ofEMT
Nodes in supercomputer(Console Program serverside)
C++ Visual Studio
Artificial Society SituationTool (ASST)
General computer (desktopapplication) C++ Visual Studio Domain experts
Open Source DataRegistration Tool (OsdRT)
General computer (withInternet Explorer clientside)
Java JSP Domain experts
6 Mathematical Problems in Engineering
M2T transformation
Modeling in domain
GME Generic Modeling Environment
Metamodeling
Instance of
FSM
Metamodeling for artificial society
DEVS
Petri Net State charts
Agentmetamodel
Environmentmetamodel
Emergency eventmetamodel
Interventionmetamodel
DAE
Domain-specific metamodels Semantic well definedmetamodels
M2M transformation
Modeling of artificial society
Interventionmodel
Emergency eventmodel
Environmentmodel
Agentmodel
FSM based emergent andintervention models
FSM based environmentmodels
FSM based agent modelsDomain-specific models
Semantic well defined models
M2M transformation
Agent model code Environment model code
Agent behavior model
Agent mentality model
Disease model code
Basic population model Disease spreading model
Disease state transformmodel
Intervention model code
Agent social networkmodel
Basic environmentinformation model
Roadnet model
Vaccination
Infectors isolation
Contactors isolation
Close work officesAgent contact model
Domain-specific models relevant codes
MDT Model Development Tool
Implementation
Figure 4 From metamodelling and modelling by GME to implementation by MDT of artificial society
331 The Principle of the Modelling of Artificial SocietyAccording to the principle of domain-specific modellingthe modelling of artificial society contains the followingsteps first metamodelling the basic elements of artificialsociety second modelling the specific features in domain ofemergency management third implementing the models ofartificial society in codes The whole process is illustrated inFigure 4The first and second steps are implemented in GMEwhile the third step is implemented in MDT
The first step is metamodelling which mainly focuses onconstructing the metamodels of artificial society Metamod-elling tries to study the common patterns of artificial societyThe outputs of metamodelling are metamodels which repre-sents the abstraction of the whole systemThe basic elementsof artificial society are described in metamodel The processof metamodelling is divided into four phases The first is theconstruction of the domain-specific metamodels As shown
in Figure 4 agent metamodel environment metamodelemergent event metamodel and intervention metamodelcompose the metamodel of artificial society The second isthe construction of the metamodels described by typicalmodelling formalisms such as FSM DAE DEVS and PetriNet [27] These formalisms are all semantically well definedThe third is the definition of the model transformationfrom domain-specific metamodels to metamodels of typicalmodelling formalisms The transformation standardizes themetamodels of artificial society by typical modelling spec-ifications The fourth is the definition of the transforma-tion templates from metamodels to code framework Thetemplates list the basic abstract interfaces of metamodels ofartificial societyThese abstract interfaces are implemented inthe specific-domain modelling and code generations
The second step is modelling the models of artificialsociety such as agent model environment model emergent
Mathematical Problems in Engineering 7
Figure 5 The metamodels of artificial society in GME
event model and intervention model are built Actuallythe models are the instantiation of metamodels in the laststep Different from the general modelling environment likeUML [28] the domain-specific modelling provides a familiarmodelling environment for the domain experts in artificialsociety For example emergency response experts only con-cern emergent eventmodel and interventionmodel inheritedfrom metamodels After constructing the domain-specificmodels based on domain-specific metamodels domain usersexecute the model transformation defined in the first stepAll the models of artificial society are transformed into FSMmodels As a result the models are implemented in this uni-fied modelling formalism (FSM) The model transformationmakes the simulation of the models possible
The third step is the generation of executable codes ofmodels The executable code framework is generated bymapping template from metamodels to code frameworkdefined in the first step Moreover domain developers alsoadd necessary codes to the framework to integrate thedynamic semantics of the models The code frameworkoutputs the dynamic link libraries by compilingThe dynamiclink libraries are loaded in the large scale artificial societyruntime infrastructure in computational experiments
332 The Metamodelling and Modelling of Artificial Societyby GME GME is used to build metamodels and modelsin our work As mentioned before the abstraction andcommon patterns of society are represented in metamodelsAccording to the bottom-up modelling style metamodels ofagent environment and communications are described inGME Figure 5 shows part of metamodels of artificial societyThe features of an agent metamodel are extracted from thecensus figures and statistical data Environment metamodelsimulates the geospatial places for the behaviors of agentsThemetamodel of communications among agents ismodeledto simulate the interactions such as infection in epidemicsand rumor propagation in public opinion formation eventsIt is worth notifying that the metamodel of communications
includes both the emergent event metamodel and interven-tion metamodel
From the perspective of modelling the details from spe-cific domains are considered in the models by the instantia-tion frommetamodels of artificial society For example socialrelationships based on complex networks are added in agentmodel to support the communications Agent activity is alsoused to quantify the agent activity under different scenarioEnvironment models are linked with the help of transporta-tion services subways and roads are modeled while the pathsearch is encapsulated in the services Emergent event modeland intervention model are also the domain-specific modelsThe modelling of artificial Beijing in GME will be discussedin detail in next section
333 The Implementation of Models of Artificial Society byMDT As mentioned before MDT is used to implementmodels such as agent environment emergent event andintervention According to the template of code frameworkthe implementations of models are generated by MDT Theimplementations are classified into two categories FSMmodels and services FSM models such as agents andenvironments are built under the specification of FiniteState Machine (FSM) [29] in MDT while all the servicessuch as transportation are encapsulated under the PublicService Standard This standard provides a generic interfacespecification for modelers to encapsulate public commonservices in artificial society FSM models like agent are builtstatistically from the quantitatively analyzed characteristicssuch as demographic attributes social behaviors emergencybehaviors and social networks Social behaviors describe thedaily behaviors of individuals while emergency behaviorsdescribe the individual behaviors in emergent events Forexample infected individuals are all isolated in hospital inSARS Isolation ismodeled as a typical emergency behavior inour work Correspondingly services are used to simulate themacroactual society Take transportation service for instancethe path search is needed by almost every agent duringmoving from spot to spot
8 Mathematical Problems in Engineering
Artificial society description in emergency
Description ofemergency
ArtificialSociety Editor
(ASE)
Agent behavior set
Agent role setAgent status set
Environment setAgent relationship set
Statistical data of real society
Descriptionof artificial
societyframework
Terminal of agent modeling andartificial society describing
middot middot middotAgent
metamodeling
Customized agentmodeling
Relationship modelDailyEmergency
Mod
el D
evelo
pmen
t Too
l (M
DT)
Agent modeling
Agent model Encapsulate
Encapsulate
Behavior model
Intervention
Emergency
Response
Daily
Social relationshipmodeling
Socialbehaviormodeling
Socialbehaviormodel
Emergencybehavior
model
Emergencybehaviormodeling
Building
Geospatial statistical data
Population statistical data Datacollection
Actualsociety
Artificial societypopulation and
geospatial database
Organizationservice
Transportationservice+
MappingAgent model andservice repository+ emergency model
repository
Climateservice
Geographicalservice
+
Agent status in emergencyEmergency management
Emergent event set
Social relationship statistical data
Environment statistical data
Population behaviorstatistical data
Emergency statistical data
Emergency organization statistical data
Geospatial and SocialEnvironment Generation
Tool (GSeT)
Agent model repository andpopulation database
C++C++C++C++
Modeling
Statistical modelJob ageSex location
Public service modeling
Organizationmodel
Transportationmodel
Geographicalmodel
Climatemodel
Interventionservice
Publ
ic se
rvic
e sta
ndar
d
Figure 6 The editing and initialization of KD-ACP
MDT provides domain experts with a tool to obtain thecode implementations of models With the help of compileenvironment like visual studio MDT also supports thefurther programming development of the specific domaindetails which cannot be described in modelling step
Both the FSM models and services are developed by theMDT first and then stored in the agent model and servicerepository The repository manages the models according tothe requirements from emergency problems and provides themodels for EDT to make the experiment plans
34 The Editing and Initialization of Artificial Society byASE and PGET As shown in Figure 6 ASE is used to editthe scenarios of artificial society within emergent eventsThe editing is composed of two parts (1) the statisticalinformation of artificial society in daily life such as the rolesof agents the relationships of agents and the types of environ-ments and (2) the statistical information of artificial societyin emergency including the statistical data of emergent
events the emergency organization and emergency relatedbehaviors of agents
According to the requirements of the editing thesestatistical data are collected from actual society manually bythe domain experts Based on these statistical data PGETgenerates the artificial society population and geospatialenvironment database The database supports the instanti-ation of artificial society at individual level For examplethe attributes such as age and gender of each agent can befound in the database With the support of the databaseFSM models and service repository discussed before itis sufficient for domain experts to study the emergencyproblems by computational experiments
35The Computational Experiments and Parallel Execution inKD-ACP The tools of ldquoCrdquo part and ldquoPrdquo part in KD-ACP areused to support the process of computational experimentsand parallel execution The working process is shown inFigure 7 EECT and IMCT are both the starting and ending
Mathematical Problems in Engineering 9
Agent model andservice repository
repository
Emergency EventsConfigurationTool (EECT)
Open Source DataRegistration Tool (OsdRT)
Intervention MeasuresConfigurationTool (IMCT)
Load
Output Output
Internet
Emergencyresponse plans
Load
∙ Input models of emergencyevents∙ Output settings of agentbehavior in emergent events+ settings of emergent events
∙ Input models of interventionmeasures∙ Output settings of agentbehavior in interventionmeasures + settings ofintervention measures
+ emergency model
∙ Input data from internetnetworks∙ Output population data +emergency events data +intervention measures data
Artificial societypopulation and
geospatialenvironment database
Artificial societyruntime database
OutputThe most optimal
emergencyresponse plan
Actual society
OutputOutputOutput
Input
Load
Load
Load
Load
Output
Artificial SocietySituation Tool (ASST)
Emergency decisionorganization
ClusterRuntime infrastructure
Deploy
Deploy
Experimentsdesign tool (EDT)
Artificial societyinitial data file
Experimentplan
Artificial societymodels
Experiments Management
large scale artificial societyruntime infrastructure
Tool(EMT) +super computerTIANHE-1A
Load
∙ Input experiment plan +artificial society models +artificial society initial data file∙ Output artificial societyruntime state data
∙ Input description of artificialsociety + artificial society populationand geospatial environment society+ emergent events configuration +
intervention measures configuration∙ Output experiment plan + artificialsociety models + artificial societyinitial data file
∙ Input runtime statistical data∙ Output graphics charts ofartificial society statistical data +situation of artificial society
Figure 7 The computational experiments and parallel execution of KD-ACP
point The emergent events and intervention measures areconfigured by EECT and IMCT respectively The configu-rations of emergent events are used to simulate both thereal emergencies like SARS and H1N1 and the supposedemergencies for experiments Similarly the configurationsof intervention measures are also used to reproduce thereal one and simulate the supposed one The repeat of theemergency is used to verify the models while the supposedconfigurations are used to obtain the optimized decision planto the response of the possible emergencies
With the input of artificial society model and servicerepository artificial society population and geospatial envi-ronment database and the configurations discussed beforeEDT generates the experiment plans to meet the require-ments of research on emergency management The outputof EDT includes artificial society models artificial societyinitial data files and experiment plan The models aredownloaded from the repository while the data file is thecollection of data from the database to initialize the modelsWhen the models and data files are ready EMT loads theexperiment plan and deploys the models and data to thecluster or TIANHE-1A supercomputer [30] which was theworldrsquos fastest supercomputer built by National University of
Defense Technology (NUDT) in China in 2010 According tothe plan the experiments are done repeatedly on the largescale artificial society runtime infrastructure [31 32] by themultisample settingsThework process is the implementationof computational experiments in ACP approach
Traditionally emergency response plans are made byemergency management theories and experiences The onlyway to test the effective of plans is the feedback results of realworld ACP approach provides a novel method to supportemergency response plan making by parallel execution Asshown in Figure 7 the work process of KD-ACP is composedof two loops The inner loop composed by red arrowsdescribes the process of computational experiments while theouter loop of yellow arrows illustrates the process of parallelexecution During the runtime of computational experi-ments the statistical data of artificial society is collected andstored in the artificial society runtime database Based on thedatabase ASST outputs the customized situation of runningartificial society by graphics charts and situation mapsThe information is sent to the organizations of emergencydecision to support making the emergency response plansWith the help of computational experiments loop these plansare simulated repeatedly to find the most optimal one
10 Mathematical Problems in Engineering
Open Source Data Registration Tool (OsdRT)
Emergent EventsConfiguration Tool (EECT)
Intervention MeasuresConfiguration Tool (IMCT)
∙ Web information collecting
∙ Deep mining in social media
∙ Web information denoising
∙ Information filtering
Data acquisition
Actual society
Open source informationin social media
Internet
∙ Basic element extracting∙ Individuals and organizationsextraction∙ Sentiment analyzing andopinion mining∙ Social network analyzing
Data extraction
∙ Domain knowledge description∙ Information fusion and collisiondetection∙ Domain knowledge construction∙ Social media informationstandardization
Data standardization
Knowledge
Figure 8 The compositions of OsdRT [34]
Moreover the most optimal plan is used to the responseof emergency in actual society According to the idea ofparallel control in [24] the feedback of actual society ispartly collected from Internet networks by OsdRT As shownin Figure 8 OsdRT is composed of three components dataacquisition data extraction and data standardization Dataacquisition collects mines and filters information fromsocial sensing networks Data extraction includes basicelement extracting individual and organization extractingsentiment analyzing and social networks analyzing Datastandardization specifies the useful knowledge and sent it toconfiguration tools in KD-ACP
By processing in OsdRT the knowledge about emergentevents and intervention measures are analyzed first andregistered in the EECT and IMCT The registration updatesthe settings of emergent events and intervention measuresThis loop composed of yellow arrows implements the parallelexecution in ACP approach The implementation of parallelcontrol and management provides a data-driven approachthat considers both the engineering and social complexityfor modelling analysis and decision making in emergencymanagement
4 Modelling Beijing City with KD-ACP
41 How to Build the Artificial Beijing According to themodelling of artificial society in KD-ACP discussed beforethe Beijing city is modeled as follows
To meet the requirements of emergency managementsthe basic elements of artificial society are extended Asshown in Figure 9 six elements are required to simulate thecity agents environments transportation activity schedulecommunication and agent activity
411 Modelling Artificial Beijing Figure 10 shows the mainGUI of GME for the modelling of artificial modelling inpublic health events Metamodels are listed in the left area ofFigure 10 the list provides basic syntax elements for domainexperts to model artificial society Domain experts build
models based on the knowledge of their own MeanwhileGME supports hierarchy for building large scale systemsThesyntax symbol listed in GUI can be extended in new tab bydouble clicking Take agent for example the model of agentcan be detailed by edition in another tab page of agent
As shown in the center of Figure 10 the models ofartificial Beijing consist of five parts models of agent andenvironment domain models of public health events inter-ventionmodels controllermodels and services Agentmodeldescribes individuals in society it is composed of basicpopulation information action social relationships activityschedule and disease related information Activity schedulerepresents individualrsquos physical actionmodel focusing on thedaily action of agents Environment model includes physicalentities such as buildings playground transportations andagents contained in environment Domain models of publichealth events are composed of the propagation model of dis-ease disease state transition model and so on Interventionmodels include the settings of vaccination isolation and soon The models mentioned before are all FSM models Themechanisms of these models will be detailed in next sectionsController models and services are the public service mod-ules they are implemented in the development in MDT
412 Modelling Agents and Environments Under the speci-fication of FSM agent and environment models are imple-mented in two parts the state space and state transitionsThe state space is composed of the demographic attributesand behavior related attributes The transitions are triggeredwhen the conditions of states are satisfied As shown inFigure 11(a) the action of agent is changed when the ldquonexttimerdquo condition is satisfied in agent model while the agentslist is changed when the agent arrival condition is satisfied inthe environment model
413 Modelling Activities Agent activities come from theagent state transitions of actions such as movements andcommunications The actions of agents are instructed by theactivity schedule shown in Table 2 Activity schedule lists
Mathematical Problems in Engineering 11
Transportation EnvironmentsEnvironments
Agents
Artificial Beijing
Agent activityActivity schedule
Agent activity
Communication by socialrelationship networks
Communication by socialrelationship networks
Figure 9 The basic elements of artificial society
Syntax elements of DSM
Services
Models of agentenvironment andactivity schedule
Domain models of publichealth events
Intervention models
Modelingaspect ofdomainexperts
Controllermodels
Figure 10 The models of artificial Beijing in GME
all the actions with probability in one day for agents inboth normal and emergent situation [38] There are severaltypes of activity schedule in artificial Beijing student agentactivity schedule worker agent activity schedule emergentagent schedule and so on For example Table 2 gives an agentactivity schedule Upon the instruction of activity schedulestudent agent changes the actions by 119901
119894after state transitions
The119901119894in the tablemeans the action probability in the relevant
period In the duration from 0800 to 1200 a student agenteither goes to classroom to have class or goes to libraryto study The probability of class action is 119901
2while the
probability of study action is 1 minus 1199012
Agent behaviors are decided by the settings of activityschedule In addition to the daily activity schedulementionedbefore emergent activity schedules are also considered inour work Take public health events for instance an infectedagent changes schedule from a normal one to emergentone The workflow of a susceptible agent is illustratedin Figure 12 to show the change of behaviors After theinfection the agent is set in incubation phase Not all theincubation agents will become symptomatic Some of themturn back to being susceptible and some of them becomesymptomatic The symptomatic agents change their activityschedule from normal to emergent In the emergent case
12 Mathematical Problems in Engineering
Next actionNext environment (location)
Advance to next actiontime
Find next environment inactivity chain by action
Current actionCurrent environment (location)
Q
q0
120575 Find next action inactivity schedule by time
F
Σ
(a) The state transitions of agent
Q
q0
120575 F
Current agent list
Current pollution level
Next agents list
Next pollution level
Agents entered
Infected agents entered
Add agents to currentagents list
Compute pollution levelby infected agents
Σ
(b) The state transitions of environment
Figure 11 The state transitions of models
Table 2 The activity schedule of a student agent
Duration (Δ119905) Activity Location Probability 119879Location (minute)0000ndash0600 Sleep Dormitory 119875
0(100) 360
0600ndash0800 Breakfast Dormitoryrestaurant 1198751(068) 120
Sports-breakfast-travel Playground-dormitoryrestaurant 1 minus 1198751(032)
0800ndash1200 Class Classroom 1198752(077) 240
Study Library 1 minus 1198752(023)
1200ndash1400 Lunch Restaurant 1198753(090) 120
Lunch-shopping Restaurant-convenience store 1 minus 1198753(010)
1400ndash1800 Class Classroom 1198754(070) 240
Study Library 1 minus 1198754(030)
1800-1900Dinner Dormitoryrestaurant 119875
5(063)
60Sports-shopping Playgroundconvenience store 1198756(025)
Study Classroomlibrary 1 minus (1198755+ 1198756) (012)
1900ndash2200Rest Dormitory 119875
7(045)
180Dinner-rest Dormitoryrestaurant-dormitory 1198758(020)
Study Classroomlibrary 1 minus (1198757+ 1198758) (035)
2200ndash2400 Rest Dormitory 1198759(072) 120
Sleep Dormitory 1 minus 1198759(018)
infected agents go to hospital according to the treatmentschedule or stay in dormitory according to isolation scheduleAfter the treatment in hospital or self-immunoprocess agentsbecome healthy and immune of disease If the agents aretreated in the hospital they are not allowed to get out untilrecovered Moreover activity schedules are also influencedby the emergency response plans For example in the case ofisolation in emergency response plans the activity scheduleof agents who had contacted with infected is changed Onlythe locations including home and dormitory compose thisemergent schedule Likewise based on the statistical datafrom emergency response plans the additional possibilities
are added into activity schedulesThe behaviors of agents alsochange with activity schedules In the view of ACP approachthe injection of new emergency response plans implementsthe parallel execution of microagent behaviors
In another aspect contact frequency is also anothercrucial element to determine the infected rate in public healthevent It is important to model the contact frequency andcontact time of individuals Based on the studies on thecontact behavior of human being by questionnaire surveyEdmunds found that the contact frequency of individualcould be fitted approximately into a normal distributionand the mean and the standard deviation distributions are
Mathematical Problems in Engineering 13
Susceptible agent ImmuneYes
Healthy and immune
No
Choose schedule
Self-isolation Go to hospital
Staysymptomatic
No
Receive treatment
Contact withinfected agents
Symptomatic
No
Self-immunoprocess
Yes
Recovered
Yes
Incubation
Infection
No
Yes No
NoYes
Figure 12 The workflow of a susceptible agent
168 and 85 [39] So it is possible to apply normal randomvariable to model the contact frequency In our work Box-Muller method [40] is used to generate the random variableof contact frequency shown in the following
119865119862= 120583119865+ 120590119865(minus2 ln (120574
1))12 cos (2120587120574
2) (1)
in which 119865119862is the random variable of contact frequency 120583
119865
is the mean value of normal distribution 120590119865is the standard
deviation of normal distribution and 1205741and 120574
2are the
uniform random variables distributed in the interval [0 1]Based on (1) and survey data [41] the contact frequenciesof agent 119865contact(119860 119894) are discretized as in Table 3 within theconsideration of activity differences
Similarly the duration per contact between individualsis another key factor of AHC transmission The duration ofcontact could also bemodeled by the normal randomvariablein the following [42]
119879119862= 120583119879+ 120590119879(minus2 ln (120574
1))12 cos (2120587120574
2) (2)
in which 119879119862is the random variable of duration per contact
120583119879
is the mean value of normal distribution 120590119879
is thestandard deviation of normal distribution and 120574
1and 120574
2
are the uniform random variables distributed in the interval
[0 1] In addition because the time spent in specific location119879Location(119860 119894) also follows the random distribution like (3)in our work In order to make the duration of contact119879contact(119860 119894) shorter than the 119879Location(119860 119894) the mean andstandard deviation of (2) is set up in (3) and (4) [42]
120583119879=119879Location(120583119865+ 120590119865) (3)
120590119879=119879Location(10120583119865+ 10120590
119865) (4)
in which 119879Location is the time of an activity in a specificlocation set in Table 2 120583
119879and 120590
119879are the mean and standard
deviation of contact frequency As a result the duration ofagent 119879contact(119860 119894) in specific locations (listed in Table 2) isalso discretized in Table 3 within the consideration of activitydifferences
414 Modelling Transportation In artificial Beijing threekinds of travel models are considered walking inside thedistrict travelling by road networks and travelling by subwaynetworks According to the traveling models the agentmovements are implemented by the compositions of travelingmodels before So it is important to build the basic road
14 Mathematical Problems in Engineering
Table 3 The contact frequency and duration in different locations and activities by social relationships in case of student and teacher
Social relationship Activity Location 119879contact(119860 119894) (minute) 119865contact(119860 119894)
Student-student
Sleep Dormitory 0 0
Breakfast Restaurant 119873(52) 119873(32)Breakfast Dormitory 119873(52) 119873(21)Sports Playground 119873(21) 119873(53)Class Classroom 119873(21) 119873(21)Study Library 119873(21) 119873(21)Lunch Restaurant 119873(158) 119873(32)
Shopping Convenience Store 119873(203) 119873(21)Dinner Restaurant 119873(108) 119873(21)Dinner Dormitory 119873(158) 119873(32)Rest Dormitory 119873(63) 119873(52)
Teacher-student Class Classroom 119873(21) 119873(21)
Subway network of Beijing city (red lines) Road network of Beijing city (green lines)
Figure 13 The implementation of transportation for artificial Beijing
and subway networks With the help of Google map thekey points of road are sampled first and the road networksare generated by the links among point Similarly subwaynetworks are generated by the links between subway stationsmapped from the real stations in Google map Figure 13shows the road and subway networks in artificial Beijingwith the road networks in green and the subway networks inred Most of the districts in Beijing city are covered by thesetwo networks The algorithm of hierarchical route planningis proposed as follows
(i) Find the next position where agent will be located innext action
(ii) Obtain the travelingmodels from the statistical trans-portation data
(iii) Search the path from the starting position to thenearest road entry or subway stations
(iv) Search the optimal path from the starting station orroad entry to the nearest station or road exit of targetposition
(v) Search the path from the target subway station or roadexit to the target position
(vi) Attach the subway train number or the bus numberincluding the transfer information to the path accord-ing to the timetable of subway or bus
(vii) Connect all the paths obtained before Generate thepath from the starting position to the target position
Based on the algorithm for each agent in artificial Beijingthe transportation is simulated during the computationalexperiments It is worth noting that the activities insidethe buses or train cars are considered because the agentcommunication during travelling is also necessary to besimulated
Mathematical Problems in Engineering 15
(a) The implementation of agent in MDT (b) The code framework of agent
Figure 14 The agent model implementation in MDT [35]
415 The Implementation of Models by MDT All the modelsare developed in the MDT Figure 14 shows the GUI of thetool MDT is used to implement the state space and statetransitions of FSM models The code framework is shownin Figure 14(b) Using the C++ inheritance features and thetechniques of dynamic link library (dll) the models areencapsulated in the dll componentsThe flexible compositionmechanism is designed to support the models evolution incomputational experiments
416 The Description of Artificial Beijing by ASE As men-tioned in Section 34 ASE describes the artificial Beijingin macroview The statistical information of Beijing city iscollected by ASE Table 4 lists part of the description ofartificial Beijing generated by ASE
42 How to Obtain the Data of Beijing City
421 Population and Geospatial Environment Database ofBeijing City Based on the models developed for ArtificialBeijing it is necessary to generate the initial data to do thecomputational experiments Because the individual level dataare not available it is necessary to construct an individual-based population database for both accurate computationalexperiment and determining optimal decisions
According to the state spaces of models discussed inSection 412 the geospatial and population database of Arti-ficial Beijing is designed as shown in Figure 15 The kernelpart of database consists of the tables such as agent list tablegeospatial environment list table household list table andagent distribution table Agent list table is an individual leveltable used to store the data to initialize the state space of agentmodel such as id and gender age Geospatial environmentlist table is also an individual level table which stores thedata of environment attributes such as id street id andtype Household list table and agent distribution table storethe statistical data for the data generation for agent andenvironment models With the help of the geospatial andpopulation database the data generating of artificial Beijingis proposed in the next section
422 Generating Geospatial and Social Environment Datafor Artificial Beijing by PGET The algorithm of generatinggeospatial and social environment data for artificial Beijingis used to quantify the spatial distribution of population andformalize geospatial behavior of each agent The algorithmshown in Figure 16 is capable of generating a syntheticartificial society which allowsmultiresolution statistical datasocial interactive behavior and multilayer social networksto be integrated together The synthetic population canrepresent individual agents in the form of households andhousehold members and the synthetic population is statis-tically equivalent to a real population For each householdcharacteristics such as address household size family typesand relationships are generated Each person is described bycharacteristics such as age gender social role and correlatedlocations The algorithm provides an effective methodologyto reconstruct the computing environment in high resolutionby using statistical data in low resolution leading to betterprediction and management of emergencies The details ofalgorithm are illustrated in [36]
As shown in Figure 17 the implementation of the algo-rithm is embedded inside the GEPTThe collective input datasuch as the data files of population distribution is collected inthe tool first then the algorithm is activated to generate thedatabase The generation lasts almost twelve hours one timewith the specified parameters settings When the generationis finished the data of artificial Beijing is obtained in thedatabase with 19610000 agents and 16000 environments
5 H1N1 Experiments in ArtificialBeijing with KD-ACP
With the support of the artificial Beijing modeled before thecomputational experiments and parallel execution are alsoimplemented by KD-ACP The experiments are designed tofind the most optimal emergency decision response plansEpidemic in city is a typical scenario in emergency manage-ment H1N1 epidemic in Beijing in 2009 is used to be the usedcase to test KD-ACP According to Section 34 it is necessarytomodel emergencymanagement first in order to support thecomputational experiments
16 Mathematical Problems in Engineering
Table 4 The description of artificial Beijing generated by ASE
Statistical features ofartificial Beijing Descriptions of statistical features Data sourcedefault values
Agent roles Baby child primary student middle school student college student worker retired
Demographic data file ofBeijing
Environment types Super market store restaurant park stadium hospital school cliniccommercial building dormitory building residential building
Demographic data file ofBeijing
Demographic data Age distribution of population sex distribution of population householddistribution of population retired age distribution children distribution
Demographic data file ofBeijing
Environment data Distribution of all types of environments Environment distributiondata file of Beijing
Relationships Classmate relative college friendship family Demographic data file ofBeijing
Varying parameters
Age difference of couples119873(120583120590) 119873(03) isin [0 12]Minimummarriage age (male female) (22 20)Age difference of children119873(120583120590) 119873(13) isin [1 10]Enrollment rate of kindergarten 120572 08Enrollment rate of primary school 120573 1Enrollment rate of middle school 120574 1Enrollment rate of college 120575 08Entrance age of kindergarten 1205721015840 3Entrance age of primary school 1205731015840 7Entrance age of middle school 1205741015840 13Entrance age of middle school 1205751015840 19
Age distributioninformation table
Agent settable
Geospatialenvironment set table
Household settable
Figure 15 The geospatial and population database of Beijing city
51 Modelling Emergency Management The modelling ofemergency management consists of two parts the modellingof emergent events and the modelling of intervention mea-sures H1N1 and the intervention measures for H1N1 aremodeled in our case
511 Modelling H1N1 by MDT H1N1 Model in agent simu-lates the states transitions of health status and the relevantactions Referring to SIR (SusceptibleInfectiveRecovered)[43 44] models agent has three health statuses susceptibleinfected and healthy with immunity Only susceptible agentcan be infected by the contact with infected agents After
the three infected phases (incubation being symptomaticand recovery) of H1N1 model agent is healthy again It isworthy mentioning that the recovered agents can be infectediteratively when the immunity disappears gradually by timeThe state transitions are illustrated in Figure 18
According to [45] the latent period of H1N1 was in aWeibull distribution within people from one to seven days[46 47] The distribution was usually centered in the rangebetween one and three days AWeibull random variable [40]is used to model the latent period as described in
119879Lat = 120572 times [minus ln (1 minus 120574)]1120573+ 120592 (119879Lat ge 120592) (5)
Mathematical Problems in Engineering 17
Censusdata
Generate GUID
Assign individualhousehold
Generate familystructure
Assign family roleto every member
Endow individualwith age gender
social role
Distribute environmentlocations on the map
School Workplace Entertainmentlocations
Consumptionlocations
Assign school orworkplace to
individual
Correlatedindividual withother locations
Generate friendshipin school
Generate coworkerrelations inworkplace
Assign individualadministrative
region
Generate housebuildings
Locate householdto house building
Generate familyrelation
Generateneighborhood
relation
PopulationEnvironment
Environment listSocial network
Figure 16 The algorithm of generating geospatial and social environment data [36]
Edit
Edit
Edit
Edit
Edit
Edit
BrowseStatistical characteristic of population
Geospatial population data
Population data of street
Population types
Geospatial and social environment generation tool
Workspaces
Parameters distribution settings
Household types
Roles in household
Household structure
Social roles
BrowseBrowseBrowseBrowseBrowseBrowse
Population distribution data
File path
Retired distributionPop distributionHousehold numHousehold distributionSex distributionAge distributionData type
Figure 17 The GUI of Population and Geospatial Environment generation Tool (PGET)
18 Mathematical Problems in Engineering
Infected
SusceptibleHealth
withimmunity
Infected
by contac
t
with in
fected
agents
Recovery
Increase before beingsymptomatic symptomatic
Decrease after being
Decrease by time
Infectivity
Immunity
Figure 18 The transitions of agent health status
119879Lat denotes the duration of latent period 120592 120572 and 120573 arethe location parameter scale parameter and shape parameterof Weibull distribution respectively 120574 is a uniform randomnumber in the range [0 1] According to statistics [47] 120592 120572and 120573 are set as 0 18 and 121 respectively Then the mean(standard deviation) of latent period is calculated as 159 days(058 day2) Furthermore infectious period is set as 7 days
A susceptible studentrsquos probability of getting infecteddepended on the infectivity of infectious agents his ownimmunity level the duration of the contact action and soon The infectivity of an infected agent evolved with timeduring the infectious period According to the statistics ofinfectivity in the chart with red bars in Figure 19 an infectedstudent had the highest level of infectivity in the second dayafter he had the first symptom The infectivity levels in otherdays in comparison with the second day are in Table 5 Theday labeled as ldquominus1rdquo means the day before the starting of thesymptomatic period
Based on the algorithm discussed before theH1N1modelis developed with the help of MDT The settings of influenzamodel will be discussed in Section 521
512 Modelling Intervention Measures by MDT In corre-spondence intervention measures model is used to ceasethe outbreak of H1N1 As discussed in [42 48 49] if theappropriate measures are taken when the infectious diseaseemerges the transmission of the disease could be sloweddown and the damage could be decreased Therefore it isnecessary and important to design emergency interventionmeasures in the artificial Beijing According to the MinistryofHealth inChina the intervention policies [50] are designedto control the spread of influenza such as H1N1 The inter-ventionmeasures of these policies including are interventionsactivated time vaccination rate antibiotic delay to hospitalisolation duration close workspace duration and limitationof activities listed in Table 6 Similarly the models of theseintervention measures are also developed with the help ofMDT Using the reasonable data ranges listed in Table 6intervention models can be initialized to do the computa-tional experiments These experiments are used to show how
Symptomatic1 2 3 4 5 6 7minus1
Recovered
Tim
e (da
y)
Infe
ctiv
ity
Figure 19 The distribution of infectivity rate of an infected agent[37]
Table 5 The proportion of infectivity
Days (119894) minus1 1 2 3 4 5 6 6Proportion of infectivity 03 05 1 08 04 02 01 005
to restrain the H1N1 transmissionThe settings of these inter-vention measures in IMCT are introduced in Section 522
52 Experiment Settings
521 H1N1 Settings by EECT EECT is designed to initializethe emergent events models such as H1N1The settings comefrom the specified emergent event models As shown inFigure 20 in theH1N1 used case the settings are composed ofthe distribution of infection source infection rate infectionperiod and so on The infection rate changes with envi-ronments and infection periods The settings are initializedfrom both the research of medicine on H1N1 and the data ofH1N1 outbreak in Beijing in 2009The process of influenza inartificial Beijing is simulated to repeat the possible scenariosof H1N1 break During the computational experiments ofartificial Beijing theH1N1 settings such as the period of beingsymptomatic are updated by the dynamic data registrationsfrom OsdRT The updating corrects the H1N1 model duringthe experiments With the help of OsdRT artificial Beijing isable to approach actual society It is also an implementationof parallel execution
522 Intervention Measures Settings by IMCT In the H1N1case the settings of intervention measures are shown inFigure 21 in IMCT These measures are designed under theinstruction of Emergency Decision Organizations like ChinaDisease Center (CDC) The measures such as vaccinationrate the rate of closedworkspace and the starting time can beinitialized in the tool Emergency response plans are imple-mented in these settings As illustrated in Figure 7 all the pos-sible compositions within value settings in reasonable inter-vals are tested to find the most optimal emergency responseplan For example different compositions of measures suchas vaccination rates and rates of workspace closed are usedto design the plans of experiments By the analysis of theresults the best composition of interventionmeasures can beobtainedThe analysis will be discussed in case study in detail
Mathematical Problems in Engineering 19
Table 6 The intervention measures of influenza
Intervention measures ValuesInterventions activated Time 119879Intervention isin [0 80th day]Vaccination rate 119877Vaccination isin [0 1]Antibiotic rate 119877Antibiotic isin [0 1]Delay to hospital 119867ospital(119860 119894) 119879infected lt 119905 lt 119879recovered + 119879HDelay 119879HDelay isin (0 3th day]Isolation duration 119868solation(119860 119894) 119879Intervention lt 119905 lt 119879Intervention + 119879IDuration 119879IDuration isin (0 14th day]Close workspace duration 119862lose(119864119894) 119879Intervention lt 119905 lt 119879Intervention + 119879CWDuration 119864119894 isinWorkspace 119879CWDuration isin (0 7th day]Limitation of agent activity duration 119871 imitation(119860 119894) 119879Intervention lt 119905 lt 119879Intervention + 119879LAADuration 119879LAADuration isin (0 7th day]119877Vaccination means the vaccination rate of artificial Beijing while 119877Antibiotic means the antibiotic rate used for infected agents in hospital119867ospital (119860119894) means that the infected agent 119860119894 is sent to hospital he or she will receive the retreatment immediately119868solation (119860119894) means that the infected agent 119860119894 is isolated at home he or she can only contact with the other agents until isolation duration is over119862lose (119864119894) means that the environment 119864119894 typed of workspace is closed until the close workspace duration is over119871 imitation (119860119894) means that the activities of agent 119860119894 are limited only the workspace and home are allowed until limitation duration is over
Incubation Weibulldistribution
Symbolic LogNormaldistribution
Recovering LogNormaldistribution
H1N1influenza
Settings of H1N1 influenza model
Disease model
Transmissionmodel
Figure 20 The settings of H1N1 in EECT
It is worth notifying that the settings of interventionmeasures are also influenced by OsdRT For example theH1N1 outbreak areas are notified by OsdRTThe interventionmeasures such as close workspace are used only in theoutbreak areas Therefore the unnecessary computation ofintervention measures is avoided So the data injection byOsdRT not only makes artificial society approach actualsociety but also increases the performance of computationalexperiments
53 Experiments by EMT When the initialization of arti-ficial Beijing is ready EMT is used to do the computa-tional experiments EMT is composed of two parts thecentral controller in the central node and the residen-tial service deployed in the computational nodes Centralcontroller loads the experiment plans and controls the whole
experiments process the GUI of controller is shown inFigure 22 EMT manages the computational experimentsin four steps cluster configuration load experiment plansmodels deployment and running experiments
531 Cluster Configuration Before the configuration clusterinformation is collected from the residential services insidethe cluster The information includes nodes amount nodenames node IPs and CPU occupations It is used to quantifythe computing power of the clusterThe cluster configurationis based on this information The nodes are selected forexperiments Then the LP (local process) [51] number foreach node is set according to the number of the CPU kernelsAfter the configuration the experiment plans are loaded tocustomize the models and data
20 Mathematical Problems in Engineering
Streets of close workspaces
Limitations ofactivities
IsolationsRespirator
Closeworkspace
Antibiotic
VaccinationHospital
Surveillance
Figure 21 The settings of intervention measures in IMCT
Information ofexperiment plan
List of experiment plans
Start | Pause | Reset | Stop | Speedup | Slowdown | Back
NodeName | CPU occupation | Memory occupation |Network | Models | Status
Figure 22 The deployment and running of experiments in EMT
532 Load Experiment Plan As discussed in Section 35 thedescription of experiment plans is composed of the infor-mation in three aspects models and the initial data for theexperiments the deployment mapping tables from the mod-els to the nodes and the plan of experiment executionWhenthe experiment plan is loaded by the EMT the file namesand model descriptions of models and data are listed Withthe support of cluster configuration models can be mappedto the nodes in the GUI according to the computationrequirements EMT also provides the setting of experimentsitself including the running times start time and end time
533 Models and Data Deployment Models are stored inthe agent population and services repository According tothe settings in experiment plan the customized models aredownloaded from the repository first then the models suchas agent model environment model emergent events modeland intervention measures model are integrated to be thecompositionmodels of artificial BeijingThemodels and datalocations in the nodes are configured in the ldquomodel pathrdquoand ldquodata pathrdquo in the GUI of central controller Finallythe models are uploaded to the nodes in cluster with theconsideration of partition The partition of artificial society
Mathematical Problems in Engineering 21
models is different from the multiagent system in [52]Models in artificial society are divided into three levels agentsand environments grids and city Agents and environmentsare encapsulated in grids which compose the whole cityThus the partition of artificial society becomes the partitionof grids The parallelism degrees are found to support thepartition of models agent independence and grid indepen-dence According to the parallelisms agents are partitioned toLPs according to the population distribution in grids Agentmovements among grids are simulated by the movementevents among grids in different LPs As a result a two-tierparallel architecture is proposed to support the simulation ofartificial society The architecture is detailed in [53]
534 Running of Experiments When the preparation isready the experiments can be done by the central controllerof EMT The running and control commands of the experi-ments are grouped into three aspects start pause or stop theexperiments get the running information of the nodes suchas the snapshot of the desktop of node With the help of thesecommands users can run the computational experimentsto do the research on the emergency management It isworth notifying that the simulation engine is optimized inorder to support the ten million agent simulation UsingGPU as coprocessor a two-tier parallel simulation engine isdesigned with support of MPI and OpenCL through phasedsynchronization [54] One-sided communication is used forreflection of remote simulation objects and message passingbetween processes A general kernel function prototype iselaborately designed and conditionally compiled for execu-tion on both CPU and GPU Moreover the optimizationoperations like load balance is developed under the instruc-tion of activity-based simulation [55] The densely populatedgrids are given more CPU and memory resource accordingto the activity predictions
54 Results of H1N1 Experiments with KD-ACP On thebasis of artificial society models and the intervention mea-sures mentioned in Section 512 a series of computationalexperiments are performed to study the H1N1 influenza inBeijing In our case of artificial Beijing 19610000 agentsare simulated in the cluster within 48 CPU cores and 128Gmemories the nodes are connected by kilomega networks Ittakes 18 hours to simulate a 250-day disease spreading
In order to illustrate how the optimal plans are obtainedbyKD-ACP the experiments are divided into four groups theexperiments ofmodel validation the experiments of sensitiv-ity analysis of vaccination rate the experiments of sensitivityanalysis of isolation and the experiments of the combinationof interventionmeasuresThese four groups show the generalprocess of the research on emergency management in H1N1influenza Firstly the models of H1N1 are validated withthe support of historical data Secondly the traditionalmedical interventions of pandemic such as vaccination areanalyzed to find the optimal plan Thirdly the traditionalnonmedical interventions of pandemic such as isolation areanalyzed to find the optimal plan Finally the combination ofinterventions is tested to show the combined effect
It is worth mentioning that all the experiments in ourwork are performed 100 times These experiments are ini-tialized with different settings of interventions in IMCT Theresults shown in the figures are the mean values
541 The Experiments of Model Validation In our artifi-cial Beijing the models of H1N1 influenza are built basedon the existing researches Wang and her colleagues hadstudied the H1N1 influenza by SEIR (SusceptibleExposedInfectiveRecovered) models [56] With the support of theirwork we build the models of artificial Beijing in the mannerof multiagents
In Figure 23 the control group shown in red line isdrawn based on the realistic statistical data collected in H1N1influenza in Beijing in 2009 The influenza lasts more thansix months and more than 170000 people are infectedComparing with the control group the simulated resultswithout interventions are drawn in blue It is obvious that thesimulated case fits the control group well before 70 days Butthe ascending of control group is slowed after 70 days andthe peak value (17883) is much less than the simulated results(25135) It is because the vaccination is activated at day 67 incontrol group according to the intervention measures usedin 2009 The restrain by vaccination becomes effective eightdays later The change of infected number after vaccinationis shown by red spotted line The trend of control group isdifferent from the simulation data after 75 days
However the simulated results (blue line) in compu-tational experiments fit the control group (as shown inFigure 23) in the first 70 days when no intervention mea-sures are executed The key features such as effective basicreproduction number (119877
0) are also almost the same in both
cases Therefore it can be concluded that the models ofartificial society built in microview are validated by therealistic statistical data in influenza in Beijing in 2009
In this group of experiments no interventions are exe-cuted in order to validate the models Therefore the simu-lated results of this extreme case which cannot happen todaycould be obtained As shown by blue line the number ofinfected agents without interventions reaches peak value atday 91 with 25135 consisting of 8052 in symptomatic phaseand 28950 in recovery phase It is obvious that the totalnumber of infected agents grows slowly at the beginningtime Then the number increases quickly when the influenzabreaks and it reaches its maximum value with 383870 in theend This phenomenon can be explained as in the followingAt the beginning of influenza due to the limited numberof infected agents the propagation of the disease remainsslow but many agents become susceptible When the numberof infected agents increases rapidly the spreading becomessignificant However the number of infected agents decreasesafter day 91 It is because the disease spreading relies on thesocial networks and spatial contact networks [49] It meansthat the infected agents can only infect agents either in thesocial relationship networks or in the spatially contacted
542 The Experiments of Sensitivity Analysis of VaccinationRate According to Section 512 vaccination is a typical
22 Mathematical Problems in Engineering
0
Infe
cted
agen
ts
0 50 100 150 200 250
Statistical dataSimulation data
25000
20000
15000
10000
5000
Day d
Figure 23 The results with no intervention executed and the com-parison with realistic statistical data in Beijing in 2009
medical intervention in influenza Traditionally the effective-ness of vaccination can only be evaluated after the influenzaand the key settings such as vaccination rate are determinedby the experience It is always difficult for emergency man-agement organizations to find the most optimal vaccinationrate in advanceTherefore the sensitivity of vaccination rate isanalyzed in this group of experiments The settings are listedin Table 7 average results are shown in Figure 24
Vaccination rates are set linearly (10 30 50 and70) to test the changes of influenza The peak value ofinfected agents decreases from 25135 22392 15787 6907to 638 respectively It shows that the more the agents arevaccinated the smaller the peak value of the infected agentswill be In the case of 70 vaccination the total infected agentis only 13799 It means that the influenza in this case does nothappen due to the high vaccination rate
The change of peak values also shows an interestingphenomenon Vaccination rates increase linearly the peakvalues decrease slower than the ascending of vaccination ratesat beginning but the descending becomes more and morefaster with the linear increase of vaccination rate It meansthat the small scale of vaccination (less than 30) does notmake senseWith the increase of vaccination rate the numberof susceptible agents decreases significantly In sequencethe possibility of infections is lowered greatly Additionallyit could be found that the decrease of peak values comeslater than the increase of the vaccination rates This is quitereasonable according to Mei et alrsquos work in [57]
As a result it is concluded that high vaccination rate iseffective in the control of H1N1 influenza though costly
543 The Experiments of Sensitivity Analysis of IsolationAccording to Section 512 isolation is a typical nonmedicalintervention It is the most used intervention measure in
Table 7 The initial settings of experiments of vaccination
Parameters ValueAgent count 19610000Initial infected count 40Interventions activated time(119879Intervention)
0
Vaccination rate (119877Vaccination) 0 10 30 50 70
0
Infe
cted
agen
ts
Loading intervention of vaccination
No vaccination10 vaccination30 vaccination
50 vaccination70 vaccination
0 50 100 150 200 250
25000
20000
15000
10000
5000
Day d
Figure 24 The results of loading intervention of vaccination
the emergency management in influenza As mentionedbefore a problem also exists that the measure cannot bevalidated before the intervention Therefore the sensitivityof isolation is analyzed in this group of experiments It isworth mentioning that the isolation is usually combinedwith the hospital policy It means the infected individual issent to hospital meanwhile the people who contact withthe infected are all isolated As a result parameters delay tohospital and isolation duration are designed together in thesettings of experiment Delay to hospital stands for the periodfrom infection time to hospital time while Isolation durationrepresents the period when an agent cannot contact others ifhe has contact with an infected agent The settings of theseparameters are listed in Table 8 average results are shown inFigure 25
Firstly the parameter delay to hospital is simulated aloneto test the sensitivity The simulated results are shown insolid line in Figure 25 The peak value decreases from 6708to 2218 when the delay to hospital decreases from 15 daysto 1 day The half day change brings the 67 reductionMoreover the peak value almost approximates to zero inthe case of 05 days The influenza does not happen in thiscase It can be concluded that the earlier the infected agents
Mathematical Problems in Engineering 23
Table 8 The initial settings of experiments of isolation
Parameters ValueAgent count 19610000Initial infected count 40Interventions activated time (119879Intervention) 80Delay to hospital (119879HDelay) isolation duration (119879IDuration) (15 0) (1 0) (05 0) (15 3) (1 3) (05 3)
0 50 100 150 200 2500
1000
2000
3000
4000
5000
6000
7000
Infe
cted
agen
ts
Loading intervention of isolation
Delay to hospital = 15
Delay to hospital = 10
Delay to hospital = 05
Delay to hospital = 15 isolation duration = 3
Delay to hospital = 10 isolation duration = 3
Delay to hospital = 05 isolation duration = 3
Day d
Figure 25 The results of loading intervention of isolation
are sent to hospital the less the agents will be infectedSo the construction of effective mechanism to find infectedindividuals is important in the emergency management ofinfluenza
Secondly the parameter isolation duration is added tosimulate the isolation of interventionObviously the isolationmeasure decreases the infected agents greatly Compared tothe parameter delay to hospital of 15-day case the peak valuedecreases from 6708 to 362when the isolation is added in theinterventionThe reduction of 6346 agents comes from the 3-day isolation of agents who had contacted with the infectedagents However the decrease of parameter delay to hospitalcannot bring the similar reduction in the 3-day isolationcases According to the figure the peak value decreases from362 to 51 from 15-day case to 05-day case only 311 agents aresaved from infection As a result the sensitivity of simulatedresults does not change obviously with delay to hospital inthe case of isolation It can be concluded that isolation is amore effective measure Though it is not easy for emergencymanagement organizations to decrease the delay to hospitaldue to the current monitoring mechanism the influenza canstill be controlled by isolation
544 The Experiments of Combination of Intervention Mea-sures Isolation is not the only nonmedical intervention ininfluenza close workspace is another common measureWhen close workspace is activated the workspaces are allclosed agents can only stay home at the closed time Luckilyisolation and close workspace are independent from oneanother It is possible for emergency management organi-zations to use these two measures together So this groupof experiments is designed to obtain the effectiveness of thecombination of intervention measures The parameters areinitialized with different settings listed in Table 9
Parameters delay to hospital and isolation duration is setin the middle values (1 day 3 days) according to the lastsection According to the experience close duration is set asseven days It is known that close workspace is a really costintervention measure the activated time is a key parameterWith the fixed parameters mentioned before the changeof interventions activated time is analyzed in this group ofcomputational experimentsThe average results are shown inFigure 26
Compared with the simulated results of experimentsdiscussed above Figure 26 strongly illustrates that influenzais ceased when the intervention measures are combinedtogether The peak value of infected agents is 141 in the ldquo80thdays of close-workspace and isolation with 1 day delay tohospitalrdquo case while 25135 in ldquono interventionrdquo case 2218 inldquo1 day delay to hospitalrdquo case 203 in ldquo1 day delay to hospitaland isolationrdquo case The interventions bring a great reductionof infection the peak values decrease from ten thousandslevel to hundred level
On the other hand the peak value even decreases to 108when the execution time of close workspace is 10 days earlierThe date of maximum value of infected agents is advancedfrom the 135th day to the 108th day The whole circle of H1N1influenza is reducedThe results of this group of experimentsshow that intervention measures should be activated as soonas possible and the compositions of interventions are moreeffective in emergency management of H1N1 influenza
In summary based on the general process lots of exper-iments could be done by KD-ACP to study emergency man-agement in H1N1 influenza Many conclusions are obtainedbased on the analysis of the experiments in this sectionAccording to these conclusions themost optimal response inour case study is the case of ldquo70th days of close-workspace andisolation with 1 day delay to hospitalrdquo It is because this casehas the fewest infected agents relatively But the modelling ofH1N1 influenza is not sufficient Economic and social factorsof interventions such as the money cost by vaccination andisolation the influence of the close-workspace in society
24 Mathematical Problems in Engineering
Table 9 The initial settings of experiments of isolation and closeworkspace
Parameters ValueAgent count 19610000Initial infected count 40Interventions activated time (119879Intervention) 70th day 80th dayDelay to hospital (119879HDelay) 1 dayIsolation duration (119879IDuration) 3 daysClose workspace duration (119879CWDuration) 7 days
50 100 150 200 2500
50
100
150
200
Infe
cted
agen
ts
Loading intervention of isolation and close workspace
Delay to hospital = 1 isolation duration = 3Start close time = 70 close duration time = 7Delay to hospital = 1 isolation duration = 3Start close time = 80 close duration time = 7
Day d
Figure 26The results of loading interventions of isolation and closeworkspace
and even opinion transmission of H1N1 in internet are notconsidered in our case study Without the analysis of thesefactors it does not make sense to talk about the most optimalintervention measures However KD-ACP had provided asoftware framework and general process of the research onemergency management Supported by KD-ACP the expertsof emergency management are able to study emergent eventssuch as H1N1 influenza as detailed as possible Based on theexperiments and analysis again and again the most optimalintervention measures will be found ultimately
55 Parallel Execution by ASST and OsdRT As discussed inSection 35 ASST andOsdRT are used to support the parallelexecution of artificial Beijing The statistical data is collectedand stored in artificial society runtime database at runtimeASST downloads the infection data from the database anddisplays it in the map As shown in Figure 27 the infectedagents are distributed in the districts in BeijingThe snapshotgives the geospatial distribution of the H1N1 epidemic Thelegends of the agents in different status are listed in the left
right corner in Figure 27 Five statuses are shown in the figurehealth without vaccination health with vaccination incuba-tion being symptomatic and convalescence ASST providesa very direct feeling for the emergency experts Partly withthe situation from ASST Emergency Decision Organizationmakes the emergency response plans to the response of theemergency scenario With the help of plans the epidemicin actual society can be restrained In the meantime theinfluenza information is reflected in the Internet Accordingto Figure 8 the data acquisition data extraction and datastandardization of H1N1 influenza are processed in OsdRTOsdRT is implemented by Dr Caorsquos team from Institute ofAutomation from Chinese Academy of Sciences [58] Thetool provides not only the reports of H1N1 SARS hand-foot-mouth disease and intestinal diseases are all monitored Theweb page is shown in Figure 28 H1N1 reports from Internetare collected first Based on the data extraction of the reportsH1N1 case of distribution is obtained and shown in theincidencemap As a result H1N1 incidence data is regarded asthe knowledge which is sent to EECT and IMCTThe settingsof emergent events and intervention measures are modifiedin order to make artificial society approach actual society innext turn of computational experiments With the help ofOsdRT the loop of parallel execution process integrates ACPapproach
6 Conclusion
This work provides an integrated software framework forsocial Computing in emergencymanagement From a systemperspective KD-ACP provides a reliable flexible low costand effective platform for the scientists to do the research onemergency response problems The analysis and predictionof these problems with inherent complexity can be solvedby the repetition of possible alternative experiments onthe software framework Therefore KD-ACP which is anattempt to implement ACP approach seemed as an effectivecomputational framework to support the decision making inemergency response
Currently KD-ACP is used to study the H1N1 epidemicin Beijing in 2009 The most optimal compositions of inter-vention measures are testified to support the response tonext influenza However it is still a long way ahead beforethe automation systematization and practicalization of KD-ACP A lot of work will be carried out along several directionsas in the following
(i) The parallel execution process in KD-ACP is actuallynot strongly connected with the emergency responseorganization Traditional experience and theory stillplay a dominant role in the decision of emergencymanagement It is necessary to propose a mechanismof hall for workshop to facilitate the parallel executionloop in actual society
(ii) Section 33 introduces the principle and process ofmodelling artificial society Metamodels and modelsof artificial Beijing are built by GME Model transfor-mation and code generation are used to implementcodes from FSM based models But many details
Mathematical Problems in Engineering 25
Fangshan district Daxing district Fengtai district Xicheng district
Mentougou district Haidian districtShijingshan district Dongcheng district
Incubation
Symptomatic
Convalescence
Health withoutvaccination
Health withvaccinationLegend of agent
health status
Figure 27 The situation of artificial Beijing by ASST
H1N1
Districts of Beijing
H1N1 case ofdistribution
Incidence map
Distribution ofH1N1 reportsfrom internet
Incidence of H1N1
Fangshan districtLiangxiang town
Population 19000Incidence 135789
(1100000)
Figure 28 The implementation of OsdRT by Institute of Automation from Chinese Academy of Sciences
are still not illustrated due to the topic of thispaper The key techniques such as template of codeframework and model transformations from specificdomains to FSM will be focused and introduced indetail in our next paper
(iii) The performance of computational experiments inKD-ACP is still slower than expectation The opti-mization of simulation engine and models structureis necessary to improve the performance So it ispossible to run artificial society with the real worldin parallel
In summary KD-ACP paves a new way for scientists inboth emergencymanagement and computation simulation to
collaborate with each other in solving the emergency man-agement problems by social computing
Research Highlights
(i) KD-ACP is implemented based on ACP approach
(ii) Artificial Beijing is built with the help of KD-ACP
(iii) H1N1 is modeled to simulate the influenza in Beijingin 2009
(iv) Computational experiments testify the effectivenessof intervention measures
26 Mathematical Problems in Engineering
Conflict of Interests
The authors declare that there is no conflict of interestsregarding the publication of this paper
Acknowledgment
This work was supported in part by the National ScienceFoundation of China (nos 91024030 71303252 61403402and 91324013)
References
[1] W Mao and F-Y Wang Advances in Intelligence and SecurityInformatics Academic Press Oxford UK 2012
[2] E Bonabeau ldquoAgent-based modeling methods and techniquesfor simulating human systemsrdquo Proceedings of the NationalAcademy of Sciences of the United States of America vol 99 no3 pp 7280ndash7287 2002
[3] K M Carley D B Fridsma E Casman et al ldquoBioWar scalableagent-basedmodel of bioattacksrdquo IEEETransactions on SystemsMan and Cybernetics A Systems and Humans vol 36 no 2 pp252ndash265 2006
[4] J M Epstein ldquoModelling to contain pandemicsrdquo Nature vol460 no 7256 p 687 2009
[5] Los Alamos National Laboratory Reports about EPISIMS LosAlamos National Laboratory 2003
[6] C H Builder and S C Bankes ldquoArtificial societies a conceptfor basic research on the societal impacts of informationtechnologyrdquo RAND Report 149 RAND Corp 1991
[7] F-Y Wang ldquoComputational theory and method on complexsystemrdquo China Basic Science vol 6 no 5 pp 3ndash10 2004
[8] F-Y Wang ldquoToward a paradigm shift in social computing theACP approachrdquo IEEE Intelligent Systems vol 22 no 5 pp 65ndash67 2007
[9] F-Y Wang and S Tang ldquoArtificial societies for integratedand sustainable development of metropolitan systemsrdquo IEEEIntelligent Systems vol 19 no 4 pp 82ndash87 2004
[10] D Wen Y Yuan and X-R Li ldquoArtificial societies computa-tional experiments and parallel systems an investigation ona computational theory for complex socioeconomic systemsrdquoIEEE Transactions on Services Computing vol 6 no 2 pp 177ndash185 2013
[11] F Wang F Zeng and Y Yuan ldquoAn ACP-based approach forcomplexity analysis of E-commerce systemrdquo Complex Systemsand Complexity Science vol 5 no 3 pp 1ndash8 2008 (Chinese)
[12] F-Y Wang and P K Wong ldquoResearch commentary intelligentsystems and technology for integrative and predictivemedicinean ACP approachrdquoACMTransactions on Intelligent Systems andTechnology vol 4 no 2 pp 1ndash6 2013
[13] J Sifeng X Gang F Dong and H Chunpeng ldquoStudy on theemergency rescue decision support system of petrochemicalplant based on ACP theoryrdquo in Proceedings of the 6th Manage-ment Annual Meeting (MAM rsquo11) 2011 (Chinese)
[14] X Dong D Fan G Xiong F Zhu Z Zhang and Y Yao ldquoPar-allel Bus Rapid Transit (BRT) operation management systembased on ACP approachrdquo in Proceedings of the 6th ManagementAnnual Meeting (MAM rsquo11) Beijing China 2011
[15] B Ning F-Y Wang H-R Dong R-M Li D Wen and L LildquoParallel systems for Urban rail transportation based on ACP
approachrdquo Journal of Transportation Systems Engineering andInformation Technology vol 10 no 6 pp 23ndash28 2010 (Chinese)
[16] F-Y Wang J Zhao and S-X Lun ldquoArtificial power systemsfor the operation and management of complex power gridsrdquoSouthern Power System Technology vol 2 no 3 pp 1ndash11 2008(Chinese)
[17] M J North T R Howe N T Collier and J R Vos ldquoRepastsimphony development environmentrdquo in Proceedings of theAgent 2005 Conference on Generative Social Processes Modelsand Mechanisms 2005
[18] M J North T RHowe N T Collier and J R Vos ldquoRepast Sim-phony runtime systemrdquo in Proceedings of the Agent Conferenceon Generative Social Processes Models and Mechanisms 2005
[19] SWARM Development Group Swarm 211 Reference Guide2000 httpwwwswarmorg
[20] S Tisue and U Wilensky ldquoNetLogo a simple environment formodeling complexityrdquo in International Conference on ComplexSystems pp 16ndash21 2004
[21] S Luke C Cioffi-Revilla L Panait K Sullivan and G BalanldquoMASON a multiagent simulation environmentrdquo Simulationvol 81 no 7 pp 517ndash527 2005
[22] M Ichikawa H Tanuma Y Koyama andH Deguchi ldquoSOARSintroduction as a social microscope for simulations of socialinteractions and gamingrdquo in Organizing and Learning throughGaming and Simulation Proceedings of the 38th Conference ofthe International Simulation and Gaming Association (ISAGArsquo07) pp 149ndash158 2007
[23] D S Burke J M Epstein D A T Cummings et al ldquoIndividual-based computational modeling of smallpox epidemic controlstrategiesrdquo Academic Emergency Medicine vol 13 no 11 pp1142ndash1149 2006
[24] F-Y Wang ldquoParallel control and management for intelligenttransportation systems concepts architectures and applica-tionsrdquo IEEE Transactions on Intelligent Transportation Systemsvol 11 no 3 pp 630ndash638 2010
[25] Z Molnar D Balasubramanian and A Ledeczi ldquoAn introduc-tion to the genericmodeling environmentrdquo in Proceedings of theTOOLS Europe 2007 Workshop on Model-Driven DevelopmentTool Implementers Forum Zurich Switzerland 2007
[26] S Kelly and J-P Tolvanen Domain Framework in Domain-Specific Modeling Enabling Full Code Generation JohnWiley ampSons Hoboken NJ USA 2007
[27] B P Zeigler H Praehofer and T G Kim Theory of Modelingand Simulation Integrating Discrete Event and ContinuousComplex Dynamic Systems Academic Press San Diego CalifUSA 2000
[28] G Booch J Rumbaugh and I JacobsonThe Unified ModelingLanguage User Guide Addison-Wesley Reading Mass USA1999
[29] F Wagner R Schmuki T Wagne and P Wolstenholme Mod-eling Software with Finite State Machines A Practical ApproachCRC amp Taylor amp Francis Boca Raton Fla USA 2006
[30] T Espiner ldquoChina builds worldrsquos fastest supercomputerrdquo ZDNetUK October 2010
[31] B Chen and G Guo ldquoA two-tier parallel architecture for arti-ficial society simulationrdquo in Proceedings of the ACMIEEESCS26th Workshop on Principles of Advanced and DistributedSimulation (PADS rsquo12) pp 184ndash186 July 2012
[32] G Guo B Chen X G Qiu and Z Li ldquoParallel simulation oflarge-scale artificial society on CPUGPU mixed architecturerdquo
Mathematical Problems in Engineering 27
in Proceedings of the ACMIEEESCS 26th Workshop on Prin-ciples of Advanced and Distributed Simulation (PADS rsquo12) pp174ndash177 July 2012
[33] S-X Lun ldquoResearch on the classification of parallel executionmodes of ACP theoryrdquo Acta Automatica Sinica vol 38 no 10pp 1602ndash1608 2012
[34] F-YWang ldquoA framework for social signal processing and anal-ysis from social sensing networks to computational dialecticalanalyticsrdquo Science China vol 43 no 12 pp 1598ndash1611 2013
[35] G GuoOne Model Userrsquos Manual version 12 National Univer-sity of Defense University 2011
[36] Y Ge R Meng Z Cao X Qiu and K Huang ldquoVirtual citymdashanindividual-based digital environment for human mobility andinteractive behaviorrdquo Simulation Transactions of the Society forthe Modeling and Simulation International 2014
[37] M Kretzschmar and R T Mikolajczyk ldquoContact profiles ineight European countries and implications for modelling thespread of airborne infectious diseasesrdquo PLoS ONE vol 4 no6 Article ID e5931 2009
[38] W Duan and X Qiu ldquoFostering artificial societies using sociallearning and social control in parallel emergency managementsystemsrdquo Frontiers in Computer Science vol 6 no 5 pp 604ndash610 2012
[39] W J Edmunds C J OrsquoCallaghan and D J Nokes ldquoWhomixes with whom Amethod to determine the contact patternsof adults that may lead to the spread of airborne infectionsrdquoProceedings of the Royal Society B Biological Sciences vol 264no 1384 pp 949ndash957 1997
[40] J Banks J S Carson B L Neison and D M Nicol Discrete-Event System Simulation Prentice Hall Englewood Cliffs NJUSA 4th edition 2007
[41] H Yongxia ldquoInvestigation and analysis of college studentsrsquo workand restrdquo Value Engineering vol 31 no 3 2012 (Chinese)
[42] W Duan Z Cao Y Wang et al ldquoAn ACP approach to publichealth emergency management using a campus outbreak ofh1n1 influenza as a case studyrdquo IEEE Transactions on SystemsMan and Cybernetics Part A Systems and Humans vol 43 no5 pp 1028ndash1041 2013
[43] HWHethcote ldquoThemathematics of infectious diseasesrdquo SIAMReview vol 42 no 4 pp 599ndash653 2000
[44] G Chowell E Shim F Brauer and et al ldquoModelling thetransmission dynamics of acute haemorrhagic conjunctivitisapplication to the 2003 outbreak in Mexicordquo Statistics inMedicine vol 25 no 11 pp 1840ndash1857 2006
[45] Y Ge L Liu B Chen X Qiu and K Huang ldquoAgent-basedmodeling for InfluenzaH1N1 in an artificial classroomrdquo SystemsEngineering Procedia vol 2 pp 94ndash104 2011
[46] A R Tuite A L Greer MWhelan et al ldquoEstimated epidemio-logic parameters andmorbidity associatedwith pandemicH1N1influenzardquo CMAJ vol 182 no 2 pp 131ndash136 2010
[47] L Brouwers B Cakici M Camitz A Tegnell and MBoman ldquoEconomic consequences to society of pandemic H1N1influenza 2009mdashpreliminary results for Swedenrdquo Euro Surveil-lance vol 14 no 37 pp 1ndash7 2009
[48] Y Ge L Liu X Qiu H Song Y Wang and K Huang ldquoAframework of multilayer social networks for communicationbehavior with agent-based modelingrdquo Simulation Transactionsof the Society for the Modeling and Simulation International vol89 no 7 pp 810ndash828 2013
[49] W Duan X Qiu Z Cao X Zheng and K Cui ldquoHeterogeneousand stochastic agent-based models for analyzing infectious
diseasesrsquo super spreadersrdquo IEEE Intelligent Systems vol 28 no4 pp 18ndash25 2013
[50] The Preparation and Emergency Response Plans to PandemicPublished by Ministry of Health Ministry of Health 2005(Chinese)
[51] R M Fujimoto Parallel and Distributed Simulation SystemsWiley Interscience 2000
[52] F Cicirelli A Giordano and L Nigro ldquoDistributed simulationof situated multi-agent systemsrdquo in Proceedings of the 15thIEEEACM International Symposium on Distributed Simulationand Real Time Applications (DS-RT rsquo11) pp 28ndash35 September2011
[53] B Chen and G Guo ldquoA two-tier parallel architecturefor artificial society simulationrdquo in Proceedings of the 26thACMIEEESCS Workshop on Principles of Advanced and Dis-tributed Simulation (PADS rsquo12) pp 184ndash186 July 2012
[54] G Guo B Chen and X Qiu ldquoParallel simulation of large-scale artificial society with GPU as coprocessorrdquo InternationalJournal of Modeling Simulation and Scientific Computing vol4 no 2 Article ID 1350005 2013
[55] B Chen L-B Zhang X-C Liu and H Vangheluwe ldquoActivity-based simulation using DEVS increasing performance by anactivity model in parallel DEVS simulationrdquo Journal of ZhejiangUniversity Science C vol 15 no 1 pp 13ndash30 2014
[56] X-L Wang P Yang Z-D Cao et al ldquoQuantitative evaluationon the effectiveness of prevention and control measures againstpandemic influenza A (H1N1) in Beijing 2009rdquoChinese Journalof Epidemiology vol 31 no 12 pp 1374ndash1378 2010 (Chinese)
[57] S Mei A D van de Vijver L Xuan Y Zhu and P M ASloot ldquoQuantitatively evaluating interventions in the influenzaA (H1N1) epidemic on China campus grounded on individual-based simulationsrdquo Procedia Computer Science vol 1 no 1 pp1675ndash1682 2010
[58] X Qiu Z Cao and Z Li Mid-Term Reports of China NationalScience Foundation 91024030 2014
Submit your manuscripts athttpwwwhindawicom
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
MathematicsJournal of
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Mathematical Problems in Engineering
Hindawi Publishing Corporationhttpwwwhindawicom
Differential EquationsInternational Journal of
Volume 2014
Applied MathematicsJournal of
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Probability and StatisticsHindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Journal of
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Mathematical PhysicsAdvances in
Complex AnalysisJournal of
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
OptimizationJournal of
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
CombinatoricsHindawi Publishing Corporationhttpwwwhindawicom Volume 2014
International Journal of
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Operations ResearchAdvances in
Journal of
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Function Spaces
Abstract and Applied AnalysisHindawi Publishing Corporationhttpwwwhindawicom Volume 2014
International Journal of Mathematics and Mathematical Sciences
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
The Scientific World JournalHindawi Publishing Corporation httpwwwhindawicom Volume 2014
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Algebra
Discrete Dynamics in Nature and Society
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Decision SciencesAdvances in
Discrete MathematicsJournal of
Hindawi Publishing Corporationhttpwwwhindawicom
Volume 2014 Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Stochastic AnalysisInternational Journal of
2 Mathematical Problems in Engineering
a novel approach in social computing to solve the problemsin society domain ACP approach is categorized into threeaspects representing and modelling society with artificialsystems analysis and evaluation by computational experi-ments and control andmanagement of real society by parallelexecution Under the instruction of conceptual framework ofACP approach a wide spectrum of complex systems such astransportation medicine finance and environment can bestudied in the computational manner Actually many real-world applications using ACP approach have been developedto solve the real problems in domains For instance complexsocioeconomic system [10] and the research framework fore-commerce system [11] are the good applications of ACPapproach in the economic area The ACP-based frame-work for integrative medicine [12] is proposed to solve theproblems in medicine An overall framework of emergencyrescue decision support system of petrochemical plant [13]is proposed based on ACP theory to study environmentrisk accidents of petrochemical plant Parallel BRT operationmanagement system [14] based on ACP approach has beenconstructed to detect the quantity of passengers on stationsreal-time traffic flow on stations or at intersections andqueuing length of vehicles on the road A novel parallel sys-tem for Urban Rail Transportation (URT) [15] based on ACPapproach is proposed to address issues on safety efficiencyand reliability of the operation of URT An artificial powersystem [16] is set up on the models of power systems andcomplex power grids to provide a feasible approach for thecontrol and management of the modern power system
Although a lot of work has been done on the concepts andtheory framework to study problems by social computing thefollowing problems of computational experiments in emer-gencymanagement are still not solved from the perspective ofmodelling for social systems theory and software frameworkof platform implementation
(i) The modelling and simulation of emergency man-agement are not given the special considerationThe representation of society focuses on the genericmodelling of agent (represented by Repast [17 18])The description of environments is too simple tomeet the requirements from research on emergencyproblems such as the building size the place relatedagent contact frequency
(ii) The existing tools and platforms cannot support thedesign of experiment Computational experimentscannot be done systematically and automatically
(iii) The existing applications of ACP-based frameworksare still domain specific A generic workflow andintegrated toolkit are needed to implement ACPapproach especially in the application of emergencymanagement
Therefore it is necessary to develop an ACP-based softwareframework for the research on emergency management Theartificial system is the projection of real world in the emergentscenarioThemodelling of the system including the emergentevents modelling and intervention measures modelling thedesign of computational experiments considering the settings
of emergency parameters the settings of large samples exper-iments and the parallel execution with loose connection ofreal society should be considered inside the software frame-work
As a result the purpose of this paper is to proposea software framework called KD-ACP applying the ACP-based computational theory and corresponding methods instudying emergency management problems KD is short forthe Chinese phonetic alphabets of China National Univer-sity of Defense Technology KD-ACP means the softwareframework is developed by National University of DefenseTechnology to implement ACP approach The remainder ofthis paper is organized as follows Section 2 summarizes theexisting agent-based modelling and simulation platformsSection 3 introduces ACP approach first and proposes theKD-ACP platform Section 4 illustrates the modelling ofBeijing city with KD-ACP both the agent models and initialdata are considered Section 5 shows how to do experimentswith KD-ACP using the H1N1 case study in artificial BeijingIn the end the paper is concluded in Section 6
2 Related Works
There have beenmany efforts on social computing especiallyon the emergency management Agent-based modelling andsimulation are popular in the implementation of socialcomputation The related works are mainly categorized intoSWARM-like agent-based modelling and simulation plat-forms and agent-based platforms for emergency manage-ment
21 SWARM-Like Agent-Based Modelling and SimulationPlatforms SWARM [19] originally proposed by Santa Feinstitute is widely used in many research areas such asbiology ecology and society The tool provides a simulationenvironment for simulating agent with the support of a seriesof class libraries It is worth noting that SWARM is the pre-cursor ofmultiagent simulation tool it influences lots ofmul-tiagent simulation platforms such as NetLogo [20] RePast(REursive Porus Agent Simulation Toolkit) MASON [21]and SOARS (Spot Oriented Agent Role Simulator) [22]
NetLogo is a multiagent programming language andmodelling environment for simulating natural and socialphenomena It is particularly well suited for modelling com-plex systems evolving over timeThe language is easy to studyand the agent-based complex systems could be built rapidlyRePast is a software framework for agent-based simulationcreated at theUniversity of Chicago An extensible simulationpackage makes RePast become a generic multiagent simula-tion platform in social science research computing MASONdesigned by George Mason University is used to serve as thebasis for a wide range of multiagent simulation tasks rangingfrom swarm robotics to machine learning to social complex-ity environments The tool is a fast discrete-event multiagentsimulation toolkit in Java SOARS is designed by TokyoInstitute of Technology to describe agent activities under theroles of social and organizational structureDecomposition ofmultiagent interaction is the most important characteristicsin this framework
Mathematical Problems in Engineering 3
All the SWARM-like agent-based modelling and simu-lation platforms provide a portable lightweight and easilyextensible environment for simulating agents in arbitraryresearch areas However the heterogeneity in specific socialcomputing domain is not considered Furthermore most ofthe platforms cannotwell support large scale agent simulationbecause of the lightweight engine The engine cannot affordthe simulation of super cities like Beijing andNewYorkwhichhave millions of people
22 Agent-Based Platforms for Emergency ManagementBiowar proposed by Carnegie Mellon University simulatesthe impact of background diseases bioterrorism attackswithin a city 62 diseases are modeled in this platform tosimulate the outbreaks on the populationrsquos behavior GASM(Global-Scale Agent Model) is designed to study the spread-ing of H1N1 65 billion population is modeled with thesupport of official statistical data A global H1N1 spread fromTokyo is simulated in GASM EpiSimS proposed by LosAlamos lab simulates the spread of disease in regions suchas cities allowing for the assessment of disease preventionintervention and response strategies The daily movementsand interactions of synthetic individuals are representedexplicitly Burke and Epstein propose a computational modelof smallpox epidemic transmission and control [23] Theagents in thismodel interact locally with one another in socialunits such as homes workplaces schools and hospitals
However these platforms cannot provide a generic soft-ware framework to study emergency problems Biowar onlyfocuses on social networks individuals are all modeled as thenodes of social networks Agents in GASM and EpisimS areisomorphic without considering the heterogeneity in specificdomains
To sum up this section briefly reviews the existing mul-tiagent simulation platforms including SWARM-like agent-based modelling and simulation platforms and agent-basedplatforms for emergency management However they can-not satisfy the requirements of simulation performanceadaptability of software framework and heterogeneity ofindividuals in research of different emergency scenarios
3 KD-ACP
KD-ACP is an integrated software framework designed andimplemented based on the principle of the ACP approachshown in Figure 1
31The ACPApproach ACP approach is a social computing-based research paradigm It is composed of three componentsas its name artificial society for A computational experi-ments for C and parallel execution for P The basic idea ofACP approach is listed as follows
(i) Model the complex societies involving human behav-ior and social organizations as artificial societies usingmultiagent modelling techniques in a ldquobottom-uprdquofashion Artificial societies are regarded as a researchplatform to study emergency management
Actual societies Artificial societies
Managementand control
Experimentationand evaluation
Training andlearning
TestingTesting Operation
Figure 1 The parallel execution of ACP approach [8 33]
(ii) Utilize innovative computing technologies to evaluateand analyze various factors in emergency manage-ment quantitatively the computers are regarded asthe experimental social laboratories for investigatingemergency management problems
(iii) Provide an effective mechanism for the control andmanagement of complex actual social society throughcomparison evaluation and interaction with artifi-cial society
It is worth notifying that ldquoPrdquo here is not the ldquoparallelrdquo inldquoparallel simulationrdquo but the representation of ldquoparallel exe-cutionrdquo The idea of parallel execution is to build the parallelscenarios by paralleling the actual societies and artificialsocieties Consequently parallel control and management ofactual societies are implemented with the help of interactionsbetween parallel scenarios The goal of parallel executionis to find the best plans to adjust the methods of controland management based on the comparison and analysis ofdifferences between actual and artificial societies Artificialsocieties provide possible simulated results of evolutions byrepeated computational experiments The simulated resultsprovide evidences for the adjustment plans These plans areused in the control and management of actual societies suchas emergency management After the application of theseplans the observations from actual societies are collected forthe comparison with expectation The differences are used tofeedback to artificial societies The new turn of comparisonand analysis to find best adjustments of control and manage-ment is repeated
The mechanism of ldquoparallel executionrdquo has been provedto be effective for use in networked complex traffic systemsand is closely related to emerging technologies in cloud com-puting social computing and cyber-physical-social systems[24] In order to promote the development of parallel controland management in emergency management the artificialsociety is proposed in ACP approach which is the expansionof ldquoartificial traffic systemsrdquo
Instructed by the ACP approach KD-ACP is also com-posed by three components The details of the architectureand implementation of KD-ACP are discussed below
32 The Software Architecture of KD-ACP The architectureof KD-ACP is shown in Figure 2 the software framework iscomposed of a series of tools These tools are grouped intothree parts to support artificial society modelling computa-tional experiments and parallel execution
4 Mathematical Problems in Engineering
A part
C part
P part
Population and GeospatialEnvironment Generation Tool (PGET)
Generic ModelingEnvironment (GME)
Artificial society population andgeospatial environment database
Initialize
InitializeInitialize
Model DevelopmentTool (MDT)
FSM based models
Artificial SocietyEditor (ASE)
Scenario ofartificial society
Artificial societyruntime database
Agent and emergencymodel repository
Experiments DesignTool (EDT)
Experiment plans
InitializeSettings of
emergent eventsEmergent Events
Configuration Tool (EECT)Intervention Measures
Configuration Tool (IMCT) Settings of intervention measuresemergency response plans
Experiments ManagementTool (EMT)
Population data emergent eventsdata and intervention measures
Open source dataRegistration Tool (OsdRT)
Actu
al so
ciet
y Artificial SocietySituation Tool (ASST)
Optimal emergencyresponse plan
Internet
C++ code of modelsdll of models
Figure 2 The software architecture of KD-ACP
In the ldquoArdquo part Generic Modeling Environment (GME)[25] and Model Development Tool (MDT) are the kerneltools in the modelling of artificial society GME is anopen source modelling tool which supports domain-specificmodelling The domains of artificial society are created byGME in our work Models such as agent environmentemergent event and intervention measure are described inspecific domains first in GME With the help of modeltransformation thesemodels are all transformed to the FiniteState Machine (FSM) models Meanwhile code generationsare supported byMDT and thesemodels are all implementedin C++ Artificial Society Editor (ASE) is used to describethe concrete scenario of actual society which defines thescope of models set for artificial society Population andGeospatial Environment generation Tool (PGET) generatesthe population and geospatial environment data with thesupport of statistical data from actual society
In the ldquoCrdquo part Emergency Events Configuration Tool(EECT) initializes the models of emergent events whileInterventionMeasures Configuration Tool (IMCT) initializesthe models of intervention measures Experiments plans aregenerated byExperimentsDesignTool (EDT) Based on theseplans Experiments Management Tool (EMT) is used to runand manage the computational experiments to study theemergency problems
In the ldquoPrdquo part Artificial Society Situation Tool (ASST)seemed as the monitor of running artificial society Thestatistical data and situation are shown by ASST at runtimeIn the meantime the emergency response plans are made by
emergency decision organizations Parts of the influences ofemergency plans are reflected on Internet Open source dataRegistration Tool (OsdRT) is used to register the open sourcedata from Internet to artificial society
KD-ACP is developed using the BrowserServer archi-tecture the tools are integrated in the home page of KD-ACP as shown in Figure 3 Each tool is activated by the clickon the link For example Artificial Society Editor is startedwhen the link of ASE is clicked The working environmentand programming languages of tools in KD-ACP are listed inTable 1
Moreover the implementation of KD-ACP is mainlycomposed of modelling phase and computational experi-ments phase It will be discussed in the next section
33 The Modelling of Artificial Society in KD-ACP It is acritical problem to focus on the key parts of society in socialcomputing Based on the ACP approach the bottom-upmodelling is used to build the artificial society As a resultmodelling of artificial society is composed of three basic ele-ments agents environments and rules for interactionsHow-ever we still meet the problem that specific features shouldbe supported in artificial society For example emergentevents and intervention measures are required in artificialsociety for emergency management The modelling of onlybasic elements cannot cover the specific features in domainsTherefore domain-specific modelling [26] is introduced tosolve the problems in modelling artificial society
Mathematical Problems in Engineering 5
Model DevelopmentTool (MDT)
Artificial SocietyEditor (ASE)
Population and GeospatialEnvironment Generation Tool (PGET)
Emergent EventsConfiguration Tool (EECT)
Intervention MeasuresConfiguration Tool (IMCT)
Experiments DesignTool (EDT)
Experiments ManagementTool (EMT)
Artificial SocietySituation Tool (ASST)
Open Source DataRegistration Tool (OsdRT)
Generic ModelingEnvironment (GME) Artificial
society (A)Computationalexperiments (c)
MDT EECT IMCT
ASST OsdRT
ASE PGET EDT EMT
Parallel execution (P)
Figure 3 The implementation of KD-ACP
Table 1 The working environment and programming languages of tools
Tools Working environment Programminglanguages
Developmentplatform User type
Generic ModelingEnvironment (GME)
General computer (desktopapplication) NULL NULL Domain experts
Model Development Tool(MDT)
General computer (desktopapplication) C++ Visual Studio Model developers
Artificial Society Editor(ASE)
General computer (desktopapplication) C Visual Studio Domain experts
Population and GeospatialEnvironment generationTool
General computer (desktopapplication) C Visual Studio Domain experts
Emergency EventsConfiguration Tool (EECT)
General computer (withInternet Explorer clientside)
C ASPNET Domain experts
Intervention MeasuresConfiguration Tool (IMCT)
General computer (withInternet Explorer clientside)
C ASPNET Domain experts
Experiments ManagementTool (EMT)
General computer (withInternet Explorer clientside)
C ASPNET
Users of computationexperimentServer of EMT
Nodes in supercomputer(Console Program serverside)
C++ Visual Studio
Runtime Infrastructure ofEMT
Nodes in supercomputer(Console Program serverside)
C++ Visual Studio
Artificial Society SituationTool (ASST)
General computer (desktopapplication) C++ Visual Studio Domain experts
Open Source DataRegistration Tool (OsdRT)
General computer (withInternet Explorer clientside)
Java JSP Domain experts
6 Mathematical Problems in Engineering
M2T transformation
Modeling in domain
GME Generic Modeling Environment
Metamodeling
Instance of
FSM
Metamodeling for artificial society
DEVS
Petri Net State charts
Agentmetamodel
Environmentmetamodel
Emergency eventmetamodel
Interventionmetamodel
DAE
Domain-specific metamodels Semantic well definedmetamodels
M2M transformation
Modeling of artificial society
Interventionmodel
Emergency eventmodel
Environmentmodel
Agentmodel
FSM based emergent andintervention models
FSM based environmentmodels
FSM based agent modelsDomain-specific models
Semantic well defined models
M2M transformation
Agent model code Environment model code
Agent behavior model
Agent mentality model
Disease model code
Basic population model Disease spreading model
Disease state transformmodel
Intervention model code
Agent social networkmodel
Basic environmentinformation model
Roadnet model
Vaccination
Infectors isolation
Contactors isolation
Close work officesAgent contact model
Domain-specific models relevant codes
MDT Model Development Tool
Implementation
Figure 4 From metamodelling and modelling by GME to implementation by MDT of artificial society
331 The Principle of the Modelling of Artificial SocietyAccording to the principle of domain-specific modellingthe modelling of artificial society contains the followingsteps first metamodelling the basic elements of artificialsociety second modelling the specific features in domain ofemergency management third implementing the models ofartificial society in codes The whole process is illustrated inFigure 4The first and second steps are implemented in GMEwhile the third step is implemented in MDT
The first step is metamodelling which mainly focuses onconstructing the metamodels of artificial society Metamod-elling tries to study the common patterns of artificial societyThe outputs of metamodelling are metamodels which repre-sents the abstraction of the whole systemThe basic elementsof artificial society are described in metamodel The processof metamodelling is divided into four phases The first is theconstruction of the domain-specific metamodels As shown
in Figure 4 agent metamodel environment metamodelemergent event metamodel and intervention metamodelcompose the metamodel of artificial society The second isthe construction of the metamodels described by typicalmodelling formalisms such as FSM DAE DEVS and PetriNet [27] These formalisms are all semantically well definedThe third is the definition of the model transformationfrom domain-specific metamodels to metamodels of typicalmodelling formalisms The transformation standardizes themetamodels of artificial society by typical modelling spec-ifications The fourth is the definition of the transforma-tion templates from metamodels to code framework Thetemplates list the basic abstract interfaces of metamodels ofartificial societyThese abstract interfaces are implemented inthe specific-domain modelling and code generations
The second step is modelling the models of artificialsociety such as agent model environment model emergent
Mathematical Problems in Engineering 7
Figure 5 The metamodels of artificial society in GME
event model and intervention model are built Actuallythe models are the instantiation of metamodels in the laststep Different from the general modelling environment likeUML [28] the domain-specific modelling provides a familiarmodelling environment for the domain experts in artificialsociety For example emergency response experts only con-cern emergent eventmodel and interventionmodel inheritedfrom metamodels After constructing the domain-specificmodels based on domain-specific metamodels domain usersexecute the model transformation defined in the first stepAll the models of artificial society are transformed into FSMmodels As a result the models are implemented in this uni-fied modelling formalism (FSM) The model transformationmakes the simulation of the models possible
The third step is the generation of executable codes ofmodels The executable code framework is generated bymapping template from metamodels to code frameworkdefined in the first step Moreover domain developers alsoadd necessary codes to the framework to integrate thedynamic semantics of the models The code frameworkoutputs the dynamic link libraries by compilingThe dynamiclink libraries are loaded in the large scale artificial societyruntime infrastructure in computational experiments
332 The Metamodelling and Modelling of Artificial Societyby GME GME is used to build metamodels and modelsin our work As mentioned before the abstraction andcommon patterns of society are represented in metamodelsAccording to the bottom-up modelling style metamodels ofagent environment and communications are described inGME Figure 5 shows part of metamodels of artificial societyThe features of an agent metamodel are extracted from thecensus figures and statistical data Environment metamodelsimulates the geospatial places for the behaviors of agentsThemetamodel of communications among agents ismodeledto simulate the interactions such as infection in epidemicsand rumor propagation in public opinion formation eventsIt is worth notifying that the metamodel of communications
includes both the emergent event metamodel and interven-tion metamodel
From the perspective of modelling the details from spe-cific domains are considered in the models by the instantia-tion frommetamodels of artificial society For example socialrelationships based on complex networks are added in agentmodel to support the communications Agent activity is alsoused to quantify the agent activity under different scenarioEnvironment models are linked with the help of transporta-tion services subways and roads are modeled while the pathsearch is encapsulated in the services Emergent event modeland intervention model are also the domain-specific modelsThe modelling of artificial Beijing in GME will be discussedin detail in next section
333 The Implementation of Models of Artificial Society byMDT As mentioned before MDT is used to implementmodels such as agent environment emergent event andintervention According to the template of code frameworkthe implementations of models are generated by MDT Theimplementations are classified into two categories FSMmodels and services FSM models such as agents andenvironments are built under the specification of FiniteState Machine (FSM) [29] in MDT while all the servicessuch as transportation are encapsulated under the PublicService Standard This standard provides a generic interfacespecification for modelers to encapsulate public commonservices in artificial society FSM models like agent are builtstatistically from the quantitatively analyzed characteristicssuch as demographic attributes social behaviors emergencybehaviors and social networks Social behaviors describe thedaily behaviors of individuals while emergency behaviorsdescribe the individual behaviors in emergent events Forexample infected individuals are all isolated in hospital inSARS Isolation ismodeled as a typical emergency behavior inour work Correspondingly services are used to simulate themacroactual society Take transportation service for instancethe path search is needed by almost every agent duringmoving from spot to spot
8 Mathematical Problems in Engineering
Artificial society description in emergency
Description ofemergency
ArtificialSociety Editor
(ASE)
Agent behavior set
Agent role setAgent status set
Environment setAgent relationship set
Statistical data of real society
Descriptionof artificial
societyframework
Terminal of agent modeling andartificial society describing
middot middot middotAgent
metamodeling
Customized agentmodeling
Relationship modelDailyEmergency
Mod
el D
evelo
pmen
t Too
l (M
DT)
Agent modeling
Agent model Encapsulate
Encapsulate
Behavior model
Intervention
Emergency
Response
Daily
Social relationshipmodeling
Socialbehaviormodeling
Socialbehaviormodel
Emergencybehavior
model
Emergencybehaviormodeling
Building
Geospatial statistical data
Population statistical data Datacollection
Actualsociety
Artificial societypopulation and
geospatial database
Organizationservice
Transportationservice+
MappingAgent model andservice repository+ emergency model
repository
Climateservice
Geographicalservice
+
Agent status in emergencyEmergency management
Emergent event set
Social relationship statistical data
Environment statistical data
Population behaviorstatistical data
Emergency statistical data
Emergency organization statistical data
Geospatial and SocialEnvironment Generation
Tool (GSeT)
Agent model repository andpopulation database
C++C++C++C++
Modeling
Statistical modelJob ageSex location
Public service modeling
Organizationmodel
Transportationmodel
Geographicalmodel
Climatemodel
Interventionservice
Publ
ic se
rvic
e sta
ndar
d
Figure 6 The editing and initialization of KD-ACP
MDT provides domain experts with a tool to obtain thecode implementations of models With the help of compileenvironment like visual studio MDT also supports thefurther programming development of the specific domaindetails which cannot be described in modelling step
Both the FSM models and services are developed by theMDT first and then stored in the agent model and servicerepository The repository manages the models according tothe requirements from emergency problems and provides themodels for EDT to make the experiment plans
34 The Editing and Initialization of Artificial Society byASE and PGET As shown in Figure 6 ASE is used to editthe scenarios of artificial society within emergent eventsThe editing is composed of two parts (1) the statisticalinformation of artificial society in daily life such as the rolesof agents the relationships of agents and the types of environ-ments and (2) the statistical information of artificial societyin emergency including the statistical data of emergent
events the emergency organization and emergency relatedbehaviors of agents
According to the requirements of the editing thesestatistical data are collected from actual society manually bythe domain experts Based on these statistical data PGETgenerates the artificial society population and geospatialenvironment database The database supports the instanti-ation of artificial society at individual level For examplethe attributes such as age and gender of each agent can befound in the database With the support of the databaseFSM models and service repository discussed before itis sufficient for domain experts to study the emergencyproblems by computational experiments
35The Computational Experiments and Parallel Execution inKD-ACP The tools of ldquoCrdquo part and ldquoPrdquo part in KD-ACP areused to support the process of computational experimentsand parallel execution The working process is shown inFigure 7 EECT and IMCT are both the starting and ending
Mathematical Problems in Engineering 9
Agent model andservice repository
repository
Emergency EventsConfigurationTool (EECT)
Open Source DataRegistration Tool (OsdRT)
Intervention MeasuresConfigurationTool (IMCT)
Load
Output Output
Internet
Emergencyresponse plans
Load
∙ Input models of emergencyevents∙ Output settings of agentbehavior in emergent events+ settings of emergent events
∙ Input models of interventionmeasures∙ Output settings of agentbehavior in interventionmeasures + settings ofintervention measures
+ emergency model
∙ Input data from internetnetworks∙ Output population data +emergency events data +intervention measures data
Artificial societypopulation and
geospatialenvironment database
Artificial societyruntime database
OutputThe most optimal
emergencyresponse plan
Actual society
OutputOutputOutput
Input
Load
Load
Load
Load
Output
Artificial SocietySituation Tool (ASST)
Emergency decisionorganization
ClusterRuntime infrastructure
Deploy
Deploy
Experimentsdesign tool (EDT)
Artificial societyinitial data file
Experimentplan
Artificial societymodels
Experiments Management
large scale artificial societyruntime infrastructure
Tool(EMT) +super computerTIANHE-1A
Load
∙ Input experiment plan +artificial society models +artificial society initial data file∙ Output artificial societyruntime state data
∙ Input description of artificialsociety + artificial society populationand geospatial environment society+ emergent events configuration +
intervention measures configuration∙ Output experiment plan + artificialsociety models + artificial societyinitial data file
∙ Input runtime statistical data∙ Output graphics charts ofartificial society statistical data +situation of artificial society
Figure 7 The computational experiments and parallel execution of KD-ACP
point The emergent events and intervention measures areconfigured by EECT and IMCT respectively The configu-rations of emergent events are used to simulate both thereal emergencies like SARS and H1N1 and the supposedemergencies for experiments Similarly the configurationsof intervention measures are also used to reproduce thereal one and simulate the supposed one The repeat of theemergency is used to verify the models while the supposedconfigurations are used to obtain the optimized decision planto the response of the possible emergencies
With the input of artificial society model and servicerepository artificial society population and geospatial envi-ronment database and the configurations discussed beforeEDT generates the experiment plans to meet the require-ments of research on emergency management The outputof EDT includes artificial society models artificial societyinitial data files and experiment plan The models aredownloaded from the repository while the data file is thecollection of data from the database to initialize the modelsWhen the models and data files are ready EMT loads theexperiment plan and deploys the models and data to thecluster or TIANHE-1A supercomputer [30] which was theworldrsquos fastest supercomputer built by National University of
Defense Technology (NUDT) in China in 2010 According tothe plan the experiments are done repeatedly on the largescale artificial society runtime infrastructure [31 32] by themultisample settingsThework process is the implementationof computational experiments in ACP approach
Traditionally emergency response plans are made byemergency management theories and experiences The onlyway to test the effective of plans is the feedback results of realworld ACP approach provides a novel method to supportemergency response plan making by parallel execution Asshown in Figure 7 the work process of KD-ACP is composedof two loops The inner loop composed by red arrowsdescribes the process of computational experiments while theouter loop of yellow arrows illustrates the process of parallelexecution During the runtime of computational experi-ments the statistical data of artificial society is collected andstored in the artificial society runtime database Based on thedatabase ASST outputs the customized situation of runningartificial society by graphics charts and situation mapsThe information is sent to the organizations of emergencydecision to support making the emergency response plansWith the help of computational experiments loop these plansare simulated repeatedly to find the most optimal one
10 Mathematical Problems in Engineering
Open Source Data Registration Tool (OsdRT)
Emergent EventsConfiguration Tool (EECT)
Intervention MeasuresConfiguration Tool (IMCT)
∙ Web information collecting
∙ Deep mining in social media
∙ Web information denoising
∙ Information filtering
Data acquisition
Actual society
Open source informationin social media
Internet
∙ Basic element extracting∙ Individuals and organizationsextraction∙ Sentiment analyzing andopinion mining∙ Social network analyzing
Data extraction
∙ Domain knowledge description∙ Information fusion and collisiondetection∙ Domain knowledge construction∙ Social media informationstandardization
Data standardization
Knowledge
Figure 8 The compositions of OsdRT [34]
Moreover the most optimal plan is used to the responseof emergency in actual society According to the idea ofparallel control in [24] the feedback of actual society ispartly collected from Internet networks by OsdRT As shownin Figure 8 OsdRT is composed of three components dataacquisition data extraction and data standardization Dataacquisition collects mines and filters information fromsocial sensing networks Data extraction includes basicelement extracting individual and organization extractingsentiment analyzing and social networks analyzing Datastandardization specifies the useful knowledge and sent it toconfiguration tools in KD-ACP
By processing in OsdRT the knowledge about emergentevents and intervention measures are analyzed first andregistered in the EECT and IMCT The registration updatesthe settings of emergent events and intervention measuresThis loop composed of yellow arrows implements the parallelexecution in ACP approach The implementation of parallelcontrol and management provides a data-driven approachthat considers both the engineering and social complexityfor modelling analysis and decision making in emergencymanagement
4 Modelling Beijing City with KD-ACP
41 How to Build the Artificial Beijing According to themodelling of artificial society in KD-ACP discussed beforethe Beijing city is modeled as follows
To meet the requirements of emergency managementsthe basic elements of artificial society are extended Asshown in Figure 9 six elements are required to simulate thecity agents environments transportation activity schedulecommunication and agent activity
411 Modelling Artificial Beijing Figure 10 shows the mainGUI of GME for the modelling of artificial modelling inpublic health events Metamodels are listed in the left area ofFigure 10 the list provides basic syntax elements for domainexperts to model artificial society Domain experts build
models based on the knowledge of their own MeanwhileGME supports hierarchy for building large scale systemsThesyntax symbol listed in GUI can be extended in new tab bydouble clicking Take agent for example the model of agentcan be detailed by edition in another tab page of agent
As shown in the center of Figure 10 the models ofartificial Beijing consist of five parts models of agent andenvironment domain models of public health events inter-ventionmodels controllermodels and services Agentmodeldescribes individuals in society it is composed of basicpopulation information action social relationships activityschedule and disease related information Activity schedulerepresents individualrsquos physical actionmodel focusing on thedaily action of agents Environment model includes physicalentities such as buildings playground transportations andagents contained in environment Domain models of publichealth events are composed of the propagation model of dis-ease disease state transition model and so on Interventionmodels include the settings of vaccination isolation and soon The models mentioned before are all FSM models Themechanisms of these models will be detailed in next sectionsController models and services are the public service mod-ules they are implemented in the development in MDT
412 Modelling Agents and Environments Under the speci-fication of FSM agent and environment models are imple-mented in two parts the state space and state transitionsThe state space is composed of the demographic attributesand behavior related attributes The transitions are triggeredwhen the conditions of states are satisfied As shown inFigure 11(a) the action of agent is changed when the ldquonexttimerdquo condition is satisfied in agent model while the agentslist is changed when the agent arrival condition is satisfied inthe environment model
413 Modelling Activities Agent activities come from theagent state transitions of actions such as movements andcommunications The actions of agents are instructed by theactivity schedule shown in Table 2 Activity schedule lists
Mathematical Problems in Engineering 11
Transportation EnvironmentsEnvironments
Agents
Artificial Beijing
Agent activityActivity schedule
Agent activity
Communication by socialrelationship networks
Communication by socialrelationship networks
Figure 9 The basic elements of artificial society
Syntax elements of DSM
Services
Models of agentenvironment andactivity schedule
Domain models of publichealth events
Intervention models
Modelingaspect ofdomainexperts
Controllermodels
Figure 10 The models of artificial Beijing in GME
all the actions with probability in one day for agents inboth normal and emergent situation [38] There are severaltypes of activity schedule in artificial Beijing student agentactivity schedule worker agent activity schedule emergentagent schedule and so on For example Table 2 gives an agentactivity schedule Upon the instruction of activity schedulestudent agent changes the actions by 119901
119894after state transitions
The119901119894in the tablemeans the action probability in the relevant
period In the duration from 0800 to 1200 a student agenteither goes to classroom to have class or goes to libraryto study The probability of class action is 119901
2while the
probability of study action is 1 minus 1199012
Agent behaviors are decided by the settings of activityschedule In addition to the daily activity schedulementionedbefore emergent activity schedules are also considered inour work Take public health events for instance an infectedagent changes schedule from a normal one to emergentone The workflow of a susceptible agent is illustratedin Figure 12 to show the change of behaviors After theinfection the agent is set in incubation phase Not all theincubation agents will become symptomatic Some of themturn back to being susceptible and some of them becomesymptomatic The symptomatic agents change their activityschedule from normal to emergent In the emergent case
12 Mathematical Problems in Engineering
Next actionNext environment (location)
Advance to next actiontime
Find next environment inactivity chain by action
Current actionCurrent environment (location)
Q
q0
120575 Find next action inactivity schedule by time
F
Σ
(a) The state transitions of agent
Q
q0
120575 F
Current agent list
Current pollution level
Next agents list
Next pollution level
Agents entered
Infected agents entered
Add agents to currentagents list
Compute pollution levelby infected agents
Σ
(b) The state transitions of environment
Figure 11 The state transitions of models
Table 2 The activity schedule of a student agent
Duration (Δ119905) Activity Location Probability 119879Location (minute)0000ndash0600 Sleep Dormitory 119875
0(100) 360
0600ndash0800 Breakfast Dormitoryrestaurant 1198751(068) 120
Sports-breakfast-travel Playground-dormitoryrestaurant 1 minus 1198751(032)
0800ndash1200 Class Classroom 1198752(077) 240
Study Library 1 minus 1198752(023)
1200ndash1400 Lunch Restaurant 1198753(090) 120
Lunch-shopping Restaurant-convenience store 1 minus 1198753(010)
1400ndash1800 Class Classroom 1198754(070) 240
Study Library 1 minus 1198754(030)
1800-1900Dinner Dormitoryrestaurant 119875
5(063)
60Sports-shopping Playgroundconvenience store 1198756(025)
Study Classroomlibrary 1 minus (1198755+ 1198756) (012)
1900ndash2200Rest Dormitory 119875
7(045)
180Dinner-rest Dormitoryrestaurant-dormitory 1198758(020)
Study Classroomlibrary 1 minus (1198757+ 1198758) (035)
2200ndash2400 Rest Dormitory 1198759(072) 120
Sleep Dormitory 1 minus 1198759(018)
infected agents go to hospital according to the treatmentschedule or stay in dormitory according to isolation scheduleAfter the treatment in hospital or self-immunoprocess agentsbecome healthy and immune of disease If the agents aretreated in the hospital they are not allowed to get out untilrecovered Moreover activity schedules are also influencedby the emergency response plans For example in the case ofisolation in emergency response plans the activity scheduleof agents who had contacted with infected is changed Onlythe locations including home and dormitory compose thisemergent schedule Likewise based on the statistical datafrom emergency response plans the additional possibilities
are added into activity schedulesThe behaviors of agents alsochange with activity schedules In the view of ACP approachthe injection of new emergency response plans implementsthe parallel execution of microagent behaviors
In another aspect contact frequency is also anothercrucial element to determine the infected rate in public healthevent It is important to model the contact frequency andcontact time of individuals Based on the studies on thecontact behavior of human being by questionnaire surveyEdmunds found that the contact frequency of individualcould be fitted approximately into a normal distributionand the mean and the standard deviation distributions are
Mathematical Problems in Engineering 13
Susceptible agent ImmuneYes
Healthy and immune
No
Choose schedule
Self-isolation Go to hospital
Staysymptomatic
No
Receive treatment
Contact withinfected agents
Symptomatic
No
Self-immunoprocess
Yes
Recovered
Yes
Incubation
Infection
No
Yes No
NoYes
Figure 12 The workflow of a susceptible agent
168 and 85 [39] So it is possible to apply normal randomvariable to model the contact frequency In our work Box-Muller method [40] is used to generate the random variableof contact frequency shown in the following
119865119862= 120583119865+ 120590119865(minus2 ln (120574
1))12 cos (2120587120574
2) (1)
in which 119865119862is the random variable of contact frequency 120583
119865
is the mean value of normal distribution 120590119865is the standard
deviation of normal distribution and 1205741and 120574
2are the
uniform random variables distributed in the interval [0 1]Based on (1) and survey data [41] the contact frequenciesof agent 119865contact(119860 119894) are discretized as in Table 3 within theconsideration of activity differences
Similarly the duration per contact between individualsis another key factor of AHC transmission The duration ofcontact could also bemodeled by the normal randomvariablein the following [42]
119879119862= 120583119879+ 120590119879(minus2 ln (120574
1))12 cos (2120587120574
2) (2)
in which 119879119862is the random variable of duration per contact
120583119879
is the mean value of normal distribution 120590119879
is thestandard deviation of normal distribution and 120574
1and 120574
2
are the uniform random variables distributed in the interval
[0 1] In addition because the time spent in specific location119879Location(119860 119894) also follows the random distribution like (3)in our work In order to make the duration of contact119879contact(119860 119894) shorter than the 119879Location(119860 119894) the mean andstandard deviation of (2) is set up in (3) and (4) [42]
120583119879=119879Location(120583119865+ 120590119865) (3)
120590119879=119879Location(10120583119865+ 10120590
119865) (4)
in which 119879Location is the time of an activity in a specificlocation set in Table 2 120583
119879and 120590
119879are the mean and standard
deviation of contact frequency As a result the duration ofagent 119879contact(119860 119894) in specific locations (listed in Table 2) isalso discretized in Table 3 within the consideration of activitydifferences
414 Modelling Transportation In artificial Beijing threekinds of travel models are considered walking inside thedistrict travelling by road networks and travelling by subwaynetworks According to the traveling models the agentmovements are implemented by the compositions of travelingmodels before So it is important to build the basic road
14 Mathematical Problems in Engineering
Table 3 The contact frequency and duration in different locations and activities by social relationships in case of student and teacher
Social relationship Activity Location 119879contact(119860 119894) (minute) 119865contact(119860 119894)
Student-student
Sleep Dormitory 0 0
Breakfast Restaurant 119873(52) 119873(32)Breakfast Dormitory 119873(52) 119873(21)Sports Playground 119873(21) 119873(53)Class Classroom 119873(21) 119873(21)Study Library 119873(21) 119873(21)Lunch Restaurant 119873(158) 119873(32)
Shopping Convenience Store 119873(203) 119873(21)Dinner Restaurant 119873(108) 119873(21)Dinner Dormitory 119873(158) 119873(32)Rest Dormitory 119873(63) 119873(52)
Teacher-student Class Classroom 119873(21) 119873(21)
Subway network of Beijing city (red lines) Road network of Beijing city (green lines)
Figure 13 The implementation of transportation for artificial Beijing
and subway networks With the help of Google map thekey points of road are sampled first and the road networksare generated by the links among point Similarly subwaynetworks are generated by the links between subway stationsmapped from the real stations in Google map Figure 13shows the road and subway networks in artificial Beijingwith the road networks in green and the subway networks inred Most of the districts in Beijing city are covered by thesetwo networks The algorithm of hierarchical route planningis proposed as follows
(i) Find the next position where agent will be located innext action
(ii) Obtain the travelingmodels from the statistical trans-portation data
(iii) Search the path from the starting position to thenearest road entry or subway stations
(iv) Search the optimal path from the starting station orroad entry to the nearest station or road exit of targetposition
(v) Search the path from the target subway station or roadexit to the target position
(vi) Attach the subway train number or the bus numberincluding the transfer information to the path accord-ing to the timetable of subway or bus
(vii) Connect all the paths obtained before Generate thepath from the starting position to the target position
Based on the algorithm for each agent in artificial Beijingthe transportation is simulated during the computationalexperiments It is worth noting that the activities insidethe buses or train cars are considered because the agentcommunication during travelling is also necessary to besimulated
Mathematical Problems in Engineering 15
(a) The implementation of agent in MDT (b) The code framework of agent
Figure 14 The agent model implementation in MDT [35]
415 The Implementation of Models by MDT All the modelsare developed in the MDT Figure 14 shows the GUI of thetool MDT is used to implement the state space and statetransitions of FSM models The code framework is shownin Figure 14(b) Using the C++ inheritance features and thetechniques of dynamic link library (dll) the models areencapsulated in the dll componentsThe flexible compositionmechanism is designed to support the models evolution incomputational experiments
416 The Description of Artificial Beijing by ASE As men-tioned in Section 34 ASE describes the artificial Beijingin macroview The statistical information of Beijing city iscollected by ASE Table 4 lists part of the description ofartificial Beijing generated by ASE
42 How to Obtain the Data of Beijing City
421 Population and Geospatial Environment Database ofBeijing City Based on the models developed for ArtificialBeijing it is necessary to generate the initial data to do thecomputational experiments Because the individual level dataare not available it is necessary to construct an individual-based population database for both accurate computationalexperiment and determining optimal decisions
According to the state spaces of models discussed inSection 412 the geospatial and population database of Arti-ficial Beijing is designed as shown in Figure 15 The kernelpart of database consists of the tables such as agent list tablegeospatial environment list table household list table andagent distribution table Agent list table is an individual leveltable used to store the data to initialize the state space of agentmodel such as id and gender age Geospatial environmentlist table is also an individual level table which stores thedata of environment attributes such as id street id andtype Household list table and agent distribution table storethe statistical data for the data generation for agent andenvironment models With the help of the geospatial andpopulation database the data generating of artificial Beijingis proposed in the next section
422 Generating Geospatial and Social Environment Datafor Artificial Beijing by PGET The algorithm of generatinggeospatial and social environment data for artificial Beijingis used to quantify the spatial distribution of population andformalize geospatial behavior of each agent The algorithmshown in Figure 16 is capable of generating a syntheticartificial society which allowsmultiresolution statistical datasocial interactive behavior and multilayer social networksto be integrated together The synthetic population canrepresent individual agents in the form of households andhousehold members and the synthetic population is statis-tically equivalent to a real population For each householdcharacteristics such as address household size family typesand relationships are generated Each person is described bycharacteristics such as age gender social role and correlatedlocations The algorithm provides an effective methodologyto reconstruct the computing environment in high resolutionby using statistical data in low resolution leading to betterprediction and management of emergencies The details ofalgorithm are illustrated in [36]
As shown in Figure 17 the implementation of the algo-rithm is embedded inside the GEPTThe collective input datasuch as the data files of population distribution is collected inthe tool first then the algorithm is activated to generate thedatabase The generation lasts almost twelve hours one timewith the specified parameters settings When the generationis finished the data of artificial Beijing is obtained in thedatabase with 19610000 agents and 16000 environments
5 H1N1 Experiments in ArtificialBeijing with KD-ACP
With the support of the artificial Beijing modeled before thecomputational experiments and parallel execution are alsoimplemented by KD-ACP The experiments are designed tofind the most optimal emergency decision response plansEpidemic in city is a typical scenario in emergency manage-ment H1N1 epidemic in Beijing in 2009 is used to be the usedcase to test KD-ACP According to Section 34 it is necessarytomodel emergencymanagement first in order to support thecomputational experiments
16 Mathematical Problems in Engineering
Table 4 The description of artificial Beijing generated by ASE
Statistical features ofartificial Beijing Descriptions of statistical features Data sourcedefault values
Agent roles Baby child primary student middle school student college student worker retired
Demographic data file ofBeijing
Environment types Super market store restaurant park stadium hospital school cliniccommercial building dormitory building residential building
Demographic data file ofBeijing
Demographic data Age distribution of population sex distribution of population householddistribution of population retired age distribution children distribution
Demographic data file ofBeijing
Environment data Distribution of all types of environments Environment distributiondata file of Beijing
Relationships Classmate relative college friendship family Demographic data file ofBeijing
Varying parameters
Age difference of couples119873(120583120590) 119873(03) isin [0 12]Minimummarriage age (male female) (22 20)Age difference of children119873(120583120590) 119873(13) isin [1 10]Enrollment rate of kindergarten 120572 08Enrollment rate of primary school 120573 1Enrollment rate of middle school 120574 1Enrollment rate of college 120575 08Entrance age of kindergarten 1205721015840 3Entrance age of primary school 1205731015840 7Entrance age of middle school 1205741015840 13Entrance age of middle school 1205751015840 19
Age distributioninformation table
Agent settable
Geospatialenvironment set table
Household settable
Figure 15 The geospatial and population database of Beijing city
51 Modelling Emergency Management The modelling ofemergency management consists of two parts the modellingof emergent events and the modelling of intervention mea-sures H1N1 and the intervention measures for H1N1 aremodeled in our case
511 Modelling H1N1 by MDT H1N1 Model in agent simu-lates the states transitions of health status and the relevantactions Referring to SIR (SusceptibleInfectiveRecovered)[43 44] models agent has three health statuses susceptibleinfected and healthy with immunity Only susceptible agentcan be infected by the contact with infected agents After
the three infected phases (incubation being symptomaticand recovery) of H1N1 model agent is healthy again It isworthy mentioning that the recovered agents can be infectediteratively when the immunity disappears gradually by timeThe state transitions are illustrated in Figure 18
According to [45] the latent period of H1N1 was in aWeibull distribution within people from one to seven days[46 47] The distribution was usually centered in the rangebetween one and three days AWeibull random variable [40]is used to model the latent period as described in
119879Lat = 120572 times [minus ln (1 minus 120574)]1120573+ 120592 (119879Lat ge 120592) (5)
Mathematical Problems in Engineering 17
Censusdata
Generate GUID
Assign individualhousehold
Generate familystructure
Assign family roleto every member
Endow individualwith age gender
social role
Distribute environmentlocations on the map
School Workplace Entertainmentlocations
Consumptionlocations
Assign school orworkplace to
individual
Correlatedindividual withother locations
Generate friendshipin school
Generate coworkerrelations inworkplace
Assign individualadministrative
region
Generate housebuildings
Locate householdto house building
Generate familyrelation
Generateneighborhood
relation
PopulationEnvironment
Environment listSocial network
Figure 16 The algorithm of generating geospatial and social environment data [36]
Edit
Edit
Edit
Edit
Edit
Edit
BrowseStatistical characteristic of population
Geospatial population data
Population data of street
Population types
Geospatial and social environment generation tool
Workspaces
Parameters distribution settings
Household types
Roles in household
Household structure
Social roles
BrowseBrowseBrowseBrowseBrowseBrowse
Population distribution data
File path
Retired distributionPop distributionHousehold numHousehold distributionSex distributionAge distributionData type
Figure 17 The GUI of Population and Geospatial Environment generation Tool (PGET)
18 Mathematical Problems in Engineering
Infected
SusceptibleHealth
withimmunity
Infected
by contac
t
with in
fected
agents
Recovery
Increase before beingsymptomatic symptomatic
Decrease after being
Decrease by time
Infectivity
Immunity
Figure 18 The transitions of agent health status
119879Lat denotes the duration of latent period 120592 120572 and 120573 arethe location parameter scale parameter and shape parameterof Weibull distribution respectively 120574 is a uniform randomnumber in the range [0 1] According to statistics [47] 120592 120572and 120573 are set as 0 18 and 121 respectively Then the mean(standard deviation) of latent period is calculated as 159 days(058 day2) Furthermore infectious period is set as 7 days
A susceptible studentrsquos probability of getting infecteddepended on the infectivity of infectious agents his ownimmunity level the duration of the contact action and soon The infectivity of an infected agent evolved with timeduring the infectious period According to the statistics ofinfectivity in the chart with red bars in Figure 19 an infectedstudent had the highest level of infectivity in the second dayafter he had the first symptom The infectivity levels in otherdays in comparison with the second day are in Table 5 Theday labeled as ldquominus1rdquo means the day before the starting of thesymptomatic period
Based on the algorithm discussed before theH1N1modelis developed with the help of MDT The settings of influenzamodel will be discussed in Section 521
512 Modelling Intervention Measures by MDT In corre-spondence intervention measures model is used to ceasethe outbreak of H1N1 As discussed in [42 48 49] if theappropriate measures are taken when the infectious diseaseemerges the transmission of the disease could be sloweddown and the damage could be decreased Therefore it isnecessary and important to design emergency interventionmeasures in the artificial Beijing According to the MinistryofHealth inChina the intervention policies [50] are designedto control the spread of influenza such as H1N1 The inter-ventionmeasures of these policies including are interventionsactivated time vaccination rate antibiotic delay to hospitalisolation duration close workspace duration and limitationof activities listed in Table 6 Similarly the models of theseintervention measures are also developed with the help ofMDT Using the reasonable data ranges listed in Table 6intervention models can be initialized to do the computa-tional experiments These experiments are used to show how
Symptomatic1 2 3 4 5 6 7minus1
Recovered
Tim
e (da
y)
Infe
ctiv
ity
Figure 19 The distribution of infectivity rate of an infected agent[37]
Table 5 The proportion of infectivity
Days (119894) minus1 1 2 3 4 5 6 6Proportion of infectivity 03 05 1 08 04 02 01 005
to restrain the H1N1 transmissionThe settings of these inter-vention measures in IMCT are introduced in Section 522
52 Experiment Settings
521 H1N1 Settings by EECT EECT is designed to initializethe emergent events models such as H1N1The settings comefrom the specified emergent event models As shown inFigure 20 in theH1N1 used case the settings are composed ofthe distribution of infection source infection rate infectionperiod and so on The infection rate changes with envi-ronments and infection periods The settings are initializedfrom both the research of medicine on H1N1 and the data ofH1N1 outbreak in Beijing in 2009The process of influenza inartificial Beijing is simulated to repeat the possible scenariosof H1N1 break During the computational experiments ofartificial Beijing theH1N1 settings such as the period of beingsymptomatic are updated by the dynamic data registrationsfrom OsdRT The updating corrects the H1N1 model duringthe experiments With the help of OsdRT artificial Beijing isable to approach actual society It is also an implementationof parallel execution
522 Intervention Measures Settings by IMCT In the H1N1case the settings of intervention measures are shown inFigure 21 in IMCT These measures are designed under theinstruction of Emergency Decision Organizations like ChinaDisease Center (CDC) The measures such as vaccinationrate the rate of closedworkspace and the starting time can beinitialized in the tool Emergency response plans are imple-mented in these settings As illustrated in Figure 7 all the pos-sible compositions within value settings in reasonable inter-vals are tested to find the most optimal emergency responseplan For example different compositions of measures suchas vaccination rates and rates of workspace closed are usedto design the plans of experiments By the analysis of theresults the best composition of interventionmeasures can beobtainedThe analysis will be discussed in case study in detail
Mathematical Problems in Engineering 19
Table 6 The intervention measures of influenza
Intervention measures ValuesInterventions activated Time 119879Intervention isin [0 80th day]Vaccination rate 119877Vaccination isin [0 1]Antibiotic rate 119877Antibiotic isin [0 1]Delay to hospital 119867ospital(119860 119894) 119879infected lt 119905 lt 119879recovered + 119879HDelay 119879HDelay isin (0 3th day]Isolation duration 119868solation(119860 119894) 119879Intervention lt 119905 lt 119879Intervention + 119879IDuration 119879IDuration isin (0 14th day]Close workspace duration 119862lose(119864119894) 119879Intervention lt 119905 lt 119879Intervention + 119879CWDuration 119864119894 isinWorkspace 119879CWDuration isin (0 7th day]Limitation of agent activity duration 119871 imitation(119860 119894) 119879Intervention lt 119905 lt 119879Intervention + 119879LAADuration 119879LAADuration isin (0 7th day]119877Vaccination means the vaccination rate of artificial Beijing while 119877Antibiotic means the antibiotic rate used for infected agents in hospital119867ospital (119860119894) means that the infected agent 119860119894 is sent to hospital he or she will receive the retreatment immediately119868solation (119860119894) means that the infected agent 119860119894 is isolated at home he or she can only contact with the other agents until isolation duration is over119862lose (119864119894) means that the environment 119864119894 typed of workspace is closed until the close workspace duration is over119871 imitation (119860119894) means that the activities of agent 119860119894 are limited only the workspace and home are allowed until limitation duration is over
Incubation Weibulldistribution
Symbolic LogNormaldistribution
Recovering LogNormaldistribution
H1N1influenza
Settings of H1N1 influenza model
Disease model
Transmissionmodel
Figure 20 The settings of H1N1 in EECT
It is worth notifying that the settings of interventionmeasures are also influenced by OsdRT For example theH1N1 outbreak areas are notified by OsdRTThe interventionmeasures such as close workspace are used only in theoutbreak areas Therefore the unnecessary computation ofintervention measures is avoided So the data injection byOsdRT not only makes artificial society approach actualsociety but also increases the performance of computationalexperiments
53 Experiments by EMT When the initialization of arti-ficial Beijing is ready EMT is used to do the computa-tional experiments EMT is composed of two parts thecentral controller in the central node and the residen-tial service deployed in the computational nodes Centralcontroller loads the experiment plans and controls the whole
experiments process the GUI of controller is shown inFigure 22 EMT manages the computational experimentsin four steps cluster configuration load experiment plansmodels deployment and running experiments
531 Cluster Configuration Before the configuration clusterinformation is collected from the residential services insidethe cluster The information includes nodes amount nodenames node IPs and CPU occupations It is used to quantifythe computing power of the clusterThe cluster configurationis based on this information The nodes are selected forexperiments Then the LP (local process) [51] number foreach node is set according to the number of the CPU kernelsAfter the configuration the experiment plans are loaded tocustomize the models and data
20 Mathematical Problems in Engineering
Streets of close workspaces
Limitations ofactivities
IsolationsRespirator
Closeworkspace
Antibiotic
VaccinationHospital
Surveillance
Figure 21 The settings of intervention measures in IMCT
Information ofexperiment plan
List of experiment plans
Start | Pause | Reset | Stop | Speedup | Slowdown | Back
NodeName | CPU occupation | Memory occupation |Network | Models | Status
Figure 22 The deployment and running of experiments in EMT
532 Load Experiment Plan As discussed in Section 35 thedescription of experiment plans is composed of the infor-mation in three aspects models and the initial data for theexperiments the deployment mapping tables from the mod-els to the nodes and the plan of experiment executionWhenthe experiment plan is loaded by the EMT the file namesand model descriptions of models and data are listed Withthe support of cluster configuration models can be mappedto the nodes in the GUI according to the computationrequirements EMT also provides the setting of experimentsitself including the running times start time and end time
533 Models and Data Deployment Models are stored inthe agent population and services repository According tothe settings in experiment plan the customized models aredownloaded from the repository first then the models suchas agent model environment model emergent events modeland intervention measures model are integrated to be thecompositionmodels of artificial BeijingThemodels and datalocations in the nodes are configured in the ldquomodel pathrdquoand ldquodata pathrdquo in the GUI of central controller Finallythe models are uploaded to the nodes in cluster with theconsideration of partition The partition of artificial society
Mathematical Problems in Engineering 21
models is different from the multiagent system in [52]Models in artificial society are divided into three levels agentsand environments grids and city Agents and environmentsare encapsulated in grids which compose the whole cityThus the partition of artificial society becomes the partitionof grids The parallelism degrees are found to support thepartition of models agent independence and grid indepen-dence According to the parallelisms agents are partitioned toLPs according to the population distribution in grids Agentmovements among grids are simulated by the movementevents among grids in different LPs As a result a two-tierparallel architecture is proposed to support the simulation ofartificial society The architecture is detailed in [53]
534 Running of Experiments When the preparation isready the experiments can be done by the central controllerof EMT The running and control commands of the experi-ments are grouped into three aspects start pause or stop theexperiments get the running information of the nodes suchas the snapshot of the desktop of node With the help of thesecommands users can run the computational experimentsto do the research on the emergency management It isworth notifying that the simulation engine is optimized inorder to support the ten million agent simulation UsingGPU as coprocessor a two-tier parallel simulation engine isdesigned with support of MPI and OpenCL through phasedsynchronization [54] One-sided communication is used forreflection of remote simulation objects and message passingbetween processes A general kernel function prototype iselaborately designed and conditionally compiled for execu-tion on both CPU and GPU Moreover the optimizationoperations like load balance is developed under the instruc-tion of activity-based simulation [55] The densely populatedgrids are given more CPU and memory resource accordingto the activity predictions
54 Results of H1N1 Experiments with KD-ACP On thebasis of artificial society models and the intervention mea-sures mentioned in Section 512 a series of computationalexperiments are performed to study the H1N1 influenza inBeijing In our case of artificial Beijing 19610000 agentsare simulated in the cluster within 48 CPU cores and 128Gmemories the nodes are connected by kilomega networks Ittakes 18 hours to simulate a 250-day disease spreading
In order to illustrate how the optimal plans are obtainedbyKD-ACP the experiments are divided into four groups theexperiments ofmodel validation the experiments of sensitiv-ity analysis of vaccination rate the experiments of sensitivityanalysis of isolation and the experiments of the combinationof interventionmeasuresThese four groups show the generalprocess of the research on emergency management in H1N1influenza Firstly the models of H1N1 are validated withthe support of historical data Secondly the traditionalmedical interventions of pandemic such as vaccination areanalyzed to find the optimal plan Thirdly the traditionalnonmedical interventions of pandemic such as isolation areanalyzed to find the optimal plan Finally the combination ofinterventions is tested to show the combined effect
It is worth mentioning that all the experiments in ourwork are performed 100 times These experiments are ini-tialized with different settings of interventions in IMCT Theresults shown in the figures are the mean values
541 The Experiments of Model Validation In our artifi-cial Beijing the models of H1N1 influenza are built basedon the existing researches Wang and her colleagues hadstudied the H1N1 influenza by SEIR (SusceptibleExposedInfectiveRecovered) models [56] With the support of theirwork we build the models of artificial Beijing in the mannerof multiagents
In Figure 23 the control group shown in red line isdrawn based on the realistic statistical data collected in H1N1influenza in Beijing in 2009 The influenza lasts more thansix months and more than 170000 people are infectedComparing with the control group the simulated resultswithout interventions are drawn in blue It is obvious that thesimulated case fits the control group well before 70 days Butthe ascending of control group is slowed after 70 days andthe peak value (17883) is much less than the simulated results(25135) It is because the vaccination is activated at day 67 incontrol group according to the intervention measures usedin 2009 The restrain by vaccination becomes effective eightdays later The change of infected number after vaccinationis shown by red spotted line The trend of control group isdifferent from the simulation data after 75 days
However the simulated results (blue line) in compu-tational experiments fit the control group (as shown inFigure 23) in the first 70 days when no intervention mea-sures are executed The key features such as effective basicreproduction number (119877
0) are also almost the same in both
cases Therefore it can be concluded that the models ofartificial society built in microview are validated by therealistic statistical data in influenza in Beijing in 2009
In this group of experiments no interventions are exe-cuted in order to validate the models Therefore the simu-lated results of this extreme case which cannot happen todaycould be obtained As shown by blue line the number ofinfected agents without interventions reaches peak value atday 91 with 25135 consisting of 8052 in symptomatic phaseand 28950 in recovery phase It is obvious that the totalnumber of infected agents grows slowly at the beginningtime Then the number increases quickly when the influenzabreaks and it reaches its maximum value with 383870 in theend This phenomenon can be explained as in the followingAt the beginning of influenza due to the limited numberof infected agents the propagation of the disease remainsslow but many agents become susceptible When the numberof infected agents increases rapidly the spreading becomessignificant However the number of infected agents decreasesafter day 91 It is because the disease spreading relies on thesocial networks and spatial contact networks [49] It meansthat the infected agents can only infect agents either in thesocial relationship networks or in the spatially contacted
542 The Experiments of Sensitivity Analysis of VaccinationRate According to Section 512 vaccination is a typical
22 Mathematical Problems in Engineering
0
Infe
cted
agen
ts
0 50 100 150 200 250
Statistical dataSimulation data
25000
20000
15000
10000
5000
Day d
Figure 23 The results with no intervention executed and the com-parison with realistic statistical data in Beijing in 2009
medical intervention in influenza Traditionally the effective-ness of vaccination can only be evaluated after the influenzaand the key settings such as vaccination rate are determinedby the experience It is always difficult for emergency man-agement organizations to find the most optimal vaccinationrate in advanceTherefore the sensitivity of vaccination rate isanalyzed in this group of experiments The settings are listedin Table 7 average results are shown in Figure 24
Vaccination rates are set linearly (10 30 50 and70) to test the changes of influenza The peak value ofinfected agents decreases from 25135 22392 15787 6907to 638 respectively It shows that the more the agents arevaccinated the smaller the peak value of the infected agentswill be In the case of 70 vaccination the total infected agentis only 13799 It means that the influenza in this case does nothappen due to the high vaccination rate
The change of peak values also shows an interestingphenomenon Vaccination rates increase linearly the peakvalues decrease slower than the ascending of vaccination ratesat beginning but the descending becomes more and morefaster with the linear increase of vaccination rate It meansthat the small scale of vaccination (less than 30) does notmake senseWith the increase of vaccination rate the numberof susceptible agents decreases significantly In sequencethe possibility of infections is lowered greatly Additionallyit could be found that the decrease of peak values comeslater than the increase of the vaccination rates This is quitereasonable according to Mei et alrsquos work in [57]
As a result it is concluded that high vaccination rate iseffective in the control of H1N1 influenza though costly
543 The Experiments of Sensitivity Analysis of IsolationAccording to Section 512 isolation is a typical nonmedicalintervention It is the most used intervention measure in
Table 7 The initial settings of experiments of vaccination
Parameters ValueAgent count 19610000Initial infected count 40Interventions activated time(119879Intervention)
0
Vaccination rate (119877Vaccination) 0 10 30 50 70
0
Infe
cted
agen
ts
Loading intervention of vaccination
No vaccination10 vaccination30 vaccination
50 vaccination70 vaccination
0 50 100 150 200 250
25000
20000
15000
10000
5000
Day d
Figure 24 The results of loading intervention of vaccination
the emergency management in influenza As mentionedbefore a problem also exists that the measure cannot bevalidated before the intervention Therefore the sensitivityof isolation is analyzed in this group of experiments It isworth mentioning that the isolation is usually combinedwith the hospital policy It means the infected individual issent to hospital meanwhile the people who contact withthe infected are all isolated As a result parameters delay tohospital and isolation duration are designed together in thesettings of experiment Delay to hospital stands for the periodfrom infection time to hospital time while Isolation durationrepresents the period when an agent cannot contact others ifhe has contact with an infected agent The settings of theseparameters are listed in Table 8 average results are shown inFigure 25
Firstly the parameter delay to hospital is simulated aloneto test the sensitivity The simulated results are shown insolid line in Figure 25 The peak value decreases from 6708to 2218 when the delay to hospital decreases from 15 daysto 1 day The half day change brings the 67 reductionMoreover the peak value almost approximates to zero inthe case of 05 days The influenza does not happen in thiscase It can be concluded that the earlier the infected agents
Mathematical Problems in Engineering 23
Table 8 The initial settings of experiments of isolation
Parameters ValueAgent count 19610000Initial infected count 40Interventions activated time (119879Intervention) 80Delay to hospital (119879HDelay) isolation duration (119879IDuration) (15 0) (1 0) (05 0) (15 3) (1 3) (05 3)
0 50 100 150 200 2500
1000
2000
3000
4000
5000
6000
7000
Infe
cted
agen
ts
Loading intervention of isolation
Delay to hospital = 15
Delay to hospital = 10
Delay to hospital = 05
Delay to hospital = 15 isolation duration = 3
Delay to hospital = 10 isolation duration = 3
Delay to hospital = 05 isolation duration = 3
Day d
Figure 25 The results of loading intervention of isolation
are sent to hospital the less the agents will be infectedSo the construction of effective mechanism to find infectedindividuals is important in the emergency management ofinfluenza
Secondly the parameter isolation duration is added tosimulate the isolation of interventionObviously the isolationmeasure decreases the infected agents greatly Compared tothe parameter delay to hospital of 15-day case the peak valuedecreases from 6708 to 362when the isolation is added in theinterventionThe reduction of 6346 agents comes from the 3-day isolation of agents who had contacted with the infectedagents However the decrease of parameter delay to hospitalcannot bring the similar reduction in the 3-day isolationcases According to the figure the peak value decreases from362 to 51 from 15-day case to 05-day case only 311 agents aresaved from infection As a result the sensitivity of simulatedresults does not change obviously with delay to hospital inthe case of isolation It can be concluded that isolation is amore effective measure Though it is not easy for emergencymanagement organizations to decrease the delay to hospitaldue to the current monitoring mechanism the influenza canstill be controlled by isolation
544 The Experiments of Combination of Intervention Mea-sures Isolation is not the only nonmedical intervention ininfluenza close workspace is another common measureWhen close workspace is activated the workspaces are allclosed agents can only stay home at the closed time Luckilyisolation and close workspace are independent from oneanother It is possible for emergency management organi-zations to use these two measures together So this groupof experiments is designed to obtain the effectiveness of thecombination of intervention measures The parameters areinitialized with different settings listed in Table 9
Parameters delay to hospital and isolation duration is setin the middle values (1 day 3 days) according to the lastsection According to the experience close duration is set asseven days It is known that close workspace is a really costintervention measure the activated time is a key parameterWith the fixed parameters mentioned before the changeof interventions activated time is analyzed in this group ofcomputational experimentsThe average results are shown inFigure 26
Compared with the simulated results of experimentsdiscussed above Figure 26 strongly illustrates that influenzais ceased when the intervention measures are combinedtogether The peak value of infected agents is 141 in the ldquo80thdays of close-workspace and isolation with 1 day delay tohospitalrdquo case while 25135 in ldquono interventionrdquo case 2218 inldquo1 day delay to hospitalrdquo case 203 in ldquo1 day delay to hospitaland isolationrdquo case The interventions bring a great reductionof infection the peak values decrease from ten thousandslevel to hundred level
On the other hand the peak value even decreases to 108when the execution time of close workspace is 10 days earlierThe date of maximum value of infected agents is advancedfrom the 135th day to the 108th day The whole circle of H1N1influenza is reducedThe results of this group of experimentsshow that intervention measures should be activated as soonas possible and the compositions of interventions are moreeffective in emergency management of H1N1 influenza
In summary based on the general process lots of exper-iments could be done by KD-ACP to study emergency man-agement in H1N1 influenza Many conclusions are obtainedbased on the analysis of the experiments in this sectionAccording to these conclusions themost optimal response inour case study is the case of ldquo70th days of close-workspace andisolation with 1 day delay to hospitalrdquo It is because this casehas the fewest infected agents relatively But the modelling ofH1N1 influenza is not sufficient Economic and social factorsof interventions such as the money cost by vaccination andisolation the influence of the close-workspace in society
24 Mathematical Problems in Engineering
Table 9 The initial settings of experiments of isolation and closeworkspace
Parameters ValueAgent count 19610000Initial infected count 40Interventions activated time (119879Intervention) 70th day 80th dayDelay to hospital (119879HDelay) 1 dayIsolation duration (119879IDuration) 3 daysClose workspace duration (119879CWDuration) 7 days
50 100 150 200 2500
50
100
150
200
Infe
cted
agen
ts
Loading intervention of isolation and close workspace
Delay to hospital = 1 isolation duration = 3Start close time = 70 close duration time = 7Delay to hospital = 1 isolation duration = 3Start close time = 80 close duration time = 7
Day d
Figure 26The results of loading interventions of isolation and closeworkspace
and even opinion transmission of H1N1 in internet are notconsidered in our case study Without the analysis of thesefactors it does not make sense to talk about the most optimalintervention measures However KD-ACP had provided asoftware framework and general process of the research onemergency management Supported by KD-ACP the expertsof emergency management are able to study emergent eventssuch as H1N1 influenza as detailed as possible Based on theexperiments and analysis again and again the most optimalintervention measures will be found ultimately
55 Parallel Execution by ASST and OsdRT As discussed inSection 35 ASST andOsdRT are used to support the parallelexecution of artificial Beijing The statistical data is collectedand stored in artificial society runtime database at runtimeASST downloads the infection data from the database anddisplays it in the map As shown in Figure 27 the infectedagents are distributed in the districts in BeijingThe snapshotgives the geospatial distribution of the H1N1 epidemic Thelegends of the agents in different status are listed in the left
right corner in Figure 27 Five statuses are shown in the figurehealth without vaccination health with vaccination incuba-tion being symptomatic and convalescence ASST providesa very direct feeling for the emergency experts Partly withthe situation from ASST Emergency Decision Organizationmakes the emergency response plans to the response of theemergency scenario With the help of plans the epidemicin actual society can be restrained In the meantime theinfluenza information is reflected in the Internet Accordingto Figure 8 the data acquisition data extraction and datastandardization of H1N1 influenza are processed in OsdRTOsdRT is implemented by Dr Caorsquos team from Institute ofAutomation from Chinese Academy of Sciences [58] Thetool provides not only the reports of H1N1 SARS hand-foot-mouth disease and intestinal diseases are all monitored Theweb page is shown in Figure 28 H1N1 reports from Internetare collected first Based on the data extraction of the reportsH1N1 case of distribution is obtained and shown in theincidencemap As a result H1N1 incidence data is regarded asthe knowledge which is sent to EECT and IMCTThe settingsof emergent events and intervention measures are modifiedin order to make artificial society approach actual society innext turn of computational experiments With the help ofOsdRT the loop of parallel execution process integrates ACPapproach
6 Conclusion
This work provides an integrated software framework forsocial Computing in emergencymanagement From a systemperspective KD-ACP provides a reliable flexible low costand effective platform for the scientists to do the research onemergency response problems The analysis and predictionof these problems with inherent complexity can be solvedby the repetition of possible alternative experiments onthe software framework Therefore KD-ACP which is anattempt to implement ACP approach seemed as an effectivecomputational framework to support the decision making inemergency response
Currently KD-ACP is used to study the H1N1 epidemicin Beijing in 2009 The most optimal compositions of inter-vention measures are testified to support the response tonext influenza However it is still a long way ahead beforethe automation systematization and practicalization of KD-ACP A lot of work will be carried out along several directionsas in the following
(i) The parallel execution process in KD-ACP is actuallynot strongly connected with the emergency responseorganization Traditional experience and theory stillplay a dominant role in the decision of emergencymanagement It is necessary to propose a mechanismof hall for workshop to facilitate the parallel executionloop in actual society
(ii) Section 33 introduces the principle and process ofmodelling artificial society Metamodels and modelsof artificial Beijing are built by GME Model transfor-mation and code generation are used to implementcodes from FSM based models But many details
Mathematical Problems in Engineering 25
Fangshan district Daxing district Fengtai district Xicheng district
Mentougou district Haidian districtShijingshan district Dongcheng district
Incubation
Symptomatic
Convalescence
Health withoutvaccination
Health withvaccinationLegend of agent
health status
Figure 27 The situation of artificial Beijing by ASST
H1N1
Districts of Beijing
H1N1 case ofdistribution
Incidence map
Distribution ofH1N1 reportsfrom internet
Incidence of H1N1
Fangshan districtLiangxiang town
Population 19000Incidence 135789
(1100000)
Figure 28 The implementation of OsdRT by Institute of Automation from Chinese Academy of Sciences
are still not illustrated due to the topic of thispaper The key techniques such as template of codeframework and model transformations from specificdomains to FSM will be focused and introduced indetail in our next paper
(iii) The performance of computational experiments inKD-ACP is still slower than expectation The opti-mization of simulation engine and models structureis necessary to improve the performance So it ispossible to run artificial society with the real worldin parallel
In summary KD-ACP paves a new way for scientists inboth emergencymanagement and computation simulation to
collaborate with each other in solving the emergency man-agement problems by social computing
Research Highlights
(i) KD-ACP is implemented based on ACP approach
(ii) Artificial Beijing is built with the help of KD-ACP
(iii) H1N1 is modeled to simulate the influenza in Beijingin 2009
(iv) Computational experiments testify the effectivenessof intervention measures
26 Mathematical Problems in Engineering
Conflict of Interests
The authors declare that there is no conflict of interestsregarding the publication of this paper
Acknowledgment
This work was supported in part by the National ScienceFoundation of China (nos 91024030 71303252 61403402and 91324013)
References
[1] W Mao and F-Y Wang Advances in Intelligence and SecurityInformatics Academic Press Oxford UK 2012
[2] E Bonabeau ldquoAgent-based modeling methods and techniquesfor simulating human systemsrdquo Proceedings of the NationalAcademy of Sciences of the United States of America vol 99 no3 pp 7280ndash7287 2002
[3] K M Carley D B Fridsma E Casman et al ldquoBioWar scalableagent-basedmodel of bioattacksrdquo IEEETransactions on SystemsMan and Cybernetics A Systems and Humans vol 36 no 2 pp252ndash265 2006
[4] J M Epstein ldquoModelling to contain pandemicsrdquo Nature vol460 no 7256 p 687 2009
[5] Los Alamos National Laboratory Reports about EPISIMS LosAlamos National Laboratory 2003
[6] C H Builder and S C Bankes ldquoArtificial societies a conceptfor basic research on the societal impacts of informationtechnologyrdquo RAND Report 149 RAND Corp 1991
[7] F-Y Wang ldquoComputational theory and method on complexsystemrdquo China Basic Science vol 6 no 5 pp 3ndash10 2004
[8] F-Y Wang ldquoToward a paradigm shift in social computing theACP approachrdquo IEEE Intelligent Systems vol 22 no 5 pp 65ndash67 2007
[9] F-Y Wang and S Tang ldquoArtificial societies for integratedand sustainable development of metropolitan systemsrdquo IEEEIntelligent Systems vol 19 no 4 pp 82ndash87 2004
[10] D Wen Y Yuan and X-R Li ldquoArtificial societies computa-tional experiments and parallel systems an investigation ona computational theory for complex socioeconomic systemsrdquoIEEE Transactions on Services Computing vol 6 no 2 pp 177ndash185 2013
[11] F Wang F Zeng and Y Yuan ldquoAn ACP-based approach forcomplexity analysis of E-commerce systemrdquo Complex Systemsand Complexity Science vol 5 no 3 pp 1ndash8 2008 (Chinese)
[12] F-Y Wang and P K Wong ldquoResearch commentary intelligentsystems and technology for integrative and predictivemedicinean ACP approachrdquoACMTransactions on Intelligent Systems andTechnology vol 4 no 2 pp 1ndash6 2013
[13] J Sifeng X Gang F Dong and H Chunpeng ldquoStudy on theemergency rescue decision support system of petrochemicalplant based on ACP theoryrdquo in Proceedings of the 6th Manage-ment Annual Meeting (MAM rsquo11) 2011 (Chinese)
[14] X Dong D Fan G Xiong F Zhu Z Zhang and Y Yao ldquoPar-allel Bus Rapid Transit (BRT) operation management systembased on ACP approachrdquo in Proceedings of the 6th ManagementAnnual Meeting (MAM rsquo11) Beijing China 2011
[15] B Ning F-Y Wang H-R Dong R-M Li D Wen and L LildquoParallel systems for Urban rail transportation based on ACP
approachrdquo Journal of Transportation Systems Engineering andInformation Technology vol 10 no 6 pp 23ndash28 2010 (Chinese)
[16] F-Y Wang J Zhao and S-X Lun ldquoArtificial power systemsfor the operation and management of complex power gridsrdquoSouthern Power System Technology vol 2 no 3 pp 1ndash11 2008(Chinese)
[17] M J North T R Howe N T Collier and J R Vos ldquoRepastsimphony development environmentrdquo in Proceedings of theAgent 2005 Conference on Generative Social Processes Modelsand Mechanisms 2005
[18] M J North T RHowe N T Collier and J R Vos ldquoRepast Sim-phony runtime systemrdquo in Proceedings of the Agent Conferenceon Generative Social Processes Models and Mechanisms 2005
[19] SWARM Development Group Swarm 211 Reference Guide2000 httpwwwswarmorg
[20] S Tisue and U Wilensky ldquoNetLogo a simple environment formodeling complexityrdquo in International Conference on ComplexSystems pp 16ndash21 2004
[21] S Luke C Cioffi-Revilla L Panait K Sullivan and G BalanldquoMASON a multiagent simulation environmentrdquo Simulationvol 81 no 7 pp 517ndash527 2005
[22] M Ichikawa H Tanuma Y Koyama andH Deguchi ldquoSOARSintroduction as a social microscope for simulations of socialinteractions and gamingrdquo in Organizing and Learning throughGaming and Simulation Proceedings of the 38th Conference ofthe International Simulation and Gaming Association (ISAGArsquo07) pp 149ndash158 2007
[23] D S Burke J M Epstein D A T Cummings et al ldquoIndividual-based computational modeling of smallpox epidemic controlstrategiesrdquo Academic Emergency Medicine vol 13 no 11 pp1142ndash1149 2006
[24] F-Y Wang ldquoParallel control and management for intelligenttransportation systems concepts architectures and applica-tionsrdquo IEEE Transactions on Intelligent Transportation Systemsvol 11 no 3 pp 630ndash638 2010
[25] Z Molnar D Balasubramanian and A Ledeczi ldquoAn introduc-tion to the genericmodeling environmentrdquo in Proceedings of theTOOLS Europe 2007 Workshop on Model-Driven DevelopmentTool Implementers Forum Zurich Switzerland 2007
[26] S Kelly and J-P Tolvanen Domain Framework in Domain-Specific Modeling Enabling Full Code Generation JohnWiley ampSons Hoboken NJ USA 2007
[27] B P Zeigler H Praehofer and T G Kim Theory of Modelingand Simulation Integrating Discrete Event and ContinuousComplex Dynamic Systems Academic Press San Diego CalifUSA 2000
[28] G Booch J Rumbaugh and I JacobsonThe Unified ModelingLanguage User Guide Addison-Wesley Reading Mass USA1999
[29] F Wagner R Schmuki T Wagne and P Wolstenholme Mod-eling Software with Finite State Machines A Practical ApproachCRC amp Taylor amp Francis Boca Raton Fla USA 2006
[30] T Espiner ldquoChina builds worldrsquos fastest supercomputerrdquo ZDNetUK October 2010
[31] B Chen and G Guo ldquoA two-tier parallel architecture for arti-ficial society simulationrdquo in Proceedings of the ACMIEEESCS26th Workshop on Principles of Advanced and DistributedSimulation (PADS rsquo12) pp 184ndash186 July 2012
[32] G Guo B Chen X G Qiu and Z Li ldquoParallel simulation oflarge-scale artificial society on CPUGPU mixed architecturerdquo
Mathematical Problems in Engineering 27
in Proceedings of the ACMIEEESCS 26th Workshop on Prin-ciples of Advanced and Distributed Simulation (PADS rsquo12) pp174ndash177 July 2012
[33] S-X Lun ldquoResearch on the classification of parallel executionmodes of ACP theoryrdquo Acta Automatica Sinica vol 38 no 10pp 1602ndash1608 2012
[34] F-YWang ldquoA framework for social signal processing and anal-ysis from social sensing networks to computational dialecticalanalyticsrdquo Science China vol 43 no 12 pp 1598ndash1611 2013
[35] G GuoOne Model Userrsquos Manual version 12 National Univer-sity of Defense University 2011
[36] Y Ge R Meng Z Cao X Qiu and K Huang ldquoVirtual citymdashanindividual-based digital environment for human mobility andinteractive behaviorrdquo Simulation Transactions of the Society forthe Modeling and Simulation International 2014
[37] M Kretzschmar and R T Mikolajczyk ldquoContact profiles ineight European countries and implications for modelling thespread of airborne infectious diseasesrdquo PLoS ONE vol 4 no6 Article ID e5931 2009
[38] W Duan and X Qiu ldquoFostering artificial societies using sociallearning and social control in parallel emergency managementsystemsrdquo Frontiers in Computer Science vol 6 no 5 pp 604ndash610 2012
[39] W J Edmunds C J OrsquoCallaghan and D J Nokes ldquoWhomixes with whom Amethod to determine the contact patternsof adults that may lead to the spread of airborne infectionsrdquoProceedings of the Royal Society B Biological Sciences vol 264no 1384 pp 949ndash957 1997
[40] J Banks J S Carson B L Neison and D M Nicol Discrete-Event System Simulation Prentice Hall Englewood Cliffs NJUSA 4th edition 2007
[41] H Yongxia ldquoInvestigation and analysis of college studentsrsquo workand restrdquo Value Engineering vol 31 no 3 2012 (Chinese)
[42] W Duan Z Cao Y Wang et al ldquoAn ACP approach to publichealth emergency management using a campus outbreak ofh1n1 influenza as a case studyrdquo IEEE Transactions on SystemsMan and Cybernetics Part A Systems and Humans vol 43 no5 pp 1028ndash1041 2013
[43] HWHethcote ldquoThemathematics of infectious diseasesrdquo SIAMReview vol 42 no 4 pp 599ndash653 2000
[44] G Chowell E Shim F Brauer and et al ldquoModelling thetransmission dynamics of acute haemorrhagic conjunctivitisapplication to the 2003 outbreak in Mexicordquo Statistics inMedicine vol 25 no 11 pp 1840ndash1857 2006
[45] Y Ge L Liu B Chen X Qiu and K Huang ldquoAgent-basedmodeling for InfluenzaH1N1 in an artificial classroomrdquo SystemsEngineering Procedia vol 2 pp 94ndash104 2011
[46] A R Tuite A L Greer MWhelan et al ldquoEstimated epidemio-logic parameters andmorbidity associatedwith pandemicH1N1influenzardquo CMAJ vol 182 no 2 pp 131ndash136 2010
[47] L Brouwers B Cakici M Camitz A Tegnell and MBoman ldquoEconomic consequences to society of pandemic H1N1influenza 2009mdashpreliminary results for Swedenrdquo Euro Surveil-lance vol 14 no 37 pp 1ndash7 2009
[48] Y Ge L Liu X Qiu H Song Y Wang and K Huang ldquoAframework of multilayer social networks for communicationbehavior with agent-based modelingrdquo Simulation Transactionsof the Society for the Modeling and Simulation International vol89 no 7 pp 810ndash828 2013
[49] W Duan X Qiu Z Cao X Zheng and K Cui ldquoHeterogeneousand stochastic agent-based models for analyzing infectious
diseasesrsquo super spreadersrdquo IEEE Intelligent Systems vol 28 no4 pp 18ndash25 2013
[50] The Preparation and Emergency Response Plans to PandemicPublished by Ministry of Health Ministry of Health 2005(Chinese)
[51] R M Fujimoto Parallel and Distributed Simulation SystemsWiley Interscience 2000
[52] F Cicirelli A Giordano and L Nigro ldquoDistributed simulationof situated multi-agent systemsrdquo in Proceedings of the 15thIEEEACM International Symposium on Distributed Simulationand Real Time Applications (DS-RT rsquo11) pp 28ndash35 September2011
[53] B Chen and G Guo ldquoA two-tier parallel architecturefor artificial society simulationrdquo in Proceedings of the 26thACMIEEESCS Workshop on Principles of Advanced and Dis-tributed Simulation (PADS rsquo12) pp 184ndash186 July 2012
[54] G Guo B Chen and X Qiu ldquoParallel simulation of large-scale artificial society with GPU as coprocessorrdquo InternationalJournal of Modeling Simulation and Scientific Computing vol4 no 2 Article ID 1350005 2013
[55] B Chen L-B Zhang X-C Liu and H Vangheluwe ldquoActivity-based simulation using DEVS increasing performance by anactivity model in parallel DEVS simulationrdquo Journal of ZhejiangUniversity Science C vol 15 no 1 pp 13ndash30 2014
[56] X-L Wang P Yang Z-D Cao et al ldquoQuantitative evaluationon the effectiveness of prevention and control measures againstpandemic influenza A (H1N1) in Beijing 2009rdquoChinese Journalof Epidemiology vol 31 no 12 pp 1374ndash1378 2010 (Chinese)
[57] S Mei A D van de Vijver L Xuan Y Zhu and P M ASloot ldquoQuantitatively evaluating interventions in the influenzaA (H1N1) epidemic on China campus grounded on individual-based simulationsrdquo Procedia Computer Science vol 1 no 1 pp1675ndash1682 2010
[58] X Qiu Z Cao and Z Li Mid-Term Reports of China NationalScience Foundation 91024030 2014
Submit your manuscripts athttpwwwhindawicom
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
MathematicsJournal of
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Mathematical Problems in Engineering
Hindawi Publishing Corporationhttpwwwhindawicom
Differential EquationsInternational Journal of
Volume 2014
Applied MathematicsJournal of
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Probability and StatisticsHindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Journal of
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Mathematical PhysicsAdvances in
Complex AnalysisJournal of
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
OptimizationJournal of
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
CombinatoricsHindawi Publishing Corporationhttpwwwhindawicom Volume 2014
International Journal of
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Operations ResearchAdvances in
Journal of
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Function Spaces
Abstract and Applied AnalysisHindawi Publishing Corporationhttpwwwhindawicom Volume 2014
International Journal of Mathematics and Mathematical Sciences
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
The Scientific World JournalHindawi Publishing Corporation httpwwwhindawicom Volume 2014
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Algebra
Discrete Dynamics in Nature and Society
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Decision SciencesAdvances in
Discrete MathematicsJournal of
Hindawi Publishing Corporationhttpwwwhindawicom
Volume 2014 Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Stochastic AnalysisInternational Journal of
Mathematical Problems in Engineering 3
All the SWARM-like agent-based modelling and simu-lation platforms provide a portable lightweight and easilyextensible environment for simulating agents in arbitraryresearch areas However the heterogeneity in specific socialcomputing domain is not considered Furthermore most ofthe platforms cannotwell support large scale agent simulationbecause of the lightweight engine The engine cannot affordthe simulation of super cities like Beijing andNewYorkwhichhave millions of people
22 Agent-Based Platforms for Emergency ManagementBiowar proposed by Carnegie Mellon University simulatesthe impact of background diseases bioterrorism attackswithin a city 62 diseases are modeled in this platform tosimulate the outbreaks on the populationrsquos behavior GASM(Global-Scale Agent Model) is designed to study the spread-ing of H1N1 65 billion population is modeled with thesupport of official statistical data A global H1N1 spread fromTokyo is simulated in GASM EpiSimS proposed by LosAlamos lab simulates the spread of disease in regions suchas cities allowing for the assessment of disease preventionintervention and response strategies The daily movementsand interactions of synthetic individuals are representedexplicitly Burke and Epstein propose a computational modelof smallpox epidemic transmission and control [23] Theagents in thismodel interact locally with one another in socialunits such as homes workplaces schools and hospitals
However these platforms cannot provide a generic soft-ware framework to study emergency problems Biowar onlyfocuses on social networks individuals are all modeled as thenodes of social networks Agents in GASM and EpisimS areisomorphic without considering the heterogeneity in specificdomains
To sum up this section briefly reviews the existing mul-tiagent simulation platforms including SWARM-like agent-based modelling and simulation platforms and agent-basedplatforms for emergency management However they can-not satisfy the requirements of simulation performanceadaptability of software framework and heterogeneity ofindividuals in research of different emergency scenarios
3 KD-ACP
KD-ACP is an integrated software framework designed andimplemented based on the principle of the ACP approachshown in Figure 1
31The ACPApproach ACP approach is a social computing-based research paradigm It is composed of three componentsas its name artificial society for A computational experi-ments for C and parallel execution for P The basic idea ofACP approach is listed as follows
(i) Model the complex societies involving human behav-ior and social organizations as artificial societies usingmultiagent modelling techniques in a ldquobottom-uprdquofashion Artificial societies are regarded as a researchplatform to study emergency management
Actual societies Artificial societies
Managementand control
Experimentationand evaluation
Training andlearning
TestingTesting Operation
Figure 1 The parallel execution of ACP approach [8 33]
(ii) Utilize innovative computing technologies to evaluateand analyze various factors in emergency manage-ment quantitatively the computers are regarded asthe experimental social laboratories for investigatingemergency management problems
(iii) Provide an effective mechanism for the control andmanagement of complex actual social society throughcomparison evaluation and interaction with artifi-cial society
It is worth notifying that ldquoPrdquo here is not the ldquoparallelrdquo inldquoparallel simulationrdquo but the representation of ldquoparallel exe-cutionrdquo The idea of parallel execution is to build the parallelscenarios by paralleling the actual societies and artificialsocieties Consequently parallel control and management ofactual societies are implemented with the help of interactionsbetween parallel scenarios The goal of parallel executionis to find the best plans to adjust the methods of controland management based on the comparison and analysis ofdifferences between actual and artificial societies Artificialsocieties provide possible simulated results of evolutions byrepeated computational experiments The simulated resultsprovide evidences for the adjustment plans These plans areused in the control and management of actual societies suchas emergency management After the application of theseplans the observations from actual societies are collected forthe comparison with expectation The differences are used tofeedback to artificial societies The new turn of comparisonand analysis to find best adjustments of control and manage-ment is repeated
The mechanism of ldquoparallel executionrdquo has been provedto be effective for use in networked complex traffic systemsand is closely related to emerging technologies in cloud com-puting social computing and cyber-physical-social systems[24] In order to promote the development of parallel controland management in emergency management the artificialsociety is proposed in ACP approach which is the expansionof ldquoartificial traffic systemsrdquo
Instructed by the ACP approach KD-ACP is also com-posed by three components The details of the architectureand implementation of KD-ACP are discussed below
32 The Software Architecture of KD-ACP The architectureof KD-ACP is shown in Figure 2 the software framework iscomposed of a series of tools These tools are grouped intothree parts to support artificial society modelling computa-tional experiments and parallel execution
4 Mathematical Problems in Engineering
A part
C part
P part
Population and GeospatialEnvironment Generation Tool (PGET)
Generic ModelingEnvironment (GME)
Artificial society population andgeospatial environment database
Initialize
InitializeInitialize
Model DevelopmentTool (MDT)
FSM based models
Artificial SocietyEditor (ASE)
Scenario ofartificial society
Artificial societyruntime database
Agent and emergencymodel repository
Experiments DesignTool (EDT)
Experiment plans
InitializeSettings of
emergent eventsEmergent Events
Configuration Tool (EECT)Intervention Measures
Configuration Tool (IMCT) Settings of intervention measuresemergency response plans
Experiments ManagementTool (EMT)
Population data emergent eventsdata and intervention measures
Open source dataRegistration Tool (OsdRT)
Actu
al so
ciet
y Artificial SocietySituation Tool (ASST)
Optimal emergencyresponse plan
Internet
C++ code of modelsdll of models
Figure 2 The software architecture of KD-ACP
In the ldquoArdquo part Generic Modeling Environment (GME)[25] and Model Development Tool (MDT) are the kerneltools in the modelling of artificial society GME is anopen source modelling tool which supports domain-specificmodelling The domains of artificial society are created byGME in our work Models such as agent environmentemergent event and intervention measure are described inspecific domains first in GME With the help of modeltransformation thesemodels are all transformed to the FiniteState Machine (FSM) models Meanwhile code generationsare supported byMDT and thesemodels are all implementedin C++ Artificial Society Editor (ASE) is used to describethe concrete scenario of actual society which defines thescope of models set for artificial society Population andGeospatial Environment generation Tool (PGET) generatesthe population and geospatial environment data with thesupport of statistical data from actual society
In the ldquoCrdquo part Emergency Events Configuration Tool(EECT) initializes the models of emergent events whileInterventionMeasures Configuration Tool (IMCT) initializesthe models of intervention measures Experiments plans aregenerated byExperimentsDesignTool (EDT) Based on theseplans Experiments Management Tool (EMT) is used to runand manage the computational experiments to study theemergency problems
In the ldquoPrdquo part Artificial Society Situation Tool (ASST)seemed as the monitor of running artificial society Thestatistical data and situation are shown by ASST at runtimeIn the meantime the emergency response plans are made by
emergency decision organizations Parts of the influences ofemergency plans are reflected on Internet Open source dataRegistration Tool (OsdRT) is used to register the open sourcedata from Internet to artificial society
KD-ACP is developed using the BrowserServer archi-tecture the tools are integrated in the home page of KD-ACP as shown in Figure 3 Each tool is activated by the clickon the link For example Artificial Society Editor is startedwhen the link of ASE is clicked The working environmentand programming languages of tools in KD-ACP are listed inTable 1
Moreover the implementation of KD-ACP is mainlycomposed of modelling phase and computational experi-ments phase It will be discussed in the next section
33 The Modelling of Artificial Society in KD-ACP It is acritical problem to focus on the key parts of society in socialcomputing Based on the ACP approach the bottom-upmodelling is used to build the artificial society As a resultmodelling of artificial society is composed of three basic ele-ments agents environments and rules for interactionsHow-ever we still meet the problem that specific features shouldbe supported in artificial society For example emergentevents and intervention measures are required in artificialsociety for emergency management The modelling of onlybasic elements cannot cover the specific features in domainsTherefore domain-specific modelling [26] is introduced tosolve the problems in modelling artificial society
Mathematical Problems in Engineering 5
Model DevelopmentTool (MDT)
Artificial SocietyEditor (ASE)
Population and GeospatialEnvironment Generation Tool (PGET)
Emergent EventsConfiguration Tool (EECT)
Intervention MeasuresConfiguration Tool (IMCT)
Experiments DesignTool (EDT)
Experiments ManagementTool (EMT)
Artificial SocietySituation Tool (ASST)
Open Source DataRegistration Tool (OsdRT)
Generic ModelingEnvironment (GME) Artificial
society (A)Computationalexperiments (c)
MDT EECT IMCT
ASST OsdRT
ASE PGET EDT EMT
Parallel execution (P)
Figure 3 The implementation of KD-ACP
Table 1 The working environment and programming languages of tools
Tools Working environment Programminglanguages
Developmentplatform User type
Generic ModelingEnvironment (GME)
General computer (desktopapplication) NULL NULL Domain experts
Model Development Tool(MDT)
General computer (desktopapplication) C++ Visual Studio Model developers
Artificial Society Editor(ASE)
General computer (desktopapplication) C Visual Studio Domain experts
Population and GeospatialEnvironment generationTool
General computer (desktopapplication) C Visual Studio Domain experts
Emergency EventsConfiguration Tool (EECT)
General computer (withInternet Explorer clientside)
C ASPNET Domain experts
Intervention MeasuresConfiguration Tool (IMCT)
General computer (withInternet Explorer clientside)
C ASPNET Domain experts
Experiments ManagementTool (EMT)
General computer (withInternet Explorer clientside)
C ASPNET
Users of computationexperimentServer of EMT
Nodes in supercomputer(Console Program serverside)
C++ Visual Studio
Runtime Infrastructure ofEMT
Nodes in supercomputer(Console Program serverside)
C++ Visual Studio
Artificial Society SituationTool (ASST)
General computer (desktopapplication) C++ Visual Studio Domain experts
Open Source DataRegistration Tool (OsdRT)
General computer (withInternet Explorer clientside)
Java JSP Domain experts
6 Mathematical Problems in Engineering
M2T transformation
Modeling in domain
GME Generic Modeling Environment
Metamodeling
Instance of
FSM
Metamodeling for artificial society
DEVS
Petri Net State charts
Agentmetamodel
Environmentmetamodel
Emergency eventmetamodel
Interventionmetamodel
DAE
Domain-specific metamodels Semantic well definedmetamodels
M2M transformation
Modeling of artificial society
Interventionmodel
Emergency eventmodel
Environmentmodel
Agentmodel
FSM based emergent andintervention models
FSM based environmentmodels
FSM based agent modelsDomain-specific models
Semantic well defined models
M2M transformation
Agent model code Environment model code
Agent behavior model
Agent mentality model
Disease model code
Basic population model Disease spreading model
Disease state transformmodel
Intervention model code
Agent social networkmodel
Basic environmentinformation model
Roadnet model
Vaccination
Infectors isolation
Contactors isolation
Close work officesAgent contact model
Domain-specific models relevant codes
MDT Model Development Tool
Implementation
Figure 4 From metamodelling and modelling by GME to implementation by MDT of artificial society
331 The Principle of the Modelling of Artificial SocietyAccording to the principle of domain-specific modellingthe modelling of artificial society contains the followingsteps first metamodelling the basic elements of artificialsociety second modelling the specific features in domain ofemergency management third implementing the models ofartificial society in codes The whole process is illustrated inFigure 4The first and second steps are implemented in GMEwhile the third step is implemented in MDT
The first step is metamodelling which mainly focuses onconstructing the metamodels of artificial society Metamod-elling tries to study the common patterns of artificial societyThe outputs of metamodelling are metamodels which repre-sents the abstraction of the whole systemThe basic elementsof artificial society are described in metamodel The processof metamodelling is divided into four phases The first is theconstruction of the domain-specific metamodels As shown
in Figure 4 agent metamodel environment metamodelemergent event metamodel and intervention metamodelcompose the metamodel of artificial society The second isthe construction of the metamodels described by typicalmodelling formalisms such as FSM DAE DEVS and PetriNet [27] These formalisms are all semantically well definedThe third is the definition of the model transformationfrom domain-specific metamodels to metamodels of typicalmodelling formalisms The transformation standardizes themetamodels of artificial society by typical modelling spec-ifications The fourth is the definition of the transforma-tion templates from metamodels to code framework Thetemplates list the basic abstract interfaces of metamodels ofartificial societyThese abstract interfaces are implemented inthe specific-domain modelling and code generations
The second step is modelling the models of artificialsociety such as agent model environment model emergent
Mathematical Problems in Engineering 7
Figure 5 The metamodels of artificial society in GME
event model and intervention model are built Actuallythe models are the instantiation of metamodels in the laststep Different from the general modelling environment likeUML [28] the domain-specific modelling provides a familiarmodelling environment for the domain experts in artificialsociety For example emergency response experts only con-cern emergent eventmodel and interventionmodel inheritedfrom metamodels After constructing the domain-specificmodels based on domain-specific metamodels domain usersexecute the model transformation defined in the first stepAll the models of artificial society are transformed into FSMmodels As a result the models are implemented in this uni-fied modelling formalism (FSM) The model transformationmakes the simulation of the models possible
The third step is the generation of executable codes ofmodels The executable code framework is generated bymapping template from metamodels to code frameworkdefined in the first step Moreover domain developers alsoadd necessary codes to the framework to integrate thedynamic semantics of the models The code frameworkoutputs the dynamic link libraries by compilingThe dynamiclink libraries are loaded in the large scale artificial societyruntime infrastructure in computational experiments
332 The Metamodelling and Modelling of Artificial Societyby GME GME is used to build metamodels and modelsin our work As mentioned before the abstraction andcommon patterns of society are represented in metamodelsAccording to the bottom-up modelling style metamodels ofagent environment and communications are described inGME Figure 5 shows part of metamodels of artificial societyThe features of an agent metamodel are extracted from thecensus figures and statistical data Environment metamodelsimulates the geospatial places for the behaviors of agentsThemetamodel of communications among agents ismodeledto simulate the interactions such as infection in epidemicsand rumor propagation in public opinion formation eventsIt is worth notifying that the metamodel of communications
includes both the emergent event metamodel and interven-tion metamodel
From the perspective of modelling the details from spe-cific domains are considered in the models by the instantia-tion frommetamodels of artificial society For example socialrelationships based on complex networks are added in agentmodel to support the communications Agent activity is alsoused to quantify the agent activity under different scenarioEnvironment models are linked with the help of transporta-tion services subways and roads are modeled while the pathsearch is encapsulated in the services Emergent event modeland intervention model are also the domain-specific modelsThe modelling of artificial Beijing in GME will be discussedin detail in next section
333 The Implementation of Models of Artificial Society byMDT As mentioned before MDT is used to implementmodels such as agent environment emergent event andintervention According to the template of code frameworkthe implementations of models are generated by MDT Theimplementations are classified into two categories FSMmodels and services FSM models such as agents andenvironments are built under the specification of FiniteState Machine (FSM) [29] in MDT while all the servicessuch as transportation are encapsulated under the PublicService Standard This standard provides a generic interfacespecification for modelers to encapsulate public commonservices in artificial society FSM models like agent are builtstatistically from the quantitatively analyzed characteristicssuch as demographic attributes social behaviors emergencybehaviors and social networks Social behaviors describe thedaily behaviors of individuals while emergency behaviorsdescribe the individual behaviors in emergent events Forexample infected individuals are all isolated in hospital inSARS Isolation ismodeled as a typical emergency behavior inour work Correspondingly services are used to simulate themacroactual society Take transportation service for instancethe path search is needed by almost every agent duringmoving from spot to spot
8 Mathematical Problems in Engineering
Artificial society description in emergency
Description ofemergency
ArtificialSociety Editor
(ASE)
Agent behavior set
Agent role setAgent status set
Environment setAgent relationship set
Statistical data of real society
Descriptionof artificial
societyframework
Terminal of agent modeling andartificial society describing
middot middot middotAgent
metamodeling
Customized agentmodeling
Relationship modelDailyEmergency
Mod
el D
evelo
pmen
t Too
l (M
DT)
Agent modeling
Agent model Encapsulate
Encapsulate
Behavior model
Intervention
Emergency
Response
Daily
Social relationshipmodeling
Socialbehaviormodeling
Socialbehaviormodel
Emergencybehavior
model
Emergencybehaviormodeling
Building
Geospatial statistical data
Population statistical data Datacollection
Actualsociety
Artificial societypopulation and
geospatial database
Organizationservice
Transportationservice+
MappingAgent model andservice repository+ emergency model
repository
Climateservice
Geographicalservice
+
Agent status in emergencyEmergency management
Emergent event set
Social relationship statistical data
Environment statistical data
Population behaviorstatistical data
Emergency statistical data
Emergency organization statistical data
Geospatial and SocialEnvironment Generation
Tool (GSeT)
Agent model repository andpopulation database
C++C++C++C++
Modeling
Statistical modelJob ageSex location
Public service modeling
Organizationmodel
Transportationmodel
Geographicalmodel
Climatemodel
Interventionservice
Publ
ic se
rvic
e sta
ndar
d
Figure 6 The editing and initialization of KD-ACP
MDT provides domain experts with a tool to obtain thecode implementations of models With the help of compileenvironment like visual studio MDT also supports thefurther programming development of the specific domaindetails which cannot be described in modelling step
Both the FSM models and services are developed by theMDT first and then stored in the agent model and servicerepository The repository manages the models according tothe requirements from emergency problems and provides themodels for EDT to make the experiment plans
34 The Editing and Initialization of Artificial Society byASE and PGET As shown in Figure 6 ASE is used to editthe scenarios of artificial society within emergent eventsThe editing is composed of two parts (1) the statisticalinformation of artificial society in daily life such as the rolesof agents the relationships of agents and the types of environ-ments and (2) the statistical information of artificial societyin emergency including the statistical data of emergent
events the emergency organization and emergency relatedbehaviors of agents
According to the requirements of the editing thesestatistical data are collected from actual society manually bythe domain experts Based on these statistical data PGETgenerates the artificial society population and geospatialenvironment database The database supports the instanti-ation of artificial society at individual level For examplethe attributes such as age and gender of each agent can befound in the database With the support of the databaseFSM models and service repository discussed before itis sufficient for domain experts to study the emergencyproblems by computational experiments
35The Computational Experiments and Parallel Execution inKD-ACP The tools of ldquoCrdquo part and ldquoPrdquo part in KD-ACP areused to support the process of computational experimentsand parallel execution The working process is shown inFigure 7 EECT and IMCT are both the starting and ending
Mathematical Problems in Engineering 9
Agent model andservice repository
repository
Emergency EventsConfigurationTool (EECT)
Open Source DataRegistration Tool (OsdRT)
Intervention MeasuresConfigurationTool (IMCT)
Load
Output Output
Internet
Emergencyresponse plans
Load
∙ Input models of emergencyevents∙ Output settings of agentbehavior in emergent events+ settings of emergent events
∙ Input models of interventionmeasures∙ Output settings of agentbehavior in interventionmeasures + settings ofintervention measures
+ emergency model
∙ Input data from internetnetworks∙ Output population data +emergency events data +intervention measures data
Artificial societypopulation and
geospatialenvironment database
Artificial societyruntime database
OutputThe most optimal
emergencyresponse plan
Actual society
OutputOutputOutput
Input
Load
Load
Load
Load
Output
Artificial SocietySituation Tool (ASST)
Emergency decisionorganization
ClusterRuntime infrastructure
Deploy
Deploy
Experimentsdesign tool (EDT)
Artificial societyinitial data file
Experimentplan
Artificial societymodels
Experiments Management
large scale artificial societyruntime infrastructure
Tool(EMT) +super computerTIANHE-1A
Load
∙ Input experiment plan +artificial society models +artificial society initial data file∙ Output artificial societyruntime state data
∙ Input description of artificialsociety + artificial society populationand geospatial environment society+ emergent events configuration +
intervention measures configuration∙ Output experiment plan + artificialsociety models + artificial societyinitial data file
∙ Input runtime statistical data∙ Output graphics charts ofartificial society statistical data +situation of artificial society
Figure 7 The computational experiments and parallel execution of KD-ACP
point The emergent events and intervention measures areconfigured by EECT and IMCT respectively The configu-rations of emergent events are used to simulate both thereal emergencies like SARS and H1N1 and the supposedemergencies for experiments Similarly the configurationsof intervention measures are also used to reproduce thereal one and simulate the supposed one The repeat of theemergency is used to verify the models while the supposedconfigurations are used to obtain the optimized decision planto the response of the possible emergencies
With the input of artificial society model and servicerepository artificial society population and geospatial envi-ronment database and the configurations discussed beforeEDT generates the experiment plans to meet the require-ments of research on emergency management The outputof EDT includes artificial society models artificial societyinitial data files and experiment plan The models aredownloaded from the repository while the data file is thecollection of data from the database to initialize the modelsWhen the models and data files are ready EMT loads theexperiment plan and deploys the models and data to thecluster or TIANHE-1A supercomputer [30] which was theworldrsquos fastest supercomputer built by National University of
Defense Technology (NUDT) in China in 2010 According tothe plan the experiments are done repeatedly on the largescale artificial society runtime infrastructure [31 32] by themultisample settingsThework process is the implementationof computational experiments in ACP approach
Traditionally emergency response plans are made byemergency management theories and experiences The onlyway to test the effective of plans is the feedback results of realworld ACP approach provides a novel method to supportemergency response plan making by parallel execution Asshown in Figure 7 the work process of KD-ACP is composedof two loops The inner loop composed by red arrowsdescribes the process of computational experiments while theouter loop of yellow arrows illustrates the process of parallelexecution During the runtime of computational experi-ments the statistical data of artificial society is collected andstored in the artificial society runtime database Based on thedatabase ASST outputs the customized situation of runningartificial society by graphics charts and situation mapsThe information is sent to the organizations of emergencydecision to support making the emergency response plansWith the help of computational experiments loop these plansare simulated repeatedly to find the most optimal one
10 Mathematical Problems in Engineering
Open Source Data Registration Tool (OsdRT)
Emergent EventsConfiguration Tool (EECT)
Intervention MeasuresConfiguration Tool (IMCT)
∙ Web information collecting
∙ Deep mining in social media
∙ Web information denoising
∙ Information filtering
Data acquisition
Actual society
Open source informationin social media
Internet
∙ Basic element extracting∙ Individuals and organizationsextraction∙ Sentiment analyzing andopinion mining∙ Social network analyzing
Data extraction
∙ Domain knowledge description∙ Information fusion and collisiondetection∙ Domain knowledge construction∙ Social media informationstandardization
Data standardization
Knowledge
Figure 8 The compositions of OsdRT [34]
Moreover the most optimal plan is used to the responseof emergency in actual society According to the idea ofparallel control in [24] the feedback of actual society ispartly collected from Internet networks by OsdRT As shownin Figure 8 OsdRT is composed of three components dataacquisition data extraction and data standardization Dataacquisition collects mines and filters information fromsocial sensing networks Data extraction includes basicelement extracting individual and organization extractingsentiment analyzing and social networks analyzing Datastandardization specifies the useful knowledge and sent it toconfiguration tools in KD-ACP
By processing in OsdRT the knowledge about emergentevents and intervention measures are analyzed first andregistered in the EECT and IMCT The registration updatesthe settings of emergent events and intervention measuresThis loop composed of yellow arrows implements the parallelexecution in ACP approach The implementation of parallelcontrol and management provides a data-driven approachthat considers both the engineering and social complexityfor modelling analysis and decision making in emergencymanagement
4 Modelling Beijing City with KD-ACP
41 How to Build the Artificial Beijing According to themodelling of artificial society in KD-ACP discussed beforethe Beijing city is modeled as follows
To meet the requirements of emergency managementsthe basic elements of artificial society are extended Asshown in Figure 9 six elements are required to simulate thecity agents environments transportation activity schedulecommunication and agent activity
411 Modelling Artificial Beijing Figure 10 shows the mainGUI of GME for the modelling of artificial modelling inpublic health events Metamodels are listed in the left area ofFigure 10 the list provides basic syntax elements for domainexperts to model artificial society Domain experts build
models based on the knowledge of their own MeanwhileGME supports hierarchy for building large scale systemsThesyntax symbol listed in GUI can be extended in new tab bydouble clicking Take agent for example the model of agentcan be detailed by edition in another tab page of agent
As shown in the center of Figure 10 the models ofartificial Beijing consist of five parts models of agent andenvironment domain models of public health events inter-ventionmodels controllermodels and services Agentmodeldescribes individuals in society it is composed of basicpopulation information action social relationships activityschedule and disease related information Activity schedulerepresents individualrsquos physical actionmodel focusing on thedaily action of agents Environment model includes physicalentities such as buildings playground transportations andagents contained in environment Domain models of publichealth events are composed of the propagation model of dis-ease disease state transition model and so on Interventionmodels include the settings of vaccination isolation and soon The models mentioned before are all FSM models Themechanisms of these models will be detailed in next sectionsController models and services are the public service mod-ules they are implemented in the development in MDT
412 Modelling Agents and Environments Under the speci-fication of FSM agent and environment models are imple-mented in two parts the state space and state transitionsThe state space is composed of the demographic attributesand behavior related attributes The transitions are triggeredwhen the conditions of states are satisfied As shown inFigure 11(a) the action of agent is changed when the ldquonexttimerdquo condition is satisfied in agent model while the agentslist is changed when the agent arrival condition is satisfied inthe environment model
413 Modelling Activities Agent activities come from theagent state transitions of actions such as movements andcommunications The actions of agents are instructed by theactivity schedule shown in Table 2 Activity schedule lists
Mathematical Problems in Engineering 11
Transportation EnvironmentsEnvironments
Agents
Artificial Beijing
Agent activityActivity schedule
Agent activity
Communication by socialrelationship networks
Communication by socialrelationship networks
Figure 9 The basic elements of artificial society
Syntax elements of DSM
Services
Models of agentenvironment andactivity schedule
Domain models of publichealth events
Intervention models
Modelingaspect ofdomainexperts
Controllermodels
Figure 10 The models of artificial Beijing in GME
all the actions with probability in one day for agents inboth normal and emergent situation [38] There are severaltypes of activity schedule in artificial Beijing student agentactivity schedule worker agent activity schedule emergentagent schedule and so on For example Table 2 gives an agentactivity schedule Upon the instruction of activity schedulestudent agent changes the actions by 119901
119894after state transitions
The119901119894in the tablemeans the action probability in the relevant
period In the duration from 0800 to 1200 a student agenteither goes to classroom to have class or goes to libraryto study The probability of class action is 119901
2while the
probability of study action is 1 minus 1199012
Agent behaviors are decided by the settings of activityschedule In addition to the daily activity schedulementionedbefore emergent activity schedules are also considered inour work Take public health events for instance an infectedagent changes schedule from a normal one to emergentone The workflow of a susceptible agent is illustratedin Figure 12 to show the change of behaviors After theinfection the agent is set in incubation phase Not all theincubation agents will become symptomatic Some of themturn back to being susceptible and some of them becomesymptomatic The symptomatic agents change their activityschedule from normal to emergent In the emergent case
12 Mathematical Problems in Engineering
Next actionNext environment (location)
Advance to next actiontime
Find next environment inactivity chain by action
Current actionCurrent environment (location)
Q
q0
120575 Find next action inactivity schedule by time
F
Σ
(a) The state transitions of agent
Q
q0
120575 F
Current agent list
Current pollution level
Next agents list
Next pollution level
Agents entered
Infected agents entered
Add agents to currentagents list
Compute pollution levelby infected agents
Σ
(b) The state transitions of environment
Figure 11 The state transitions of models
Table 2 The activity schedule of a student agent
Duration (Δ119905) Activity Location Probability 119879Location (minute)0000ndash0600 Sleep Dormitory 119875
0(100) 360
0600ndash0800 Breakfast Dormitoryrestaurant 1198751(068) 120
Sports-breakfast-travel Playground-dormitoryrestaurant 1 minus 1198751(032)
0800ndash1200 Class Classroom 1198752(077) 240
Study Library 1 minus 1198752(023)
1200ndash1400 Lunch Restaurant 1198753(090) 120
Lunch-shopping Restaurant-convenience store 1 minus 1198753(010)
1400ndash1800 Class Classroom 1198754(070) 240
Study Library 1 minus 1198754(030)
1800-1900Dinner Dormitoryrestaurant 119875
5(063)
60Sports-shopping Playgroundconvenience store 1198756(025)
Study Classroomlibrary 1 minus (1198755+ 1198756) (012)
1900ndash2200Rest Dormitory 119875
7(045)
180Dinner-rest Dormitoryrestaurant-dormitory 1198758(020)
Study Classroomlibrary 1 minus (1198757+ 1198758) (035)
2200ndash2400 Rest Dormitory 1198759(072) 120
Sleep Dormitory 1 minus 1198759(018)
infected agents go to hospital according to the treatmentschedule or stay in dormitory according to isolation scheduleAfter the treatment in hospital or self-immunoprocess agentsbecome healthy and immune of disease If the agents aretreated in the hospital they are not allowed to get out untilrecovered Moreover activity schedules are also influencedby the emergency response plans For example in the case ofisolation in emergency response plans the activity scheduleof agents who had contacted with infected is changed Onlythe locations including home and dormitory compose thisemergent schedule Likewise based on the statistical datafrom emergency response plans the additional possibilities
are added into activity schedulesThe behaviors of agents alsochange with activity schedules In the view of ACP approachthe injection of new emergency response plans implementsthe parallel execution of microagent behaviors
In another aspect contact frequency is also anothercrucial element to determine the infected rate in public healthevent It is important to model the contact frequency andcontact time of individuals Based on the studies on thecontact behavior of human being by questionnaire surveyEdmunds found that the contact frequency of individualcould be fitted approximately into a normal distributionand the mean and the standard deviation distributions are
Mathematical Problems in Engineering 13
Susceptible agent ImmuneYes
Healthy and immune
No
Choose schedule
Self-isolation Go to hospital
Staysymptomatic
No
Receive treatment
Contact withinfected agents
Symptomatic
No
Self-immunoprocess
Yes
Recovered
Yes
Incubation
Infection
No
Yes No
NoYes
Figure 12 The workflow of a susceptible agent
168 and 85 [39] So it is possible to apply normal randomvariable to model the contact frequency In our work Box-Muller method [40] is used to generate the random variableof contact frequency shown in the following
119865119862= 120583119865+ 120590119865(minus2 ln (120574
1))12 cos (2120587120574
2) (1)
in which 119865119862is the random variable of contact frequency 120583
119865
is the mean value of normal distribution 120590119865is the standard
deviation of normal distribution and 1205741and 120574
2are the
uniform random variables distributed in the interval [0 1]Based on (1) and survey data [41] the contact frequenciesof agent 119865contact(119860 119894) are discretized as in Table 3 within theconsideration of activity differences
Similarly the duration per contact between individualsis another key factor of AHC transmission The duration ofcontact could also bemodeled by the normal randomvariablein the following [42]
119879119862= 120583119879+ 120590119879(minus2 ln (120574
1))12 cos (2120587120574
2) (2)
in which 119879119862is the random variable of duration per contact
120583119879
is the mean value of normal distribution 120590119879
is thestandard deviation of normal distribution and 120574
1and 120574
2
are the uniform random variables distributed in the interval
[0 1] In addition because the time spent in specific location119879Location(119860 119894) also follows the random distribution like (3)in our work In order to make the duration of contact119879contact(119860 119894) shorter than the 119879Location(119860 119894) the mean andstandard deviation of (2) is set up in (3) and (4) [42]
120583119879=119879Location(120583119865+ 120590119865) (3)
120590119879=119879Location(10120583119865+ 10120590
119865) (4)
in which 119879Location is the time of an activity in a specificlocation set in Table 2 120583
119879and 120590
119879are the mean and standard
deviation of contact frequency As a result the duration ofagent 119879contact(119860 119894) in specific locations (listed in Table 2) isalso discretized in Table 3 within the consideration of activitydifferences
414 Modelling Transportation In artificial Beijing threekinds of travel models are considered walking inside thedistrict travelling by road networks and travelling by subwaynetworks According to the traveling models the agentmovements are implemented by the compositions of travelingmodels before So it is important to build the basic road
14 Mathematical Problems in Engineering
Table 3 The contact frequency and duration in different locations and activities by social relationships in case of student and teacher
Social relationship Activity Location 119879contact(119860 119894) (minute) 119865contact(119860 119894)
Student-student
Sleep Dormitory 0 0
Breakfast Restaurant 119873(52) 119873(32)Breakfast Dormitory 119873(52) 119873(21)Sports Playground 119873(21) 119873(53)Class Classroom 119873(21) 119873(21)Study Library 119873(21) 119873(21)Lunch Restaurant 119873(158) 119873(32)
Shopping Convenience Store 119873(203) 119873(21)Dinner Restaurant 119873(108) 119873(21)Dinner Dormitory 119873(158) 119873(32)Rest Dormitory 119873(63) 119873(52)
Teacher-student Class Classroom 119873(21) 119873(21)
Subway network of Beijing city (red lines) Road network of Beijing city (green lines)
Figure 13 The implementation of transportation for artificial Beijing
and subway networks With the help of Google map thekey points of road are sampled first and the road networksare generated by the links among point Similarly subwaynetworks are generated by the links between subway stationsmapped from the real stations in Google map Figure 13shows the road and subway networks in artificial Beijingwith the road networks in green and the subway networks inred Most of the districts in Beijing city are covered by thesetwo networks The algorithm of hierarchical route planningis proposed as follows
(i) Find the next position where agent will be located innext action
(ii) Obtain the travelingmodels from the statistical trans-portation data
(iii) Search the path from the starting position to thenearest road entry or subway stations
(iv) Search the optimal path from the starting station orroad entry to the nearest station or road exit of targetposition
(v) Search the path from the target subway station or roadexit to the target position
(vi) Attach the subway train number or the bus numberincluding the transfer information to the path accord-ing to the timetable of subway or bus
(vii) Connect all the paths obtained before Generate thepath from the starting position to the target position
Based on the algorithm for each agent in artificial Beijingthe transportation is simulated during the computationalexperiments It is worth noting that the activities insidethe buses or train cars are considered because the agentcommunication during travelling is also necessary to besimulated
Mathematical Problems in Engineering 15
(a) The implementation of agent in MDT (b) The code framework of agent
Figure 14 The agent model implementation in MDT [35]
415 The Implementation of Models by MDT All the modelsare developed in the MDT Figure 14 shows the GUI of thetool MDT is used to implement the state space and statetransitions of FSM models The code framework is shownin Figure 14(b) Using the C++ inheritance features and thetechniques of dynamic link library (dll) the models areencapsulated in the dll componentsThe flexible compositionmechanism is designed to support the models evolution incomputational experiments
416 The Description of Artificial Beijing by ASE As men-tioned in Section 34 ASE describes the artificial Beijingin macroview The statistical information of Beijing city iscollected by ASE Table 4 lists part of the description ofartificial Beijing generated by ASE
42 How to Obtain the Data of Beijing City
421 Population and Geospatial Environment Database ofBeijing City Based on the models developed for ArtificialBeijing it is necessary to generate the initial data to do thecomputational experiments Because the individual level dataare not available it is necessary to construct an individual-based population database for both accurate computationalexperiment and determining optimal decisions
According to the state spaces of models discussed inSection 412 the geospatial and population database of Arti-ficial Beijing is designed as shown in Figure 15 The kernelpart of database consists of the tables such as agent list tablegeospatial environment list table household list table andagent distribution table Agent list table is an individual leveltable used to store the data to initialize the state space of agentmodel such as id and gender age Geospatial environmentlist table is also an individual level table which stores thedata of environment attributes such as id street id andtype Household list table and agent distribution table storethe statistical data for the data generation for agent andenvironment models With the help of the geospatial andpopulation database the data generating of artificial Beijingis proposed in the next section
422 Generating Geospatial and Social Environment Datafor Artificial Beijing by PGET The algorithm of generatinggeospatial and social environment data for artificial Beijingis used to quantify the spatial distribution of population andformalize geospatial behavior of each agent The algorithmshown in Figure 16 is capable of generating a syntheticartificial society which allowsmultiresolution statistical datasocial interactive behavior and multilayer social networksto be integrated together The synthetic population canrepresent individual agents in the form of households andhousehold members and the synthetic population is statis-tically equivalent to a real population For each householdcharacteristics such as address household size family typesand relationships are generated Each person is described bycharacteristics such as age gender social role and correlatedlocations The algorithm provides an effective methodologyto reconstruct the computing environment in high resolutionby using statistical data in low resolution leading to betterprediction and management of emergencies The details ofalgorithm are illustrated in [36]
As shown in Figure 17 the implementation of the algo-rithm is embedded inside the GEPTThe collective input datasuch as the data files of population distribution is collected inthe tool first then the algorithm is activated to generate thedatabase The generation lasts almost twelve hours one timewith the specified parameters settings When the generationis finished the data of artificial Beijing is obtained in thedatabase with 19610000 agents and 16000 environments
5 H1N1 Experiments in ArtificialBeijing with KD-ACP
With the support of the artificial Beijing modeled before thecomputational experiments and parallel execution are alsoimplemented by KD-ACP The experiments are designed tofind the most optimal emergency decision response plansEpidemic in city is a typical scenario in emergency manage-ment H1N1 epidemic in Beijing in 2009 is used to be the usedcase to test KD-ACP According to Section 34 it is necessarytomodel emergencymanagement first in order to support thecomputational experiments
16 Mathematical Problems in Engineering
Table 4 The description of artificial Beijing generated by ASE
Statistical features ofartificial Beijing Descriptions of statistical features Data sourcedefault values
Agent roles Baby child primary student middle school student college student worker retired
Demographic data file ofBeijing
Environment types Super market store restaurant park stadium hospital school cliniccommercial building dormitory building residential building
Demographic data file ofBeijing
Demographic data Age distribution of population sex distribution of population householddistribution of population retired age distribution children distribution
Demographic data file ofBeijing
Environment data Distribution of all types of environments Environment distributiondata file of Beijing
Relationships Classmate relative college friendship family Demographic data file ofBeijing
Varying parameters
Age difference of couples119873(120583120590) 119873(03) isin [0 12]Minimummarriage age (male female) (22 20)Age difference of children119873(120583120590) 119873(13) isin [1 10]Enrollment rate of kindergarten 120572 08Enrollment rate of primary school 120573 1Enrollment rate of middle school 120574 1Enrollment rate of college 120575 08Entrance age of kindergarten 1205721015840 3Entrance age of primary school 1205731015840 7Entrance age of middle school 1205741015840 13Entrance age of middle school 1205751015840 19
Age distributioninformation table
Agent settable
Geospatialenvironment set table
Household settable
Figure 15 The geospatial and population database of Beijing city
51 Modelling Emergency Management The modelling ofemergency management consists of two parts the modellingof emergent events and the modelling of intervention mea-sures H1N1 and the intervention measures for H1N1 aremodeled in our case
511 Modelling H1N1 by MDT H1N1 Model in agent simu-lates the states transitions of health status and the relevantactions Referring to SIR (SusceptibleInfectiveRecovered)[43 44] models agent has three health statuses susceptibleinfected and healthy with immunity Only susceptible agentcan be infected by the contact with infected agents After
the three infected phases (incubation being symptomaticand recovery) of H1N1 model agent is healthy again It isworthy mentioning that the recovered agents can be infectediteratively when the immunity disappears gradually by timeThe state transitions are illustrated in Figure 18
According to [45] the latent period of H1N1 was in aWeibull distribution within people from one to seven days[46 47] The distribution was usually centered in the rangebetween one and three days AWeibull random variable [40]is used to model the latent period as described in
119879Lat = 120572 times [minus ln (1 minus 120574)]1120573+ 120592 (119879Lat ge 120592) (5)
Mathematical Problems in Engineering 17
Censusdata
Generate GUID
Assign individualhousehold
Generate familystructure
Assign family roleto every member
Endow individualwith age gender
social role
Distribute environmentlocations on the map
School Workplace Entertainmentlocations
Consumptionlocations
Assign school orworkplace to
individual
Correlatedindividual withother locations
Generate friendshipin school
Generate coworkerrelations inworkplace
Assign individualadministrative
region
Generate housebuildings
Locate householdto house building
Generate familyrelation
Generateneighborhood
relation
PopulationEnvironment
Environment listSocial network
Figure 16 The algorithm of generating geospatial and social environment data [36]
Edit
Edit
Edit
Edit
Edit
Edit
BrowseStatistical characteristic of population
Geospatial population data
Population data of street
Population types
Geospatial and social environment generation tool
Workspaces
Parameters distribution settings
Household types
Roles in household
Household structure
Social roles
BrowseBrowseBrowseBrowseBrowseBrowse
Population distribution data
File path
Retired distributionPop distributionHousehold numHousehold distributionSex distributionAge distributionData type
Figure 17 The GUI of Population and Geospatial Environment generation Tool (PGET)
18 Mathematical Problems in Engineering
Infected
SusceptibleHealth
withimmunity
Infected
by contac
t
with in
fected
agents
Recovery
Increase before beingsymptomatic symptomatic
Decrease after being
Decrease by time
Infectivity
Immunity
Figure 18 The transitions of agent health status
119879Lat denotes the duration of latent period 120592 120572 and 120573 arethe location parameter scale parameter and shape parameterof Weibull distribution respectively 120574 is a uniform randomnumber in the range [0 1] According to statistics [47] 120592 120572and 120573 are set as 0 18 and 121 respectively Then the mean(standard deviation) of latent period is calculated as 159 days(058 day2) Furthermore infectious period is set as 7 days
A susceptible studentrsquos probability of getting infecteddepended on the infectivity of infectious agents his ownimmunity level the duration of the contact action and soon The infectivity of an infected agent evolved with timeduring the infectious period According to the statistics ofinfectivity in the chart with red bars in Figure 19 an infectedstudent had the highest level of infectivity in the second dayafter he had the first symptom The infectivity levels in otherdays in comparison with the second day are in Table 5 Theday labeled as ldquominus1rdquo means the day before the starting of thesymptomatic period
Based on the algorithm discussed before theH1N1modelis developed with the help of MDT The settings of influenzamodel will be discussed in Section 521
512 Modelling Intervention Measures by MDT In corre-spondence intervention measures model is used to ceasethe outbreak of H1N1 As discussed in [42 48 49] if theappropriate measures are taken when the infectious diseaseemerges the transmission of the disease could be sloweddown and the damage could be decreased Therefore it isnecessary and important to design emergency interventionmeasures in the artificial Beijing According to the MinistryofHealth inChina the intervention policies [50] are designedto control the spread of influenza such as H1N1 The inter-ventionmeasures of these policies including are interventionsactivated time vaccination rate antibiotic delay to hospitalisolation duration close workspace duration and limitationof activities listed in Table 6 Similarly the models of theseintervention measures are also developed with the help ofMDT Using the reasonable data ranges listed in Table 6intervention models can be initialized to do the computa-tional experiments These experiments are used to show how
Symptomatic1 2 3 4 5 6 7minus1
Recovered
Tim
e (da
y)
Infe
ctiv
ity
Figure 19 The distribution of infectivity rate of an infected agent[37]
Table 5 The proportion of infectivity
Days (119894) minus1 1 2 3 4 5 6 6Proportion of infectivity 03 05 1 08 04 02 01 005
to restrain the H1N1 transmissionThe settings of these inter-vention measures in IMCT are introduced in Section 522
52 Experiment Settings
521 H1N1 Settings by EECT EECT is designed to initializethe emergent events models such as H1N1The settings comefrom the specified emergent event models As shown inFigure 20 in theH1N1 used case the settings are composed ofthe distribution of infection source infection rate infectionperiod and so on The infection rate changes with envi-ronments and infection periods The settings are initializedfrom both the research of medicine on H1N1 and the data ofH1N1 outbreak in Beijing in 2009The process of influenza inartificial Beijing is simulated to repeat the possible scenariosof H1N1 break During the computational experiments ofartificial Beijing theH1N1 settings such as the period of beingsymptomatic are updated by the dynamic data registrationsfrom OsdRT The updating corrects the H1N1 model duringthe experiments With the help of OsdRT artificial Beijing isable to approach actual society It is also an implementationof parallel execution
522 Intervention Measures Settings by IMCT In the H1N1case the settings of intervention measures are shown inFigure 21 in IMCT These measures are designed under theinstruction of Emergency Decision Organizations like ChinaDisease Center (CDC) The measures such as vaccinationrate the rate of closedworkspace and the starting time can beinitialized in the tool Emergency response plans are imple-mented in these settings As illustrated in Figure 7 all the pos-sible compositions within value settings in reasonable inter-vals are tested to find the most optimal emergency responseplan For example different compositions of measures suchas vaccination rates and rates of workspace closed are usedto design the plans of experiments By the analysis of theresults the best composition of interventionmeasures can beobtainedThe analysis will be discussed in case study in detail
Mathematical Problems in Engineering 19
Table 6 The intervention measures of influenza
Intervention measures ValuesInterventions activated Time 119879Intervention isin [0 80th day]Vaccination rate 119877Vaccination isin [0 1]Antibiotic rate 119877Antibiotic isin [0 1]Delay to hospital 119867ospital(119860 119894) 119879infected lt 119905 lt 119879recovered + 119879HDelay 119879HDelay isin (0 3th day]Isolation duration 119868solation(119860 119894) 119879Intervention lt 119905 lt 119879Intervention + 119879IDuration 119879IDuration isin (0 14th day]Close workspace duration 119862lose(119864119894) 119879Intervention lt 119905 lt 119879Intervention + 119879CWDuration 119864119894 isinWorkspace 119879CWDuration isin (0 7th day]Limitation of agent activity duration 119871 imitation(119860 119894) 119879Intervention lt 119905 lt 119879Intervention + 119879LAADuration 119879LAADuration isin (0 7th day]119877Vaccination means the vaccination rate of artificial Beijing while 119877Antibiotic means the antibiotic rate used for infected agents in hospital119867ospital (119860119894) means that the infected agent 119860119894 is sent to hospital he or she will receive the retreatment immediately119868solation (119860119894) means that the infected agent 119860119894 is isolated at home he or she can only contact with the other agents until isolation duration is over119862lose (119864119894) means that the environment 119864119894 typed of workspace is closed until the close workspace duration is over119871 imitation (119860119894) means that the activities of agent 119860119894 are limited only the workspace and home are allowed until limitation duration is over
Incubation Weibulldistribution
Symbolic LogNormaldistribution
Recovering LogNormaldistribution
H1N1influenza
Settings of H1N1 influenza model
Disease model
Transmissionmodel
Figure 20 The settings of H1N1 in EECT
It is worth notifying that the settings of interventionmeasures are also influenced by OsdRT For example theH1N1 outbreak areas are notified by OsdRTThe interventionmeasures such as close workspace are used only in theoutbreak areas Therefore the unnecessary computation ofintervention measures is avoided So the data injection byOsdRT not only makes artificial society approach actualsociety but also increases the performance of computationalexperiments
53 Experiments by EMT When the initialization of arti-ficial Beijing is ready EMT is used to do the computa-tional experiments EMT is composed of two parts thecentral controller in the central node and the residen-tial service deployed in the computational nodes Centralcontroller loads the experiment plans and controls the whole
experiments process the GUI of controller is shown inFigure 22 EMT manages the computational experimentsin four steps cluster configuration load experiment plansmodels deployment and running experiments
531 Cluster Configuration Before the configuration clusterinformation is collected from the residential services insidethe cluster The information includes nodes amount nodenames node IPs and CPU occupations It is used to quantifythe computing power of the clusterThe cluster configurationis based on this information The nodes are selected forexperiments Then the LP (local process) [51] number foreach node is set according to the number of the CPU kernelsAfter the configuration the experiment plans are loaded tocustomize the models and data
20 Mathematical Problems in Engineering
Streets of close workspaces
Limitations ofactivities
IsolationsRespirator
Closeworkspace
Antibiotic
VaccinationHospital
Surveillance
Figure 21 The settings of intervention measures in IMCT
Information ofexperiment plan
List of experiment plans
Start | Pause | Reset | Stop | Speedup | Slowdown | Back
NodeName | CPU occupation | Memory occupation |Network | Models | Status
Figure 22 The deployment and running of experiments in EMT
532 Load Experiment Plan As discussed in Section 35 thedescription of experiment plans is composed of the infor-mation in three aspects models and the initial data for theexperiments the deployment mapping tables from the mod-els to the nodes and the plan of experiment executionWhenthe experiment plan is loaded by the EMT the file namesand model descriptions of models and data are listed Withthe support of cluster configuration models can be mappedto the nodes in the GUI according to the computationrequirements EMT also provides the setting of experimentsitself including the running times start time and end time
533 Models and Data Deployment Models are stored inthe agent population and services repository According tothe settings in experiment plan the customized models aredownloaded from the repository first then the models suchas agent model environment model emergent events modeland intervention measures model are integrated to be thecompositionmodels of artificial BeijingThemodels and datalocations in the nodes are configured in the ldquomodel pathrdquoand ldquodata pathrdquo in the GUI of central controller Finallythe models are uploaded to the nodes in cluster with theconsideration of partition The partition of artificial society
Mathematical Problems in Engineering 21
models is different from the multiagent system in [52]Models in artificial society are divided into three levels agentsand environments grids and city Agents and environmentsare encapsulated in grids which compose the whole cityThus the partition of artificial society becomes the partitionof grids The parallelism degrees are found to support thepartition of models agent independence and grid indepen-dence According to the parallelisms agents are partitioned toLPs according to the population distribution in grids Agentmovements among grids are simulated by the movementevents among grids in different LPs As a result a two-tierparallel architecture is proposed to support the simulation ofartificial society The architecture is detailed in [53]
534 Running of Experiments When the preparation isready the experiments can be done by the central controllerof EMT The running and control commands of the experi-ments are grouped into three aspects start pause or stop theexperiments get the running information of the nodes suchas the snapshot of the desktop of node With the help of thesecommands users can run the computational experimentsto do the research on the emergency management It isworth notifying that the simulation engine is optimized inorder to support the ten million agent simulation UsingGPU as coprocessor a two-tier parallel simulation engine isdesigned with support of MPI and OpenCL through phasedsynchronization [54] One-sided communication is used forreflection of remote simulation objects and message passingbetween processes A general kernel function prototype iselaborately designed and conditionally compiled for execu-tion on both CPU and GPU Moreover the optimizationoperations like load balance is developed under the instruc-tion of activity-based simulation [55] The densely populatedgrids are given more CPU and memory resource accordingto the activity predictions
54 Results of H1N1 Experiments with KD-ACP On thebasis of artificial society models and the intervention mea-sures mentioned in Section 512 a series of computationalexperiments are performed to study the H1N1 influenza inBeijing In our case of artificial Beijing 19610000 agentsare simulated in the cluster within 48 CPU cores and 128Gmemories the nodes are connected by kilomega networks Ittakes 18 hours to simulate a 250-day disease spreading
In order to illustrate how the optimal plans are obtainedbyKD-ACP the experiments are divided into four groups theexperiments ofmodel validation the experiments of sensitiv-ity analysis of vaccination rate the experiments of sensitivityanalysis of isolation and the experiments of the combinationof interventionmeasuresThese four groups show the generalprocess of the research on emergency management in H1N1influenza Firstly the models of H1N1 are validated withthe support of historical data Secondly the traditionalmedical interventions of pandemic such as vaccination areanalyzed to find the optimal plan Thirdly the traditionalnonmedical interventions of pandemic such as isolation areanalyzed to find the optimal plan Finally the combination ofinterventions is tested to show the combined effect
It is worth mentioning that all the experiments in ourwork are performed 100 times These experiments are ini-tialized with different settings of interventions in IMCT Theresults shown in the figures are the mean values
541 The Experiments of Model Validation In our artifi-cial Beijing the models of H1N1 influenza are built basedon the existing researches Wang and her colleagues hadstudied the H1N1 influenza by SEIR (SusceptibleExposedInfectiveRecovered) models [56] With the support of theirwork we build the models of artificial Beijing in the mannerof multiagents
In Figure 23 the control group shown in red line isdrawn based on the realistic statistical data collected in H1N1influenza in Beijing in 2009 The influenza lasts more thansix months and more than 170000 people are infectedComparing with the control group the simulated resultswithout interventions are drawn in blue It is obvious that thesimulated case fits the control group well before 70 days Butthe ascending of control group is slowed after 70 days andthe peak value (17883) is much less than the simulated results(25135) It is because the vaccination is activated at day 67 incontrol group according to the intervention measures usedin 2009 The restrain by vaccination becomes effective eightdays later The change of infected number after vaccinationis shown by red spotted line The trend of control group isdifferent from the simulation data after 75 days
However the simulated results (blue line) in compu-tational experiments fit the control group (as shown inFigure 23) in the first 70 days when no intervention mea-sures are executed The key features such as effective basicreproduction number (119877
0) are also almost the same in both
cases Therefore it can be concluded that the models ofartificial society built in microview are validated by therealistic statistical data in influenza in Beijing in 2009
In this group of experiments no interventions are exe-cuted in order to validate the models Therefore the simu-lated results of this extreme case which cannot happen todaycould be obtained As shown by blue line the number ofinfected agents without interventions reaches peak value atday 91 with 25135 consisting of 8052 in symptomatic phaseand 28950 in recovery phase It is obvious that the totalnumber of infected agents grows slowly at the beginningtime Then the number increases quickly when the influenzabreaks and it reaches its maximum value with 383870 in theend This phenomenon can be explained as in the followingAt the beginning of influenza due to the limited numberof infected agents the propagation of the disease remainsslow but many agents become susceptible When the numberof infected agents increases rapidly the spreading becomessignificant However the number of infected agents decreasesafter day 91 It is because the disease spreading relies on thesocial networks and spatial contact networks [49] It meansthat the infected agents can only infect agents either in thesocial relationship networks or in the spatially contacted
542 The Experiments of Sensitivity Analysis of VaccinationRate According to Section 512 vaccination is a typical
22 Mathematical Problems in Engineering
0
Infe
cted
agen
ts
0 50 100 150 200 250
Statistical dataSimulation data
25000
20000
15000
10000
5000
Day d
Figure 23 The results with no intervention executed and the com-parison with realistic statistical data in Beijing in 2009
medical intervention in influenza Traditionally the effective-ness of vaccination can only be evaluated after the influenzaand the key settings such as vaccination rate are determinedby the experience It is always difficult for emergency man-agement organizations to find the most optimal vaccinationrate in advanceTherefore the sensitivity of vaccination rate isanalyzed in this group of experiments The settings are listedin Table 7 average results are shown in Figure 24
Vaccination rates are set linearly (10 30 50 and70) to test the changes of influenza The peak value ofinfected agents decreases from 25135 22392 15787 6907to 638 respectively It shows that the more the agents arevaccinated the smaller the peak value of the infected agentswill be In the case of 70 vaccination the total infected agentis only 13799 It means that the influenza in this case does nothappen due to the high vaccination rate
The change of peak values also shows an interestingphenomenon Vaccination rates increase linearly the peakvalues decrease slower than the ascending of vaccination ratesat beginning but the descending becomes more and morefaster with the linear increase of vaccination rate It meansthat the small scale of vaccination (less than 30) does notmake senseWith the increase of vaccination rate the numberof susceptible agents decreases significantly In sequencethe possibility of infections is lowered greatly Additionallyit could be found that the decrease of peak values comeslater than the increase of the vaccination rates This is quitereasonable according to Mei et alrsquos work in [57]
As a result it is concluded that high vaccination rate iseffective in the control of H1N1 influenza though costly
543 The Experiments of Sensitivity Analysis of IsolationAccording to Section 512 isolation is a typical nonmedicalintervention It is the most used intervention measure in
Table 7 The initial settings of experiments of vaccination
Parameters ValueAgent count 19610000Initial infected count 40Interventions activated time(119879Intervention)
0
Vaccination rate (119877Vaccination) 0 10 30 50 70
0
Infe
cted
agen
ts
Loading intervention of vaccination
No vaccination10 vaccination30 vaccination
50 vaccination70 vaccination
0 50 100 150 200 250
25000
20000
15000
10000
5000
Day d
Figure 24 The results of loading intervention of vaccination
the emergency management in influenza As mentionedbefore a problem also exists that the measure cannot bevalidated before the intervention Therefore the sensitivityof isolation is analyzed in this group of experiments It isworth mentioning that the isolation is usually combinedwith the hospital policy It means the infected individual issent to hospital meanwhile the people who contact withthe infected are all isolated As a result parameters delay tohospital and isolation duration are designed together in thesettings of experiment Delay to hospital stands for the periodfrom infection time to hospital time while Isolation durationrepresents the period when an agent cannot contact others ifhe has contact with an infected agent The settings of theseparameters are listed in Table 8 average results are shown inFigure 25
Firstly the parameter delay to hospital is simulated aloneto test the sensitivity The simulated results are shown insolid line in Figure 25 The peak value decreases from 6708to 2218 when the delay to hospital decreases from 15 daysto 1 day The half day change brings the 67 reductionMoreover the peak value almost approximates to zero inthe case of 05 days The influenza does not happen in thiscase It can be concluded that the earlier the infected agents
Mathematical Problems in Engineering 23
Table 8 The initial settings of experiments of isolation
Parameters ValueAgent count 19610000Initial infected count 40Interventions activated time (119879Intervention) 80Delay to hospital (119879HDelay) isolation duration (119879IDuration) (15 0) (1 0) (05 0) (15 3) (1 3) (05 3)
0 50 100 150 200 2500
1000
2000
3000
4000
5000
6000
7000
Infe
cted
agen
ts
Loading intervention of isolation
Delay to hospital = 15
Delay to hospital = 10
Delay to hospital = 05
Delay to hospital = 15 isolation duration = 3
Delay to hospital = 10 isolation duration = 3
Delay to hospital = 05 isolation duration = 3
Day d
Figure 25 The results of loading intervention of isolation
are sent to hospital the less the agents will be infectedSo the construction of effective mechanism to find infectedindividuals is important in the emergency management ofinfluenza
Secondly the parameter isolation duration is added tosimulate the isolation of interventionObviously the isolationmeasure decreases the infected agents greatly Compared tothe parameter delay to hospital of 15-day case the peak valuedecreases from 6708 to 362when the isolation is added in theinterventionThe reduction of 6346 agents comes from the 3-day isolation of agents who had contacted with the infectedagents However the decrease of parameter delay to hospitalcannot bring the similar reduction in the 3-day isolationcases According to the figure the peak value decreases from362 to 51 from 15-day case to 05-day case only 311 agents aresaved from infection As a result the sensitivity of simulatedresults does not change obviously with delay to hospital inthe case of isolation It can be concluded that isolation is amore effective measure Though it is not easy for emergencymanagement organizations to decrease the delay to hospitaldue to the current monitoring mechanism the influenza canstill be controlled by isolation
544 The Experiments of Combination of Intervention Mea-sures Isolation is not the only nonmedical intervention ininfluenza close workspace is another common measureWhen close workspace is activated the workspaces are allclosed agents can only stay home at the closed time Luckilyisolation and close workspace are independent from oneanother It is possible for emergency management organi-zations to use these two measures together So this groupof experiments is designed to obtain the effectiveness of thecombination of intervention measures The parameters areinitialized with different settings listed in Table 9
Parameters delay to hospital and isolation duration is setin the middle values (1 day 3 days) according to the lastsection According to the experience close duration is set asseven days It is known that close workspace is a really costintervention measure the activated time is a key parameterWith the fixed parameters mentioned before the changeof interventions activated time is analyzed in this group ofcomputational experimentsThe average results are shown inFigure 26
Compared with the simulated results of experimentsdiscussed above Figure 26 strongly illustrates that influenzais ceased when the intervention measures are combinedtogether The peak value of infected agents is 141 in the ldquo80thdays of close-workspace and isolation with 1 day delay tohospitalrdquo case while 25135 in ldquono interventionrdquo case 2218 inldquo1 day delay to hospitalrdquo case 203 in ldquo1 day delay to hospitaland isolationrdquo case The interventions bring a great reductionof infection the peak values decrease from ten thousandslevel to hundred level
On the other hand the peak value even decreases to 108when the execution time of close workspace is 10 days earlierThe date of maximum value of infected agents is advancedfrom the 135th day to the 108th day The whole circle of H1N1influenza is reducedThe results of this group of experimentsshow that intervention measures should be activated as soonas possible and the compositions of interventions are moreeffective in emergency management of H1N1 influenza
In summary based on the general process lots of exper-iments could be done by KD-ACP to study emergency man-agement in H1N1 influenza Many conclusions are obtainedbased on the analysis of the experiments in this sectionAccording to these conclusions themost optimal response inour case study is the case of ldquo70th days of close-workspace andisolation with 1 day delay to hospitalrdquo It is because this casehas the fewest infected agents relatively But the modelling ofH1N1 influenza is not sufficient Economic and social factorsof interventions such as the money cost by vaccination andisolation the influence of the close-workspace in society
24 Mathematical Problems in Engineering
Table 9 The initial settings of experiments of isolation and closeworkspace
Parameters ValueAgent count 19610000Initial infected count 40Interventions activated time (119879Intervention) 70th day 80th dayDelay to hospital (119879HDelay) 1 dayIsolation duration (119879IDuration) 3 daysClose workspace duration (119879CWDuration) 7 days
50 100 150 200 2500
50
100
150
200
Infe
cted
agen
ts
Loading intervention of isolation and close workspace
Delay to hospital = 1 isolation duration = 3Start close time = 70 close duration time = 7Delay to hospital = 1 isolation duration = 3Start close time = 80 close duration time = 7
Day d
Figure 26The results of loading interventions of isolation and closeworkspace
and even opinion transmission of H1N1 in internet are notconsidered in our case study Without the analysis of thesefactors it does not make sense to talk about the most optimalintervention measures However KD-ACP had provided asoftware framework and general process of the research onemergency management Supported by KD-ACP the expertsof emergency management are able to study emergent eventssuch as H1N1 influenza as detailed as possible Based on theexperiments and analysis again and again the most optimalintervention measures will be found ultimately
55 Parallel Execution by ASST and OsdRT As discussed inSection 35 ASST andOsdRT are used to support the parallelexecution of artificial Beijing The statistical data is collectedand stored in artificial society runtime database at runtimeASST downloads the infection data from the database anddisplays it in the map As shown in Figure 27 the infectedagents are distributed in the districts in BeijingThe snapshotgives the geospatial distribution of the H1N1 epidemic Thelegends of the agents in different status are listed in the left
right corner in Figure 27 Five statuses are shown in the figurehealth without vaccination health with vaccination incuba-tion being symptomatic and convalescence ASST providesa very direct feeling for the emergency experts Partly withthe situation from ASST Emergency Decision Organizationmakes the emergency response plans to the response of theemergency scenario With the help of plans the epidemicin actual society can be restrained In the meantime theinfluenza information is reflected in the Internet Accordingto Figure 8 the data acquisition data extraction and datastandardization of H1N1 influenza are processed in OsdRTOsdRT is implemented by Dr Caorsquos team from Institute ofAutomation from Chinese Academy of Sciences [58] Thetool provides not only the reports of H1N1 SARS hand-foot-mouth disease and intestinal diseases are all monitored Theweb page is shown in Figure 28 H1N1 reports from Internetare collected first Based on the data extraction of the reportsH1N1 case of distribution is obtained and shown in theincidencemap As a result H1N1 incidence data is regarded asthe knowledge which is sent to EECT and IMCTThe settingsof emergent events and intervention measures are modifiedin order to make artificial society approach actual society innext turn of computational experiments With the help ofOsdRT the loop of parallel execution process integrates ACPapproach
6 Conclusion
This work provides an integrated software framework forsocial Computing in emergencymanagement From a systemperspective KD-ACP provides a reliable flexible low costand effective platform for the scientists to do the research onemergency response problems The analysis and predictionof these problems with inherent complexity can be solvedby the repetition of possible alternative experiments onthe software framework Therefore KD-ACP which is anattempt to implement ACP approach seemed as an effectivecomputational framework to support the decision making inemergency response
Currently KD-ACP is used to study the H1N1 epidemicin Beijing in 2009 The most optimal compositions of inter-vention measures are testified to support the response tonext influenza However it is still a long way ahead beforethe automation systematization and practicalization of KD-ACP A lot of work will be carried out along several directionsas in the following
(i) The parallel execution process in KD-ACP is actuallynot strongly connected with the emergency responseorganization Traditional experience and theory stillplay a dominant role in the decision of emergencymanagement It is necessary to propose a mechanismof hall for workshop to facilitate the parallel executionloop in actual society
(ii) Section 33 introduces the principle and process ofmodelling artificial society Metamodels and modelsof artificial Beijing are built by GME Model transfor-mation and code generation are used to implementcodes from FSM based models But many details
Mathematical Problems in Engineering 25
Fangshan district Daxing district Fengtai district Xicheng district
Mentougou district Haidian districtShijingshan district Dongcheng district
Incubation
Symptomatic
Convalescence
Health withoutvaccination
Health withvaccinationLegend of agent
health status
Figure 27 The situation of artificial Beijing by ASST
H1N1
Districts of Beijing
H1N1 case ofdistribution
Incidence map
Distribution ofH1N1 reportsfrom internet
Incidence of H1N1
Fangshan districtLiangxiang town
Population 19000Incidence 135789
(1100000)
Figure 28 The implementation of OsdRT by Institute of Automation from Chinese Academy of Sciences
are still not illustrated due to the topic of thispaper The key techniques such as template of codeframework and model transformations from specificdomains to FSM will be focused and introduced indetail in our next paper
(iii) The performance of computational experiments inKD-ACP is still slower than expectation The opti-mization of simulation engine and models structureis necessary to improve the performance So it ispossible to run artificial society with the real worldin parallel
In summary KD-ACP paves a new way for scientists inboth emergencymanagement and computation simulation to
collaborate with each other in solving the emergency man-agement problems by social computing
Research Highlights
(i) KD-ACP is implemented based on ACP approach
(ii) Artificial Beijing is built with the help of KD-ACP
(iii) H1N1 is modeled to simulate the influenza in Beijingin 2009
(iv) Computational experiments testify the effectivenessof intervention measures
26 Mathematical Problems in Engineering
Conflict of Interests
The authors declare that there is no conflict of interestsregarding the publication of this paper
Acknowledgment
This work was supported in part by the National ScienceFoundation of China (nos 91024030 71303252 61403402and 91324013)
References
[1] W Mao and F-Y Wang Advances in Intelligence and SecurityInformatics Academic Press Oxford UK 2012
[2] E Bonabeau ldquoAgent-based modeling methods and techniquesfor simulating human systemsrdquo Proceedings of the NationalAcademy of Sciences of the United States of America vol 99 no3 pp 7280ndash7287 2002
[3] K M Carley D B Fridsma E Casman et al ldquoBioWar scalableagent-basedmodel of bioattacksrdquo IEEETransactions on SystemsMan and Cybernetics A Systems and Humans vol 36 no 2 pp252ndash265 2006
[4] J M Epstein ldquoModelling to contain pandemicsrdquo Nature vol460 no 7256 p 687 2009
[5] Los Alamos National Laboratory Reports about EPISIMS LosAlamos National Laboratory 2003
[6] C H Builder and S C Bankes ldquoArtificial societies a conceptfor basic research on the societal impacts of informationtechnologyrdquo RAND Report 149 RAND Corp 1991
[7] F-Y Wang ldquoComputational theory and method on complexsystemrdquo China Basic Science vol 6 no 5 pp 3ndash10 2004
[8] F-Y Wang ldquoToward a paradigm shift in social computing theACP approachrdquo IEEE Intelligent Systems vol 22 no 5 pp 65ndash67 2007
[9] F-Y Wang and S Tang ldquoArtificial societies for integratedand sustainable development of metropolitan systemsrdquo IEEEIntelligent Systems vol 19 no 4 pp 82ndash87 2004
[10] D Wen Y Yuan and X-R Li ldquoArtificial societies computa-tional experiments and parallel systems an investigation ona computational theory for complex socioeconomic systemsrdquoIEEE Transactions on Services Computing vol 6 no 2 pp 177ndash185 2013
[11] F Wang F Zeng and Y Yuan ldquoAn ACP-based approach forcomplexity analysis of E-commerce systemrdquo Complex Systemsand Complexity Science vol 5 no 3 pp 1ndash8 2008 (Chinese)
[12] F-Y Wang and P K Wong ldquoResearch commentary intelligentsystems and technology for integrative and predictivemedicinean ACP approachrdquoACMTransactions on Intelligent Systems andTechnology vol 4 no 2 pp 1ndash6 2013
[13] J Sifeng X Gang F Dong and H Chunpeng ldquoStudy on theemergency rescue decision support system of petrochemicalplant based on ACP theoryrdquo in Proceedings of the 6th Manage-ment Annual Meeting (MAM rsquo11) 2011 (Chinese)
[14] X Dong D Fan G Xiong F Zhu Z Zhang and Y Yao ldquoPar-allel Bus Rapid Transit (BRT) operation management systembased on ACP approachrdquo in Proceedings of the 6th ManagementAnnual Meeting (MAM rsquo11) Beijing China 2011
[15] B Ning F-Y Wang H-R Dong R-M Li D Wen and L LildquoParallel systems for Urban rail transportation based on ACP
approachrdquo Journal of Transportation Systems Engineering andInformation Technology vol 10 no 6 pp 23ndash28 2010 (Chinese)
[16] F-Y Wang J Zhao and S-X Lun ldquoArtificial power systemsfor the operation and management of complex power gridsrdquoSouthern Power System Technology vol 2 no 3 pp 1ndash11 2008(Chinese)
[17] M J North T R Howe N T Collier and J R Vos ldquoRepastsimphony development environmentrdquo in Proceedings of theAgent 2005 Conference on Generative Social Processes Modelsand Mechanisms 2005
[18] M J North T RHowe N T Collier and J R Vos ldquoRepast Sim-phony runtime systemrdquo in Proceedings of the Agent Conferenceon Generative Social Processes Models and Mechanisms 2005
[19] SWARM Development Group Swarm 211 Reference Guide2000 httpwwwswarmorg
[20] S Tisue and U Wilensky ldquoNetLogo a simple environment formodeling complexityrdquo in International Conference on ComplexSystems pp 16ndash21 2004
[21] S Luke C Cioffi-Revilla L Panait K Sullivan and G BalanldquoMASON a multiagent simulation environmentrdquo Simulationvol 81 no 7 pp 517ndash527 2005
[22] M Ichikawa H Tanuma Y Koyama andH Deguchi ldquoSOARSintroduction as a social microscope for simulations of socialinteractions and gamingrdquo in Organizing and Learning throughGaming and Simulation Proceedings of the 38th Conference ofthe International Simulation and Gaming Association (ISAGArsquo07) pp 149ndash158 2007
[23] D S Burke J M Epstein D A T Cummings et al ldquoIndividual-based computational modeling of smallpox epidemic controlstrategiesrdquo Academic Emergency Medicine vol 13 no 11 pp1142ndash1149 2006
[24] F-Y Wang ldquoParallel control and management for intelligenttransportation systems concepts architectures and applica-tionsrdquo IEEE Transactions on Intelligent Transportation Systemsvol 11 no 3 pp 630ndash638 2010
[25] Z Molnar D Balasubramanian and A Ledeczi ldquoAn introduc-tion to the genericmodeling environmentrdquo in Proceedings of theTOOLS Europe 2007 Workshop on Model-Driven DevelopmentTool Implementers Forum Zurich Switzerland 2007
[26] S Kelly and J-P Tolvanen Domain Framework in Domain-Specific Modeling Enabling Full Code Generation JohnWiley ampSons Hoboken NJ USA 2007
[27] B P Zeigler H Praehofer and T G Kim Theory of Modelingand Simulation Integrating Discrete Event and ContinuousComplex Dynamic Systems Academic Press San Diego CalifUSA 2000
[28] G Booch J Rumbaugh and I JacobsonThe Unified ModelingLanguage User Guide Addison-Wesley Reading Mass USA1999
[29] F Wagner R Schmuki T Wagne and P Wolstenholme Mod-eling Software with Finite State Machines A Practical ApproachCRC amp Taylor amp Francis Boca Raton Fla USA 2006
[30] T Espiner ldquoChina builds worldrsquos fastest supercomputerrdquo ZDNetUK October 2010
[31] B Chen and G Guo ldquoA two-tier parallel architecture for arti-ficial society simulationrdquo in Proceedings of the ACMIEEESCS26th Workshop on Principles of Advanced and DistributedSimulation (PADS rsquo12) pp 184ndash186 July 2012
[32] G Guo B Chen X G Qiu and Z Li ldquoParallel simulation oflarge-scale artificial society on CPUGPU mixed architecturerdquo
Mathematical Problems in Engineering 27
in Proceedings of the ACMIEEESCS 26th Workshop on Prin-ciples of Advanced and Distributed Simulation (PADS rsquo12) pp174ndash177 July 2012
[33] S-X Lun ldquoResearch on the classification of parallel executionmodes of ACP theoryrdquo Acta Automatica Sinica vol 38 no 10pp 1602ndash1608 2012
[34] F-YWang ldquoA framework for social signal processing and anal-ysis from social sensing networks to computational dialecticalanalyticsrdquo Science China vol 43 no 12 pp 1598ndash1611 2013
[35] G GuoOne Model Userrsquos Manual version 12 National Univer-sity of Defense University 2011
[36] Y Ge R Meng Z Cao X Qiu and K Huang ldquoVirtual citymdashanindividual-based digital environment for human mobility andinteractive behaviorrdquo Simulation Transactions of the Society forthe Modeling and Simulation International 2014
[37] M Kretzschmar and R T Mikolajczyk ldquoContact profiles ineight European countries and implications for modelling thespread of airborne infectious diseasesrdquo PLoS ONE vol 4 no6 Article ID e5931 2009
[38] W Duan and X Qiu ldquoFostering artificial societies using sociallearning and social control in parallel emergency managementsystemsrdquo Frontiers in Computer Science vol 6 no 5 pp 604ndash610 2012
[39] W J Edmunds C J OrsquoCallaghan and D J Nokes ldquoWhomixes with whom Amethod to determine the contact patternsof adults that may lead to the spread of airborne infectionsrdquoProceedings of the Royal Society B Biological Sciences vol 264no 1384 pp 949ndash957 1997
[40] J Banks J S Carson B L Neison and D M Nicol Discrete-Event System Simulation Prentice Hall Englewood Cliffs NJUSA 4th edition 2007
[41] H Yongxia ldquoInvestigation and analysis of college studentsrsquo workand restrdquo Value Engineering vol 31 no 3 2012 (Chinese)
[42] W Duan Z Cao Y Wang et al ldquoAn ACP approach to publichealth emergency management using a campus outbreak ofh1n1 influenza as a case studyrdquo IEEE Transactions on SystemsMan and Cybernetics Part A Systems and Humans vol 43 no5 pp 1028ndash1041 2013
[43] HWHethcote ldquoThemathematics of infectious diseasesrdquo SIAMReview vol 42 no 4 pp 599ndash653 2000
[44] G Chowell E Shim F Brauer and et al ldquoModelling thetransmission dynamics of acute haemorrhagic conjunctivitisapplication to the 2003 outbreak in Mexicordquo Statistics inMedicine vol 25 no 11 pp 1840ndash1857 2006
[45] Y Ge L Liu B Chen X Qiu and K Huang ldquoAgent-basedmodeling for InfluenzaH1N1 in an artificial classroomrdquo SystemsEngineering Procedia vol 2 pp 94ndash104 2011
[46] A R Tuite A L Greer MWhelan et al ldquoEstimated epidemio-logic parameters andmorbidity associatedwith pandemicH1N1influenzardquo CMAJ vol 182 no 2 pp 131ndash136 2010
[47] L Brouwers B Cakici M Camitz A Tegnell and MBoman ldquoEconomic consequences to society of pandemic H1N1influenza 2009mdashpreliminary results for Swedenrdquo Euro Surveil-lance vol 14 no 37 pp 1ndash7 2009
[48] Y Ge L Liu X Qiu H Song Y Wang and K Huang ldquoAframework of multilayer social networks for communicationbehavior with agent-based modelingrdquo Simulation Transactionsof the Society for the Modeling and Simulation International vol89 no 7 pp 810ndash828 2013
[49] W Duan X Qiu Z Cao X Zheng and K Cui ldquoHeterogeneousand stochastic agent-based models for analyzing infectious
diseasesrsquo super spreadersrdquo IEEE Intelligent Systems vol 28 no4 pp 18ndash25 2013
[50] The Preparation and Emergency Response Plans to PandemicPublished by Ministry of Health Ministry of Health 2005(Chinese)
[51] R M Fujimoto Parallel and Distributed Simulation SystemsWiley Interscience 2000
[52] F Cicirelli A Giordano and L Nigro ldquoDistributed simulationof situated multi-agent systemsrdquo in Proceedings of the 15thIEEEACM International Symposium on Distributed Simulationand Real Time Applications (DS-RT rsquo11) pp 28ndash35 September2011
[53] B Chen and G Guo ldquoA two-tier parallel architecturefor artificial society simulationrdquo in Proceedings of the 26thACMIEEESCS Workshop on Principles of Advanced and Dis-tributed Simulation (PADS rsquo12) pp 184ndash186 July 2012
[54] G Guo B Chen and X Qiu ldquoParallel simulation of large-scale artificial society with GPU as coprocessorrdquo InternationalJournal of Modeling Simulation and Scientific Computing vol4 no 2 Article ID 1350005 2013
[55] B Chen L-B Zhang X-C Liu and H Vangheluwe ldquoActivity-based simulation using DEVS increasing performance by anactivity model in parallel DEVS simulationrdquo Journal of ZhejiangUniversity Science C vol 15 no 1 pp 13ndash30 2014
[56] X-L Wang P Yang Z-D Cao et al ldquoQuantitative evaluationon the effectiveness of prevention and control measures againstpandemic influenza A (H1N1) in Beijing 2009rdquoChinese Journalof Epidemiology vol 31 no 12 pp 1374ndash1378 2010 (Chinese)
[57] S Mei A D van de Vijver L Xuan Y Zhu and P M ASloot ldquoQuantitatively evaluating interventions in the influenzaA (H1N1) epidemic on China campus grounded on individual-based simulationsrdquo Procedia Computer Science vol 1 no 1 pp1675ndash1682 2010
[58] X Qiu Z Cao and Z Li Mid-Term Reports of China NationalScience Foundation 91024030 2014
Submit your manuscripts athttpwwwhindawicom
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
MathematicsJournal of
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Mathematical Problems in Engineering
Hindawi Publishing Corporationhttpwwwhindawicom
Differential EquationsInternational Journal of
Volume 2014
Applied MathematicsJournal of
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Probability and StatisticsHindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Journal of
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Mathematical PhysicsAdvances in
Complex AnalysisJournal of
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
OptimizationJournal of
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
CombinatoricsHindawi Publishing Corporationhttpwwwhindawicom Volume 2014
International Journal of
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Operations ResearchAdvances in
Journal of
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Function Spaces
Abstract and Applied AnalysisHindawi Publishing Corporationhttpwwwhindawicom Volume 2014
International Journal of Mathematics and Mathematical Sciences
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
The Scientific World JournalHindawi Publishing Corporation httpwwwhindawicom Volume 2014
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Algebra
Discrete Dynamics in Nature and Society
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Decision SciencesAdvances in
Discrete MathematicsJournal of
Hindawi Publishing Corporationhttpwwwhindawicom
Volume 2014 Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Stochastic AnalysisInternational Journal of
4 Mathematical Problems in Engineering
A part
C part
P part
Population and GeospatialEnvironment Generation Tool (PGET)
Generic ModelingEnvironment (GME)
Artificial society population andgeospatial environment database
Initialize
InitializeInitialize
Model DevelopmentTool (MDT)
FSM based models
Artificial SocietyEditor (ASE)
Scenario ofartificial society
Artificial societyruntime database
Agent and emergencymodel repository
Experiments DesignTool (EDT)
Experiment plans
InitializeSettings of
emergent eventsEmergent Events
Configuration Tool (EECT)Intervention Measures
Configuration Tool (IMCT) Settings of intervention measuresemergency response plans
Experiments ManagementTool (EMT)
Population data emergent eventsdata and intervention measures
Open source dataRegistration Tool (OsdRT)
Actu
al so
ciet
y Artificial SocietySituation Tool (ASST)
Optimal emergencyresponse plan
Internet
C++ code of modelsdll of models
Figure 2 The software architecture of KD-ACP
In the ldquoArdquo part Generic Modeling Environment (GME)[25] and Model Development Tool (MDT) are the kerneltools in the modelling of artificial society GME is anopen source modelling tool which supports domain-specificmodelling The domains of artificial society are created byGME in our work Models such as agent environmentemergent event and intervention measure are described inspecific domains first in GME With the help of modeltransformation thesemodels are all transformed to the FiniteState Machine (FSM) models Meanwhile code generationsare supported byMDT and thesemodels are all implementedin C++ Artificial Society Editor (ASE) is used to describethe concrete scenario of actual society which defines thescope of models set for artificial society Population andGeospatial Environment generation Tool (PGET) generatesthe population and geospatial environment data with thesupport of statistical data from actual society
In the ldquoCrdquo part Emergency Events Configuration Tool(EECT) initializes the models of emergent events whileInterventionMeasures Configuration Tool (IMCT) initializesthe models of intervention measures Experiments plans aregenerated byExperimentsDesignTool (EDT) Based on theseplans Experiments Management Tool (EMT) is used to runand manage the computational experiments to study theemergency problems
In the ldquoPrdquo part Artificial Society Situation Tool (ASST)seemed as the monitor of running artificial society Thestatistical data and situation are shown by ASST at runtimeIn the meantime the emergency response plans are made by
emergency decision organizations Parts of the influences ofemergency plans are reflected on Internet Open source dataRegistration Tool (OsdRT) is used to register the open sourcedata from Internet to artificial society
KD-ACP is developed using the BrowserServer archi-tecture the tools are integrated in the home page of KD-ACP as shown in Figure 3 Each tool is activated by the clickon the link For example Artificial Society Editor is startedwhen the link of ASE is clicked The working environmentand programming languages of tools in KD-ACP are listed inTable 1
Moreover the implementation of KD-ACP is mainlycomposed of modelling phase and computational experi-ments phase It will be discussed in the next section
33 The Modelling of Artificial Society in KD-ACP It is acritical problem to focus on the key parts of society in socialcomputing Based on the ACP approach the bottom-upmodelling is used to build the artificial society As a resultmodelling of artificial society is composed of three basic ele-ments agents environments and rules for interactionsHow-ever we still meet the problem that specific features shouldbe supported in artificial society For example emergentevents and intervention measures are required in artificialsociety for emergency management The modelling of onlybasic elements cannot cover the specific features in domainsTherefore domain-specific modelling [26] is introduced tosolve the problems in modelling artificial society
Mathematical Problems in Engineering 5
Model DevelopmentTool (MDT)
Artificial SocietyEditor (ASE)
Population and GeospatialEnvironment Generation Tool (PGET)
Emergent EventsConfiguration Tool (EECT)
Intervention MeasuresConfiguration Tool (IMCT)
Experiments DesignTool (EDT)
Experiments ManagementTool (EMT)
Artificial SocietySituation Tool (ASST)
Open Source DataRegistration Tool (OsdRT)
Generic ModelingEnvironment (GME) Artificial
society (A)Computationalexperiments (c)
MDT EECT IMCT
ASST OsdRT
ASE PGET EDT EMT
Parallel execution (P)
Figure 3 The implementation of KD-ACP
Table 1 The working environment and programming languages of tools
Tools Working environment Programminglanguages
Developmentplatform User type
Generic ModelingEnvironment (GME)
General computer (desktopapplication) NULL NULL Domain experts
Model Development Tool(MDT)
General computer (desktopapplication) C++ Visual Studio Model developers
Artificial Society Editor(ASE)
General computer (desktopapplication) C Visual Studio Domain experts
Population and GeospatialEnvironment generationTool
General computer (desktopapplication) C Visual Studio Domain experts
Emergency EventsConfiguration Tool (EECT)
General computer (withInternet Explorer clientside)
C ASPNET Domain experts
Intervention MeasuresConfiguration Tool (IMCT)
General computer (withInternet Explorer clientside)
C ASPNET Domain experts
Experiments ManagementTool (EMT)
General computer (withInternet Explorer clientside)
C ASPNET
Users of computationexperimentServer of EMT
Nodes in supercomputer(Console Program serverside)
C++ Visual Studio
Runtime Infrastructure ofEMT
Nodes in supercomputer(Console Program serverside)
C++ Visual Studio
Artificial Society SituationTool (ASST)
General computer (desktopapplication) C++ Visual Studio Domain experts
Open Source DataRegistration Tool (OsdRT)
General computer (withInternet Explorer clientside)
Java JSP Domain experts
6 Mathematical Problems in Engineering
M2T transformation
Modeling in domain
GME Generic Modeling Environment
Metamodeling
Instance of
FSM
Metamodeling for artificial society
DEVS
Petri Net State charts
Agentmetamodel
Environmentmetamodel
Emergency eventmetamodel
Interventionmetamodel
DAE
Domain-specific metamodels Semantic well definedmetamodels
M2M transformation
Modeling of artificial society
Interventionmodel
Emergency eventmodel
Environmentmodel
Agentmodel
FSM based emergent andintervention models
FSM based environmentmodels
FSM based agent modelsDomain-specific models
Semantic well defined models
M2M transformation
Agent model code Environment model code
Agent behavior model
Agent mentality model
Disease model code
Basic population model Disease spreading model
Disease state transformmodel
Intervention model code
Agent social networkmodel
Basic environmentinformation model
Roadnet model
Vaccination
Infectors isolation
Contactors isolation
Close work officesAgent contact model
Domain-specific models relevant codes
MDT Model Development Tool
Implementation
Figure 4 From metamodelling and modelling by GME to implementation by MDT of artificial society
331 The Principle of the Modelling of Artificial SocietyAccording to the principle of domain-specific modellingthe modelling of artificial society contains the followingsteps first metamodelling the basic elements of artificialsociety second modelling the specific features in domain ofemergency management third implementing the models ofartificial society in codes The whole process is illustrated inFigure 4The first and second steps are implemented in GMEwhile the third step is implemented in MDT
The first step is metamodelling which mainly focuses onconstructing the metamodels of artificial society Metamod-elling tries to study the common patterns of artificial societyThe outputs of metamodelling are metamodels which repre-sents the abstraction of the whole systemThe basic elementsof artificial society are described in metamodel The processof metamodelling is divided into four phases The first is theconstruction of the domain-specific metamodels As shown
in Figure 4 agent metamodel environment metamodelemergent event metamodel and intervention metamodelcompose the metamodel of artificial society The second isthe construction of the metamodels described by typicalmodelling formalisms such as FSM DAE DEVS and PetriNet [27] These formalisms are all semantically well definedThe third is the definition of the model transformationfrom domain-specific metamodels to metamodels of typicalmodelling formalisms The transformation standardizes themetamodels of artificial society by typical modelling spec-ifications The fourth is the definition of the transforma-tion templates from metamodels to code framework Thetemplates list the basic abstract interfaces of metamodels ofartificial societyThese abstract interfaces are implemented inthe specific-domain modelling and code generations
The second step is modelling the models of artificialsociety such as agent model environment model emergent
Mathematical Problems in Engineering 7
Figure 5 The metamodels of artificial society in GME
event model and intervention model are built Actuallythe models are the instantiation of metamodels in the laststep Different from the general modelling environment likeUML [28] the domain-specific modelling provides a familiarmodelling environment for the domain experts in artificialsociety For example emergency response experts only con-cern emergent eventmodel and interventionmodel inheritedfrom metamodels After constructing the domain-specificmodels based on domain-specific metamodels domain usersexecute the model transformation defined in the first stepAll the models of artificial society are transformed into FSMmodels As a result the models are implemented in this uni-fied modelling formalism (FSM) The model transformationmakes the simulation of the models possible
The third step is the generation of executable codes ofmodels The executable code framework is generated bymapping template from metamodels to code frameworkdefined in the first step Moreover domain developers alsoadd necessary codes to the framework to integrate thedynamic semantics of the models The code frameworkoutputs the dynamic link libraries by compilingThe dynamiclink libraries are loaded in the large scale artificial societyruntime infrastructure in computational experiments
332 The Metamodelling and Modelling of Artificial Societyby GME GME is used to build metamodels and modelsin our work As mentioned before the abstraction andcommon patterns of society are represented in metamodelsAccording to the bottom-up modelling style metamodels ofagent environment and communications are described inGME Figure 5 shows part of metamodels of artificial societyThe features of an agent metamodel are extracted from thecensus figures and statistical data Environment metamodelsimulates the geospatial places for the behaviors of agentsThemetamodel of communications among agents ismodeledto simulate the interactions such as infection in epidemicsand rumor propagation in public opinion formation eventsIt is worth notifying that the metamodel of communications
includes both the emergent event metamodel and interven-tion metamodel
From the perspective of modelling the details from spe-cific domains are considered in the models by the instantia-tion frommetamodels of artificial society For example socialrelationships based on complex networks are added in agentmodel to support the communications Agent activity is alsoused to quantify the agent activity under different scenarioEnvironment models are linked with the help of transporta-tion services subways and roads are modeled while the pathsearch is encapsulated in the services Emergent event modeland intervention model are also the domain-specific modelsThe modelling of artificial Beijing in GME will be discussedin detail in next section
333 The Implementation of Models of Artificial Society byMDT As mentioned before MDT is used to implementmodels such as agent environment emergent event andintervention According to the template of code frameworkthe implementations of models are generated by MDT Theimplementations are classified into two categories FSMmodels and services FSM models such as agents andenvironments are built under the specification of FiniteState Machine (FSM) [29] in MDT while all the servicessuch as transportation are encapsulated under the PublicService Standard This standard provides a generic interfacespecification for modelers to encapsulate public commonservices in artificial society FSM models like agent are builtstatistically from the quantitatively analyzed characteristicssuch as demographic attributes social behaviors emergencybehaviors and social networks Social behaviors describe thedaily behaviors of individuals while emergency behaviorsdescribe the individual behaviors in emergent events Forexample infected individuals are all isolated in hospital inSARS Isolation ismodeled as a typical emergency behavior inour work Correspondingly services are used to simulate themacroactual society Take transportation service for instancethe path search is needed by almost every agent duringmoving from spot to spot
8 Mathematical Problems in Engineering
Artificial society description in emergency
Description ofemergency
ArtificialSociety Editor
(ASE)
Agent behavior set
Agent role setAgent status set
Environment setAgent relationship set
Statistical data of real society
Descriptionof artificial
societyframework
Terminal of agent modeling andartificial society describing
middot middot middotAgent
metamodeling
Customized agentmodeling
Relationship modelDailyEmergency
Mod
el D
evelo
pmen
t Too
l (M
DT)
Agent modeling
Agent model Encapsulate
Encapsulate
Behavior model
Intervention
Emergency
Response
Daily
Social relationshipmodeling
Socialbehaviormodeling
Socialbehaviormodel
Emergencybehavior
model
Emergencybehaviormodeling
Building
Geospatial statistical data
Population statistical data Datacollection
Actualsociety
Artificial societypopulation and
geospatial database
Organizationservice
Transportationservice+
MappingAgent model andservice repository+ emergency model
repository
Climateservice
Geographicalservice
+
Agent status in emergencyEmergency management
Emergent event set
Social relationship statistical data
Environment statistical data
Population behaviorstatistical data
Emergency statistical data
Emergency organization statistical data
Geospatial and SocialEnvironment Generation
Tool (GSeT)
Agent model repository andpopulation database
C++C++C++C++
Modeling
Statistical modelJob ageSex location
Public service modeling
Organizationmodel
Transportationmodel
Geographicalmodel
Climatemodel
Interventionservice
Publ
ic se
rvic
e sta
ndar
d
Figure 6 The editing and initialization of KD-ACP
MDT provides domain experts with a tool to obtain thecode implementations of models With the help of compileenvironment like visual studio MDT also supports thefurther programming development of the specific domaindetails which cannot be described in modelling step
Both the FSM models and services are developed by theMDT first and then stored in the agent model and servicerepository The repository manages the models according tothe requirements from emergency problems and provides themodels for EDT to make the experiment plans
34 The Editing and Initialization of Artificial Society byASE and PGET As shown in Figure 6 ASE is used to editthe scenarios of artificial society within emergent eventsThe editing is composed of two parts (1) the statisticalinformation of artificial society in daily life such as the rolesof agents the relationships of agents and the types of environ-ments and (2) the statistical information of artificial societyin emergency including the statistical data of emergent
events the emergency organization and emergency relatedbehaviors of agents
According to the requirements of the editing thesestatistical data are collected from actual society manually bythe domain experts Based on these statistical data PGETgenerates the artificial society population and geospatialenvironment database The database supports the instanti-ation of artificial society at individual level For examplethe attributes such as age and gender of each agent can befound in the database With the support of the databaseFSM models and service repository discussed before itis sufficient for domain experts to study the emergencyproblems by computational experiments
35The Computational Experiments and Parallel Execution inKD-ACP The tools of ldquoCrdquo part and ldquoPrdquo part in KD-ACP areused to support the process of computational experimentsand parallel execution The working process is shown inFigure 7 EECT and IMCT are both the starting and ending
Mathematical Problems in Engineering 9
Agent model andservice repository
repository
Emergency EventsConfigurationTool (EECT)
Open Source DataRegistration Tool (OsdRT)
Intervention MeasuresConfigurationTool (IMCT)
Load
Output Output
Internet
Emergencyresponse plans
Load
∙ Input models of emergencyevents∙ Output settings of agentbehavior in emergent events+ settings of emergent events
∙ Input models of interventionmeasures∙ Output settings of agentbehavior in interventionmeasures + settings ofintervention measures
+ emergency model
∙ Input data from internetnetworks∙ Output population data +emergency events data +intervention measures data
Artificial societypopulation and
geospatialenvironment database
Artificial societyruntime database
OutputThe most optimal
emergencyresponse plan
Actual society
OutputOutputOutput
Input
Load
Load
Load
Load
Output
Artificial SocietySituation Tool (ASST)
Emergency decisionorganization
ClusterRuntime infrastructure
Deploy
Deploy
Experimentsdesign tool (EDT)
Artificial societyinitial data file
Experimentplan
Artificial societymodels
Experiments Management
large scale artificial societyruntime infrastructure
Tool(EMT) +super computerTIANHE-1A
Load
∙ Input experiment plan +artificial society models +artificial society initial data file∙ Output artificial societyruntime state data
∙ Input description of artificialsociety + artificial society populationand geospatial environment society+ emergent events configuration +
intervention measures configuration∙ Output experiment plan + artificialsociety models + artificial societyinitial data file
∙ Input runtime statistical data∙ Output graphics charts ofartificial society statistical data +situation of artificial society
Figure 7 The computational experiments and parallel execution of KD-ACP
point The emergent events and intervention measures areconfigured by EECT and IMCT respectively The configu-rations of emergent events are used to simulate both thereal emergencies like SARS and H1N1 and the supposedemergencies for experiments Similarly the configurationsof intervention measures are also used to reproduce thereal one and simulate the supposed one The repeat of theemergency is used to verify the models while the supposedconfigurations are used to obtain the optimized decision planto the response of the possible emergencies
With the input of artificial society model and servicerepository artificial society population and geospatial envi-ronment database and the configurations discussed beforeEDT generates the experiment plans to meet the require-ments of research on emergency management The outputof EDT includes artificial society models artificial societyinitial data files and experiment plan The models aredownloaded from the repository while the data file is thecollection of data from the database to initialize the modelsWhen the models and data files are ready EMT loads theexperiment plan and deploys the models and data to thecluster or TIANHE-1A supercomputer [30] which was theworldrsquos fastest supercomputer built by National University of
Defense Technology (NUDT) in China in 2010 According tothe plan the experiments are done repeatedly on the largescale artificial society runtime infrastructure [31 32] by themultisample settingsThework process is the implementationof computational experiments in ACP approach
Traditionally emergency response plans are made byemergency management theories and experiences The onlyway to test the effective of plans is the feedback results of realworld ACP approach provides a novel method to supportemergency response plan making by parallel execution Asshown in Figure 7 the work process of KD-ACP is composedof two loops The inner loop composed by red arrowsdescribes the process of computational experiments while theouter loop of yellow arrows illustrates the process of parallelexecution During the runtime of computational experi-ments the statistical data of artificial society is collected andstored in the artificial society runtime database Based on thedatabase ASST outputs the customized situation of runningartificial society by graphics charts and situation mapsThe information is sent to the organizations of emergencydecision to support making the emergency response plansWith the help of computational experiments loop these plansare simulated repeatedly to find the most optimal one
10 Mathematical Problems in Engineering
Open Source Data Registration Tool (OsdRT)
Emergent EventsConfiguration Tool (EECT)
Intervention MeasuresConfiguration Tool (IMCT)
∙ Web information collecting
∙ Deep mining in social media
∙ Web information denoising
∙ Information filtering
Data acquisition
Actual society
Open source informationin social media
Internet
∙ Basic element extracting∙ Individuals and organizationsextraction∙ Sentiment analyzing andopinion mining∙ Social network analyzing
Data extraction
∙ Domain knowledge description∙ Information fusion and collisiondetection∙ Domain knowledge construction∙ Social media informationstandardization
Data standardization
Knowledge
Figure 8 The compositions of OsdRT [34]
Moreover the most optimal plan is used to the responseof emergency in actual society According to the idea ofparallel control in [24] the feedback of actual society ispartly collected from Internet networks by OsdRT As shownin Figure 8 OsdRT is composed of three components dataacquisition data extraction and data standardization Dataacquisition collects mines and filters information fromsocial sensing networks Data extraction includes basicelement extracting individual and organization extractingsentiment analyzing and social networks analyzing Datastandardization specifies the useful knowledge and sent it toconfiguration tools in KD-ACP
By processing in OsdRT the knowledge about emergentevents and intervention measures are analyzed first andregistered in the EECT and IMCT The registration updatesthe settings of emergent events and intervention measuresThis loop composed of yellow arrows implements the parallelexecution in ACP approach The implementation of parallelcontrol and management provides a data-driven approachthat considers both the engineering and social complexityfor modelling analysis and decision making in emergencymanagement
4 Modelling Beijing City with KD-ACP
41 How to Build the Artificial Beijing According to themodelling of artificial society in KD-ACP discussed beforethe Beijing city is modeled as follows
To meet the requirements of emergency managementsthe basic elements of artificial society are extended Asshown in Figure 9 six elements are required to simulate thecity agents environments transportation activity schedulecommunication and agent activity
411 Modelling Artificial Beijing Figure 10 shows the mainGUI of GME for the modelling of artificial modelling inpublic health events Metamodels are listed in the left area ofFigure 10 the list provides basic syntax elements for domainexperts to model artificial society Domain experts build
models based on the knowledge of their own MeanwhileGME supports hierarchy for building large scale systemsThesyntax symbol listed in GUI can be extended in new tab bydouble clicking Take agent for example the model of agentcan be detailed by edition in another tab page of agent
As shown in the center of Figure 10 the models ofartificial Beijing consist of five parts models of agent andenvironment domain models of public health events inter-ventionmodels controllermodels and services Agentmodeldescribes individuals in society it is composed of basicpopulation information action social relationships activityschedule and disease related information Activity schedulerepresents individualrsquos physical actionmodel focusing on thedaily action of agents Environment model includes physicalentities such as buildings playground transportations andagents contained in environment Domain models of publichealth events are composed of the propagation model of dis-ease disease state transition model and so on Interventionmodels include the settings of vaccination isolation and soon The models mentioned before are all FSM models Themechanisms of these models will be detailed in next sectionsController models and services are the public service mod-ules they are implemented in the development in MDT
412 Modelling Agents and Environments Under the speci-fication of FSM agent and environment models are imple-mented in two parts the state space and state transitionsThe state space is composed of the demographic attributesand behavior related attributes The transitions are triggeredwhen the conditions of states are satisfied As shown inFigure 11(a) the action of agent is changed when the ldquonexttimerdquo condition is satisfied in agent model while the agentslist is changed when the agent arrival condition is satisfied inthe environment model
413 Modelling Activities Agent activities come from theagent state transitions of actions such as movements andcommunications The actions of agents are instructed by theactivity schedule shown in Table 2 Activity schedule lists
Mathematical Problems in Engineering 11
Transportation EnvironmentsEnvironments
Agents
Artificial Beijing
Agent activityActivity schedule
Agent activity
Communication by socialrelationship networks
Communication by socialrelationship networks
Figure 9 The basic elements of artificial society
Syntax elements of DSM
Services
Models of agentenvironment andactivity schedule
Domain models of publichealth events
Intervention models
Modelingaspect ofdomainexperts
Controllermodels
Figure 10 The models of artificial Beijing in GME
all the actions with probability in one day for agents inboth normal and emergent situation [38] There are severaltypes of activity schedule in artificial Beijing student agentactivity schedule worker agent activity schedule emergentagent schedule and so on For example Table 2 gives an agentactivity schedule Upon the instruction of activity schedulestudent agent changes the actions by 119901
119894after state transitions
The119901119894in the tablemeans the action probability in the relevant
period In the duration from 0800 to 1200 a student agenteither goes to classroom to have class or goes to libraryto study The probability of class action is 119901
2while the
probability of study action is 1 minus 1199012
Agent behaviors are decided by the settings of activityschedule In addition to the daily activity schedulementionedbefore emergent activity schedules are also considered inour work Take public health events for instance an infectedagent changes schedule from a normal one to emergentone The workflow of a susceptible agent is illustratedin Figure 12 to show the change of behaviors After theinfection the agent is set in incubation phase Not all theincubation agents will become symptomatic Some of themturn back to being susceptible and some of them becomesymptomatic The symptomatic agents change their activityschedule from normal to emergent In the emergent case
12 Mathematical Problems in Engineering
Next actionNext environment (location)
Advance to next actiontime
Find next environment inactivity chain by action
Current actionCurrent environment (location)
Q
q0
120575 Find next action inactivity schedule by time
F
Σ
(a) The state transitions of agent
Q
q0
120575 F
Current agent list
Current pollution level
Next agents list
Next pollution level
Agents entered
Infected agents entered
Add agents to currentagents list
Compute pollution levelby infected agents
Σ
(b) The state transitions of environment
Figure 11 The state transitions of models
Table 2 The activity schedule of a student agent
Duration (Δ119905) Activity Location Probability 119879Location (minute)0000ndash0600 Sleep Dormitory 119875
0(100) 360
0600ndash0800 Breakfast Dormitoryrestaurant 1198751(068) 120
Sports-breakfast-travel Playground-dormitoryrestaurant 1 minus 1198751(032)
0800ndash1200 Class Classroom 1198752(077) 240
Study Library 1 minus 1198752(023)
1200ndash1400 Lunch Restaurant 1198753(090) 120
Lunch-shopping Restaurant-convenience store 1 minus 1198753(010)
1400ndash1800 Class Classroom 1198754(070) 240
Study Library 1 minus 1198754(030)
1800-1900Dinner Dormitoryrestaurant 119875
5(063)
60Sports-shopping Playgroundconvenience store 1198756(025)
Study Classroomlibrary 1 minus (1198755+ 1198756) (012)
1900ndash2200Rest Dormitory 119875
7(045)
180Dinner-rest Dormitoryrestaurant-dormitory 1198758(020)
Study Classroomlibrary 1 minus (1198757+ 1198758) (035)
2200ndash2400 Rest Dormitory 1198759(072) 120
Sleep Dormitory 1 minus 1198759(018)
infected agents go to hospital according to the treatmentschedule or stay in dormitory according to isolation scheduleAfter the treatment in hospital or self-immunoprocess agentsbecome healthy and immune of disease If the agents aretreated in the hospital they are not allowed to get out untilrecovered Moreover activity schedules are also influencedby the emergency response plans For example in the case ofisolation in emergency response plans the activity scheduleof agents who had contacted with infected is changed Onlythe locations including home and dormitory compose thisemergent schedule Likewise based on the statistical datafrom emergency response plans the additional possibilities
are added into activity schedulesThe behaviors of agents alsochange with activity schedules In the view of ACP approachthe injection of new emergency response plans implementsthe parallel execution of microagent behaviors
In another aspect contact frequency is also anothercrucial element to determine the infected rate in public healthevent It is important to model the contact frequency andcontact time of individuals Based on the studies on thecontact behavior of human being by questionnaire surveyEdmunds found that the contact frequency of individualcould be fitted approximately into a normal distributionand the mean and the standard deviation distributions are
Mathematical Problems in Engineering 13
Susceptible agent ImmuneYes
Healthy and immune
No
Choose schedule
Self-isolation Go to hospital
Staysymptomatic
No
Receive treatment
Contact withinfected agents
Symptomatic
No
Self-immunoprocess
Yes
Recovered
Yes
Incubation
Infection
No
Yes No
NoYes
Figure 12 The workflow of a susceptible agent
168 and 85 [39] So it is possible to apply normal randomvariable to model the contact frequency In our work Box-Muller method [40] is used to generate the random variableof contact frequency shown in the following
119865119862= 120583119865+ 120590119865(minus2 ln (120574
1))12 cos (2120587120574
2) (1)
in which 119865119862is the random variable of contact frequency 120583
119865
is the mean value of normal distribution 120590119865is the standard
deviation of normal distribution and 1205741and 120574
2are the
uniform random variables distributed in the interval [0 1]Based on (1) and survey data [41] the contact frequenciesof agent 119865contact(119860 119894) are discretized as in Table 3 within theconsideration of activity differences
Similarly the duration per contact between individualsis another key factor of AHC transmission The duration ofcontact could also bemodeled by the normal randomvariablein the following [42]
119879119862= 120583119879+ 120590119879(minus2 ln (120574
1))12 cos (2120587120574
2) (2)
in which 119879119862is the random variable of duration per contact
120583119879
is the mean value of normal distribution 120590119879
is thestandard deviation of normal distribution and 120574
1and 120574
2
are the uniform random variables distributed in the interval
[0 1] In addition because the time spent in specific location119879Location(119860 119894) also follows the random distribution like (3)in our work In order to make the duration of contact119879contact(119860 119894) shorter than the 119879Location(119860 119894) the mean andstandard deviation of (2) is set up in (3) and (4) [42]
120583119879=119879Location(120583119865+ 120590119865) (3)
120590119879=119879Location(10120583119865+ 10120590
119865) (4)
in which 119879Location is the time of an activity in a specificlocation set in Table 2 120583
119879and 120590
119879are the mean and standard
deviation of contact frequency As a result the duration ofagent 119879contact(119860 119894) in specific locations (listed in Table 2) isalso discretized in Table 3 within the consideration of activitydifferences
414 Modelling Transportation In artificial Beijing threekinds of travel models are considered walking inside thedistrict travelling by road networks and travelling by subwaynetworks According to the traveling models the agentmovements are implemented by the compositions of travelingmodels before So it is important to build the basic road
14 Mathematical Problems in Engineering
Table 3 The contact frequency and duration in different locations and activities by social relationships in case of student and teacher
Social relationship Activity Location 119879contact(119860 119894) (minute) 119865contact(119860 119894)
Student-student
Sleep Dormitory 0 0
Breakfast Restaurant 119873(52) 119873(32)Breakfast Dormitory 119873(52) 119873(21)Sports Playground 119873(21) 119873(53)Class Classroom 119873(21) 119873(21)Study Library 119873(21) 119873(21)Lunch Restaurant 119873(158) 119873(32)
Shopping Convenience Store 119873(203) 119873(21)Dinner Restaurant 119873(108) 119873(21)Dinner Dormitory 119873(158) 119873(32)Rest Dormitory 119873(63) 119873(52)
Teacher-student Class Classroom 119873(21) 119873(21)
Subway network of Beijing city (red lines) Road network of Beijing city (green lines)
Figure 13 The implementation of transportation for artificial Beijing
and subway networks With the help of Google map thekey points of road are sampled first and the road networksare generated by the links among point Similarly subwaynetworks are generated by the links between subway stationsmapped from the real stations in Google map Figure 13shows the road and subway networks in artificial Beijingwith the road networks in green and the subway networks inred Most of the districts in Beijing city are covered by thesetwo networks The algorithm of hierarchical route planningis proposed as follows
(i) Find the next position where agent will be located innext action
(ii) Obtain the travelingmodels from the statistical trans-portation data
(iii) Search the path from the starting position to thenearest road entry or subway stations
(iv) Search the optimal path from the starting station orroad entry to the nearest station or road exit of targetposition
(v) Search the path from the target subway station or roadexit to the target position
(vi) Attach the subway train number or the bus numberincluding the transfer information to the path accord-ing to the timetable of subway or bus
(vii) Connect all the paths obtained before Generate thepath from the starting position to the target position
Based on the algorithm for each agent in artificial Beijingthe transportation is simulated during the computationalexperiments It is worth noting that the activities insidethe buses or train cars are considered because the agentcommunication during travelling is also necessary to besimulated
Mathematical Problems in Engineering 15
(a) The implementation of agent in MDT (b) The code framework of agent
Figure 14 The agent model implementation in MDT [35]
415 The Implementation of Models by MDT All the modelsare developed in the MDT Figure 14 shows the GUI of thetool MDT is used to implement the state space and statetransitions of FSM models The code framework is shownin Figure 14(b) Using the C++ inheritance features and thetechniques of dynamic link library (dll) the models areencapsulated in the dll componentsThe flexible compositionmechanism is designed to support the models evolution incomputational experiments
416 The Description of Artificial Beijing by ASE As men-tioned in Section 34 ASE describes the artificial Beijingin macroview The statistical information of Beijing city iscollected by ASE Table 4 lists part of the description ofartificial Beijing generated by ASE
42 How to Obtain the Data of Beijing City
421 Population and Geospatial Environment Database ofBeijing City Based on the models developed for ArtificialBeijing it is necessary to generate the initial data to do thecomputational experiments Because the individual level dataare not available it is necessary to construct an individual-based population database for both accurate computationalexperiment and determining optimal decisions
According to the state spaces of models discussed inSection 412 the geospatial and population database of Arti-ficial Beijing is designed as shown in Figure 15 The kernelpart of database consists of the tables such as agent list tablegeospatial environment list table household list table andagent distribution table Agent list table is an individual leveltable used to store the data to initialize the state space of agentmodel such as id and gender age Geospatial environmentlist table is also an individual level table which stores thedata of environment attributes such as id street id andtype Household list table and agent distribution table storethe statistical data for the data generation for agent andenvironment models With the help of the geospatial andpopulation database the data generating of artificial Beijingis proposed in the next section
422 Generating Geospatial and Social Environment Datafor Artificial Beijing by PGET The algorithm of generatinggeospatial and social environment data for artificial Beijingis used to quantify the spatial distribution of population andformalize geospatial behavior of each agent The algorithmshown in Figure 16 is capable of generating a syntheticartificial society which allowsmultiresolution statistical datasocial interactive behavior and multilayer social networksto be integrated together The synthetic population canrepresent individual agents in the form of households andhousehold members and the synthetic population is statis-tically equivalent to a real population For each householdcharacteristics such as address household size family typesand relationships are generated Each person is described bycharacteristics such as age gender social role and correlatedlocations The algorithm provides an effective methodologyto reconstruct the computing environment in high resolutionby using statistical data in low resolution leading to betterprediction and management of emergencies The details ofalgorithm are illustrated in [36]
As shown in Figure 17 the implementation of the algo-rithm is embedded inside the GEPTThe collective input datasuch as the data files of population distribution is collected inthe tool first then the algorithm is activated to generate thedatabase The generation lasts almost twelve hours one timewith the specified parameters settings When the generationis finished the data of artificial Beijing is obtained in thedatabase with 19610000 agents and 16000 environments
5 H1N1 Experiments in ArtificialBeijing with KD-ACP
With the support of the artificial Beijing modeled before thecomputational experiments and parallel execution are alsoimplemented by KD-ACP The experiments are designed tofind the most optimal emergency decision response plansEpidemic in city is a typical scenario in emergency manage-ment H1N1 epidemic in Beijing in 2009 is used to be the usedcase to test KD-ACP According to Section 34 it is necessarytomodel emergencymanagement first in order to support thecomputational experiments
16 Mathematical Problems in Engineering
Table 4 The description of artificial Beijing generated by ASE
Statistical features ofartificial Beijing Descriptions of statistical features Data sourcedefault values
Agent roles Baby child primary student middle school student college student worker retired
Demographic data file ofBeijing
Environment types Super market store restaurant park stadium hospital school cliniccommercial building dormitory building residential building
Demographic data file ofBeijing
Demographic data Age distribution of population sex distribution of population householddistribution of population retired age distribution children distribution
Demographic data file ofBeijing
Environment data Distribution of all types of environments Environment distributiondata file of Beijing
Relationships Classmate relative college friendship family Demographic data file ofBeijing
Varying parameters
Age difference of couples119873(120583120590) 119873(03) isin [0 12]Minimummarriage age (male female) (22 20)Age difference of children119873(120583120590) 119873(13) isin [1 10]Enrollment rate of kindergarten 120572 08Enrollment rate of primary school 120573 1Enrollment rate of middle school 120574 1Enrollment rate of college 120575 08Entrance age of kindergarten 1205721015840 3Entrance age of primary school 1205731015840 7Entrance age of middle school 1205741015840 13Entrance age of middle school 1205751015840 19
Age distributioninformation table
Agent settable
Geospatialenvironment set table
Household settable
Figure 15 The geospatial and population database of Beijing city
51 Modelling Emergency Management The modelling ofemergency management consists of two parts the modellingof emergent events and the modelling of intervention mea-sures H1N1 and the intervention measures for H1N1 aremodeled in our case
511 Modelling H1N1 by MDT H1N1 Model in agent simu-lates the states transitions of health status and the relevantactions Referring to SIR (SusceptibleInfectiveRecovered)[43 44] models agent has three health statuses susceptibleinfected and healthy with immunity Only susceptible agentcan be infected by the contact with infected agents After
the three infected phases (incubation being symptomaticand recovery) of H1N1 model agent is healthy again It isworthy mentioning that the recovered agents can be infectediteratively when the immunity disappears gradually by timeThe state transitions are illustrated in Figure 18
According to [45] the latent period of H1N1 was in aWeibull distribution within people from one to seven days[46 47] The distribution was usually centered in the rangebetween one and three days AWeibull random variable [40]is used to model the latent period as described in
119879Lat = 120572 times [minus ln (1 minus 120574)]1120573+ 120592 (119879Lat ge 120592) (5)
Mathematical Problems in Engineering 17
Censusdata
Generate GUID
Assign individualhousehold
Generate familystructure
Assign family roleto every member
Endow individualwith age gender
social role
Distribute environmentlocations on the map
School Workplace Entertainmentlocations
Consumptionlocations
Assign school orworkplace to
individual
Correlatedindividual withother locations
Generate friendshipin school
Generate coworkerrelations inworkplace
Assign individualadministrative
region
Generate housebuildings
Locate householdto house building
Generate familyrelation
Generateneighborhood
relation
PopulationEnvironment
Environment listSocial network
Figure 16 The algorithm of generating geospatial and social environment data [36]
Edit
Edit
Edit
Edit
Edit
Edit
BrowseStatistical characteristic of population
Geospatial population data
Population data of street
Population types
Geospatial and social environment generation tool
Workspaces
Parameters distribution settings
Household types
Roles in household
Household structure
Social roles
BrowseBrowseBrowseBrowseBrowseBrowse
Population distribution data
File path
Retired distributionPop distributionHousehold numHousehold distributionSex distributionAge distributionData type
Figure 17 The GUI of Population and Geospatial Environment generation Tool (PGET)
18 Mathematical Problems in Engineering
Infected
SusceptibleHealth
withimmunity
Infected
by contac
t
with in
fected
agents
Recovery
Increase before beingsymptomatic symptomatic
Decrease after being
Decrease by time
Infectivity
Immunity
Figure 18 The transitions of agent health status
119879Lat denotes the duration of latent period 120592 120572 and 120573 arethe location parameter scale parameter and shape parameterof Weibull distribution respectively 120574 is a uniform randomnumber in the range [0 1] According to statistics [47] 120592 120572and 120573 are set as 0 18 and 121 respectively Then the mean(standard deviation) of latent period is calculated as 159 days(058 day2) Furthermore infectious period is set as 7 days
A susceptible studentrsquos probability of getting infecteddepended on the infectivity of infectious agents his ownimmunity level the duration of the contact action and soon The infectivity of an infected agent evolved with timeduring the infectious period According to the statistics ofinfectivity in the chart with red bars in Figure 19 an infectedstudent had the highest level of infectivity in the second dayafter he had the first symptom The infectivity levels in otherdays in comparison with the second day are in Table 5 Theday labeled as ldquominus1rdquo means the day before the starting of thesymptomatic period
Based on the algorithm discussed before theH1N1modelis developed with the help of MDT The settings of influenzamodel will be discussed in Section 521
512 Modelling Intervention Measures by MDT In corre-spondence intervention measures model is used to ceasethe outbreak of H1N1 As discussed in [42 48 49] if theappropriate measures are taken when the infectious diseaseemerges the transmission of the disease could be sloweddown and the damage could be decreased Therefore it isnecessary and important to design emergency interventionmeasures in the artificial Beijing According to the MinistryofHealth inChina the intervention policies [50] are designedto control the spread of influenza such as H1N1 The inter-ventionmeasures of these policies including are interventionsactivated time vaccination rate antibiotic delay to hospitalisolation duration close workspace duration and limitationof activities listed in Table 6 Similarly the models of theseintervention measures are also developed with the help ofMDT Using the reasonable data ranges listed in Table 6intervention models can be initialized to do the computa-tional experiments These experiments are used to show how
Symptomatic1 2 3 4 5 6 7minus1
Recovered
Tim
e (da
y)
Infe
ctiv
ity
Figure 19 The distribution of infectivity rate of an infected agent[37]
Table 5 The proportion of infectivity
Days (119894) minus1 1 2 3 4 5 6 6Proportion of infectivity 03 05 1 08 04 02 01 005
to restrain the H1N1 transmissionThe settings of these inter-vention measures in IMCT are introduced in Section 522
52 Experiment Settings
521 H1N1 Settings by EECT EECT is designed to initializethe emergent events models such as H1N1The settings comefrom the specified emergent event models As shown inFigure 20 in theH1N1 used case the settings are composed ofthe distribution of infection source infection rate infectionperiod and so on The infection rate changes with envi-ronments and infection periods The settings are initializedfrom both the research of medicine on H1N1 and the data ofH1N1 outbreak in Beijing in 2009The process of influenza inartificial Beijing is simulated to repeat the possible scenariosof H1N1 break During the computational experiments ofartificial Beijing theH1N1 settings such as the period of beingsymptomatic are updated by the dynamic data registrationsfrom OsdRT The updating corrects the H1N1 model duringthe experiments With the help of OsdRT artificial Beijing isable to approach actual society It is also an implementationof parallel execution
522 Intervention Measures Settings by IMCT In the H1N1case the settings of intervention measures are shown inFigure 21 in IMCT These measures are designed under theinstruction of Emergency Decision Organizations like ChinaDisease Center (CDC) The measures such as vaccinationrate the rate of closedworkspace and the starting time can beinitialized in the tool Emergency response plans are imple-mented in these settings As illustrated in Figure 7 all the pos-sible compositions within value settings in reasonable inter-vals are tested to find the most optimal emergency responseplan For example different compositions of measures suchas vaccination rates and rates of workspace closed are usedto design the plans of experiments By the analysis of theresults the best composition of interventionmeasures can beobtainedThe analysis will be discussed in case study in detail
Mathematical Problems in Engineering 19
Table 6 The intervention measures of influenza
Intervention measures ValuesInterventions activated Time 119879Intervention isin [0 80th day]Vaccination rate 119877Vaccination isin [0 1]Antibiotic rate 119877Antibiotic isin [0 1]Delay to hospital 119867ospital(119860 119894) 119879infected lt 119905 lt 119879recovered + 119879HDelay 119879HDelay isin (0 3th day]Isolation duration 119868solation(119860 119894) 119879Intervention lt 119905 lt 119879Intervention + 119879IDuration 119879IDuration isin (0 14th day]Close workspace duration 119862lose(119864119894) 119879Intervention lt 119905 lt 119879Intervention + 119879CWDuration 119864119894 isinWorkspace 119879CWDuration isin (0 7th day]Limitation of agent activity duration 119871 imitation(119860 119894) 119879Intervention lt 119905 lt 119879Intervention + 119879LAADuration 119879LAADuration isin (0 7th day]119877Vaccination means the vaccination rate of artificial Beijing while 119877Antibiotic means the antibiotic rate used for infected agents in hospital119867ospital (119860119894) means that the infected agent 119860119894 is sent to hospital he or she will receive the retreatment immediately119868solation (119860119894) means that the infected agent 119860119894 is isolated at home he or she can only contact with the other agents until isolation duration is over119862lose (119864119894) means that the environment 119864119894 typed of workspace is closed until the close workspace duration is over119871 imitation (119860119894) means that the activities of agent 119860119894 are limited only the workspace and home are allowed until limitation duration is over
Incubation Weibulldistribution
Symbolic LogNormaldistribution
Recovering LogNormaldistribution
H1N1influenza
Settings of H1N1 influenza model
Disease model
Transmissionmodel
Figure 20 The settings of H1N1 in EECT
It is worth notifying that the settings of interventionmeasures are also influenced by OsdRT For example theH1N1 outbreak areas are notified by OsdRTThe interventionmeasures such as close workspace are used only in theoutbreak areas Therefore the unnecessary computation ofintervention measures is avoided So the data injection byOsdRT not only makes artificial society approach actualsociety but also increases the performance of computationalexperiments
53 Experiments by EMT When the initialization of arti-ficial Beijing is ready EMT is used to do the computa-tional experiments EMT is composed of two parts thecentral controller in the central node and the residen-tial service deployed in the computational nodes Centralcontroller loads the experiment plans and controls the whole
experiments process the GUI of controller is shown inFigure 22 EMT manages the computational experimentsin four steps cluster configuration load experiment plansmodels deployment and running experiments
531 Cluster Configuration Before the configuration clusterinformation is collected from the residential services insidethe cluster The information includes nodes amount nodenames node IPs and CPU occupations It is used to quantifythe computing power of the clusterThe cluster configurationis based on this information The nodes are selected forexperiments Then the LP (local process) [51] number foreach node is set according to the number of the CPU kernelsAfter the configuration the experiment plans are loaded tocustomize the models and data
20 Mathematical Problems in Engineering
Streets of close workspaces
Limitations ofactivities
IsolationsRespirator
Closeworkspace
Antibiotic
VaccinationHospital
Surveillance
Figure 21 The settings of intervention measures in IMCT
Information ofexperiment plan
List of experiment plans
Start | Pause | Reset | Stop | Speedup | Slowdown | Back
NodeName | CPU occupation | Memory occupation |Network | Models | Status
Figure 22 The deployment and running of experiments in EMT
532 Load Experiment Plan As discussed in Section 35 thedescription of experiment plans is composed of the infor-mation in three aspects models and the initial data for theexperiments the deployment mapping tables from the mod-els to the nodes and the plan of experiment executionWhenthe experiment plan is loaded by the EMT the file namesand model descriptions of models and data are listed Withthe support of cluster configuration models can be mappedto the nodes in the GUI according to the computationrequirements EMT also provides the setting of experimentsitself including the running times start time and end time
533 Models and Data Deployment Models are stored inthe agent population and services repository According tothe settings in experiment plan the customized models aredownloaded from the repository first then the models suchas agent model environment model emergent events modeland intervention measures model are integrated to be thecompositionmodels of artificial BeijingThemodels and datalocations in the nodes are configured in the ldquomodel pathrdquoand ldquodata pathrdquo in the GUI of central controller Finallythe models are uploaded to the nodes in cluster with theconsideration of partition The partition of artificial society
Mathematical Problems in Engineering 21
models is different from the multiagent system in [52]Models in artificial society are divided into three levels agentsand environments grids and city Agents and environmentsare encapsulated in grids which compose the whole cityThus the partition of artificial society becomes the partitionof grids The parallelism degrees are found to support thepartition of models agent independence and grid indepen-dence According to the parallelisms agents are partitioned toLPs according to the population distribution in grids Agentmovements among grids are simulated by the movementevents among grids in different LPs As a result a two-tierparallel architecture is proposed to support the simulation ofartificial society The architecture is detailed in [53]
534 Running of Experiments When the preparation isready the experiments can be done by the central controllerof EMT The running and control commands of the experi-ments are grouped into three aspects start pause or stop theexperiments get the running information of the nodes suchas the snapshot of the desktop of node With the help of thesecommands users can run the computational experimentsto do the research on the emergency management It isworth notifying that the simulation engine is optimized inorder to support the ten million agent simulation UsingGPU as coprocessor a two-tier parallel simulation engine isdesigned with support of MPI and OpenCL through phasedsynchronization [54] One-sided communication is used forreflection of remote simulation objects and message passingbetween processes A general kernel function prototype iselaborately designed and conditionally compiled for execu-tion on both CPU and GPU Moreover the optimizationoperations like load balance is developed under the instruc-tion of activity-based simulation [55] The densely populatedgrids are given more CPU and memory resource accordingto the activity predictions
54 Results of H1N1 Experiments with KD-ACP On thebasis of artificial society models and the intervention mea-sures mentioned in Section 512 a series of computationalexperiments are performed to study the H1N1 influenza inBeijing In our case of artificial Beijing 19610000 agentsare simulated in the cluster within 48 CPU cores and 128Gmemories the nodes are connected by kilomega networks Ittakes 18 hours to simulate a 250-day disease spreading
In order to illustrate how the optimal plans are obtainedbyKD-ACP the experiments are divided into four groups theexperiments ofmodel validation the experiments of sensitiv-ity analysis of vaccination rate the experiments of sensitivityanalysis of isolation and the experiments of the combinationof interventionmeasuresThese four groups show the generalprocess of the research on emergency management in H1N1influenza Firstly the models of H1N1 are validated withthe support of historical data Secondly the traditionalmedical interventions of pandemic such as vaccination areanalyzed to find the optimal plan Thirdly the traditionalnonmedical interventions of pandemic such as isolation areanalyzed to find the optimal plan Finally the combination ofinterventions is tested to show the combined effect
It is worth mentioning that all the experiments in ourwork are performed 100 times These experiments are ini-tialized with different settings of interventions in IMCT Theresults shown in the figures are the mean values
541 The Experiments of Model Validation In our artifi-cial Beijing the models of H1N1 influenza are built basedon the existing researches Wang and her colleagues hadstudied the H1N1 influenza by SEIR (SusceptibleExposedInfectiveRecovered) models [56] With the support of theirwork we build the models of artificial Beijing in the mannerof multiagents
In Figure 23 the control group shown in red line isdrawn based on the realistic statistical data collected in H1N1influenza in Beijing in 2009 The influenza lasts more thansix months and more than 170000 people are infectedComparing with the control group the simulated resultswithout interventions are drawn in blue It is obvious that thesimulated case fits the control group well before 70 days Butthe ascending of control group is slowed after 70 days andthe peak value (17883) is much less than the simulated results(25135) It is because the vaccination is activated at day 67 incontrol group according to the intervention measures usedin 2009 The restrain by vaccination becomes effective eightdays later The change of infected number after vaccinationis shown by red spotted line The trend of control group isdifferent from the simulation data after 75 days
However the simulated results (blue line) in compu-tational experiments fit the control group (as shown inFigure 23) in the first 70 days when no intervention mea-sures are executed The key features such as effective basicreproduction number (119877
0) are also almost the same in both
cases Therefore it can be concluded that the models ofartificial society built in microview are validated by therealistic statistical data in influenza in Beijing in 2009
In this group of experiments no interventions are exe-cuted in order to validate the models Therefore the simu-lated results of this extreme case which cannot happen todaycould be obtained As shown by blue line the number ofinfected agents without interventions reaches peak value atday 91 with 25135 consisting of 8052 in symptomatic phaseand 28950 in recovery phase It is obvious that the totalnumber of infected agents grows slowly at the beginningtime Then the number increases quickly when the influenzabreaks and it reaches its maximum value with 383870 in theend This phenomenon can be explained as in the followingAt the beginning of influenza due to the limited numberof infected agents the propagation of the disease remainsslow but many agents become susceptible When the numberof infected agents increases rapidly the spreading becomessignificant However the number of infected agents decreasesafter day 91 It is because the disease spreading relies on thesocial networks and spatial contact networks [49] It meansthat the infected agents can only infect agents either in thesocial relationship networks or in the spatially contacted
542 The Experiments of Sensitivity Analysis of VaccinationRate According to Section 512 vaccination is a typical
22 Mathematical Problems in Engineering
0
Infe
cted
agen
ts
0 50 100 150 200 250
Statistical dataSimulation data
25000
20000
15000
10000
5000
Day d
Figure 23 The results with no intervention executed and the com-parison with realistic statistical data in Beijing in 2009
medical intervention in influenza Traditionally the effective-ness of vaccination can only be evaluated after the influenzaand the key settings such as vaccination rate are determinedby the experience It is always difficult for emergency man-agement organizations to find the most optimal vaccinationrate in advanceTherefore the sensitivity of vaccination rate isanalyzed in this group of experiments The settings are listedin Table 7 average results are shown in Figure 24
Vaccination rates are set linearly (10 30 50 and70) to test the changes of influenza The peak value ofinfected agents decreases from 25135 22392 15787 6907to 638 respectively It shows that the more the agents arevaccinated the smaller the peak value of the infected agentswill be In the case of 70 vaccination the total infected agentis only 13799 It means that the influenza in this case does nothappen due to the high vaccination rate
The change of peak values also shows an interestingphenomenon Vaccination rates increase linearly the peakvalues decrease slower than the ascending of vaccination ratesat beginning but the descending becomes more and morefaster with the linear increase of vaccination rate It meansthat the small scale of vaccination (less than 30) does notmake senseWith the increase of vaccination rate the numberof susceptible agents decreases significantly In sequencethe possibility of infections is lowered greatly Additionallyit could be found that the decrease of peak values comeslater than the increase of the vaccination rates This is quitereasonable according to Mei et alrsquos work in [57]
As a result it is concluded that high vaccination rate iseffective in the control of H1N1 influenza though costly
543 The Experiments of Sensitivity Analysis of IsolationAccording to Section 512 isolation is a typical nonmedicalintervention It is the most used intervention measure in
Table 7 The initial settings of experiments of vaccination
Parameters ValueAgent count 19610000Initial infected count 40Interventions activated time(119879Intervention)
0
Vaccination rate (119877Vaccination) 0 10 30 50 70
0
Infe
cted
agen
ts
Loading intervention of vaccination
No vaccination10 vaccination30 vaccination
50 vaccination70 vaccination
0 50 100 150 200 250
25000
20000
15000
10000
5000
Day d
Figure 24 The results of loading intervention of vaccination
the emergency management in influenza As mentionedbefore a problem also exists that the measure cannot bevalidated before the intervention Therefore the sensitivityof isolation is analyzed in this group of experiments It isworth mentioning that the isolation is usually combinedwith the hospital policy It means the infected individual issent to hospital meanwhile the people who contact withthe infected are all isolated As a result parameters delay tohospital and isolation duration are designed together in thesettings of experiment Delay to hospital stands for the periodfrom infection time to hospital time while Isolation durationrepresents the period when an agent cannot contact others ifhe has contact with an infected agent The settings of theseparameters are listed in Table 8 average results are shown inFigure 25
Firstly the parameter delay to hospital is simulated aloneto test the sensitivity The simulated results are shown insolid line in Figure 25 The peak value decreases from 6708to 2218 when the delay to hospital decreases from 15 daysto 1 day The half day change brings the 67 reductionMoreover the peak value almost approximates to zero inthe case of 05 days The influenza does not happen in thiscase It can be concluded that the earlier the infected agents
Mathematical Problems in Engineering 23
Table 8 The initial settings of experiments of isolation
Parameters ValueAgent count 19610000Initial infected count 40Interventions activated time (119879Intervention) 80Delay to hospital (119879HDelay) isolation duration (119879IDuration) (15 0) (1 0) (05 0) (15 3) (1 3) (05 3)
0 50 100 150 200 2500
1000
2000
3000
4000
5000
6000
7000
Infe
cted
agen
ts
Loading intervention of isolation
Delay to hospital = 15
Delay to hospital = 10
Delay to hospital = 05
Delay to hospital = 15 isolation duration = 3
Delay to hospital = 10 isolation duration = 3
Delay to hospital = 05 isolation duration = 3
Day d
Figure 25 The results of loading intervention of isolation
are sent to hospital the less the agents will be infectedSo the construction of effective mechanism to find infectedindividuals is important in the emergency management ofinfluenza
Secondly the parameter isolation duration is added tosimulate the isolation of interventionObviously the isolationmeasure decreases the infected agents greatly Compared tothe parameter delay to hospital of 15-day case the peak valuedecreases from 6708 to 362when the isolation is added in theinterventionThe reduction of 6346 agents comes from the 3-day isolation of agents who had contacted with the infectedagents However the decrease of parameter delay to hospitalcannot bring the similar reduction in the 3-day isolationcases According to the figure the peak value decreases from362 to 51 from 15-day case to 05-day case only 311 agents aresaved from infection As a result the sensitivity of simulatedresults does not change obviously with delay to hospital inthe case of isolation It can be concluded that isolation is amore effective measure Though it is not easy for emergencymanagement organizations to decrease the delay to hospitaldue to the current monitoring mechanism the influenza canstill be controlled by isolation
544 The Experiments of Combination of Intervention Mea-sures Isolation is not the only nonmedical intervention ininfluenza close workspace is another common measureWhen close workspace is activated the workspaces are allclosed agents can only stay home at the closed time Luckilyisolation and close workspace are independent from oneanother It is possible for emergency management organi-zations to use these two measures together So this groupof experiments is designed to obtain the effectiveness of thecombination of intervention measures The parameters areinitialized with different settings listed in Table 9
Parameters delay to hospital and isolation duration is setin the middle values (1 day 3 days) according to the lastsection According to the experience close duration is set asseven days It is known that close workspace is a really costintervention measure the activated time is a key parameterWith the fixed parameters mentioned before the changeof interventions activated time is analyzed in this group ofcomputational experimentsThe average results are shown inFigure 26
Compared with the simulated results of experimentsdiscussed above Figure 26 strongly illustrates that influenzais ceased when the intervention measures are combinedtogether The peak value of infected agents is 141 in the ldquo80thdays of close-workspace and isolation with 1 day delay tohospitalrdquo case while 25135 in ldquono interventionrdquo case 2218 inldquo1 day delay to hospitalrdquo case 203 in ldquo1 day delay to hospitaland isolationrdquo case The interventions bring a great reductionof infection the peak values decrease from ten thousandslevel to hundred level
On the other hand the peak value even decreases to 108when the execution time of close workspace is 10 days earlierThe date of maximum value of infected agents is advancedfrom the 135th day to the 108th day The whole circle of H1N1influenza is reducedThe results of this group of experimentsshow that intervention measures should be activated as soonas possible and the compositions of interventions are moreeffective in emergency management of H1N1 influenza
In summary based on the general process lots of exper-iments could be done by KD-ACP to study emergency man-agement in H1N1 influenza Many conclusions are obtainedbased on the analysis of the experiments in this sectionAccording to these conclusions themost optimal response inour case study is the case of ldquo70th days of close-workspace andisolation with 1 day delay to hospitalrdquo It is because this casehas the fewest infected agents relatively But the modelling ofH1N1 influenza is not sufficient Economic and social factorsof interventions such as the money cost by vaccination andisolation the influence of the close-workspace in society
24 Mathematical Problems in Engineering
Table 9 The initial settings of experiments of isolation and closeworkspace
Parameters ValueAgent count 19610000Initial infected count 40Interventions activated time (119879Intervention) 70th day 80th dayDelay to hospital (119879HDelay) 1 dayIsolation duration (119879IDuration) 3 daysClose workspace duration (119879CWDuration) 7 days
50 100 150 200 2500
50
100
150
200
Infe
cted
agen
ts
Loading intervention of isolation and close workspace
Delay to hospital = 1 isolation duration = 3Start close time = 70 close duration time = 7Delay to hospital = 1 isolation duration = 3Start close time = 80 close duration time = 7
Day d
Figure 26The results of loading interventions of isolation and closeworkspace
and even opinion transmission of H1N1 in internet are notconsidered in our case study Without the analysis of thesefactors it does not make sense to talk about the most optimalintervention measures However KD-ACP had provided asoftware framework and general process of the research onemergency management Supported by KD-ACP the expertsof emergency management are able to study emergent eventssuch as H1N1 influenza as detailed as possible Based on theexperiments and analysis again and again the most optimalintervention measures will be found ultimately
55 Parallel Execution by ASST and OsdRT As discussed inSection 35 ASST andOsdRT are used to support the parallelexecution of artificial Beijing The statistical data is collectedand stored in artificial society runtime database at runtimeASST downloads the infection data from the database anddisplays it in the map As shown in Figure 27 the infectedagents are distributed in the districts in BeijingThe snapshotgives the geospatial distribution of the H1N1 epidemic Thelegends of the agents in different status are listed in the left
right corner in Figure 27 Five statuses are shown in the figurehealth without vaccination health with vaccination incuba-tion being symptomatic and convalescence ASST providesa very direct feeling for the emergency experts Partly withthe situation from ASST Emergency Decision Organizationmakes the emergency response plans to the response of theemergency scenario With the help of plans the epidemicin actual society can be restrained In the meantime theinfluenza information is reflected in the Internet Accordingto Figure 8 the data acquisition data extraction and datastandardization of H1N1 influenza are processed in OsdRTOsdRT is implemented by Dr Caorsquos team from Institute ofAutomation from Chinese Academy of Sciences [58] Thetool provides not only the reports of H1N1 SARS hand-foot-mouth disease and intestinal diseases are all monitored Theweb page is shown in Figure 28 H1N1 reports from Internetare collected first Based on the data extraction of the reportsH1N1 case of distribution is obtained and shown in theincidencemap As a result H1N1 incidence data is regarded asthe knowledge which is sent to EECT and IMCTThe settingsof emergent events and intervention measures are modifiedin order to make artificial society approach actual society innext turn of computational experiments With the help ofOsdRT the loop of parallel execution process integrates ACPapproach
6 Conclusion
This work provides an integrated software framework forsocial Computing in emergencymanagement From a systemperspective KD-ACP provides a reliable flexible low costand effective platform for the scientists to do the research onemergency response problems The analysis and predictionof these problems with inherent complexity can be solvedby the repetition of possible alternative experiments onthe software framework Therefore KD-ACP which is anattempt to implement ACP approach seemed as an effectivecomputational framework to support the decision making inemergency response
Currently KD-ACP is used to study the H1N1 epidemicin Beijing in 2009 The most optimal compositions of inter-vention measures are testified to support the response tonext influenza However it is still a long way ahead beforethe automation systematization and practicalization of KD-ACP A lot of work will be carried out along several directionsas in the following
(i) The parallel execution process in KD-ACP is actuallynot strongly connected with the emergency responseorganization Traditional experience and theory stillplay a dominant role in the decision of emergencymanagement It is necessary to propose a mechanismof hall for workshop to facilitate the parallel executionloop in actual society
(ii) Section 33 introduces the principle and process ofmodelling artificial society Metamodels and modelsof artificial Beijing are built by GME Model transfor-mation and code generation are used to implementcodes from FSM based models But many details
Mathematical Problems in Engineering 25
Fangshan district Daxing district Fengtai district Xicheng district
Mentougou district Haidian districtShijingshan district Dongcheng district
Incubation
Symptomatic
Convalescence
Health withoutvaccination
Health withvaccinationLegend of agent
health status
Figure 27 The situation of artificial Beijing by ASST
H1N1
Districts of Beijing
H1N1 case ofdistribution
Incidence map
Distribution ofH1N1 reportsfrom internet
Incidence of H1N1
Fangshan districtLiangxiang town
Population 19000Incidence 135789
(1100000)
Figure 28 The implementation of OsdRT by Institute of Automation from Chinese Academy of Sciences
are still not illustrated due to the topic of thispaper The key techniques such as template of codeframework and model transformations from specificdomains to FSM will be focused and introduced indetail in our next paper
(iii) The performance of computational experiments inKD-ACP is still slower than expectation The opti-mization of simulation engine and models structureis necessary to improve the performance So it ispossible to run artificial society with the real worldin parallel
In summary KD-ACP paves a new way for scientists inboth emergencymanagement and computation simulation to
collaborate with each other in solving the emergency man-agement problems by social computing
Research Highlights
(i) KD-ACP is implemented based on ACP approach
(ii) Artificial Beijing is built with the help of KD-ACP
(iii) H1N1 is modeled to simulate the influenza in Beijingin 2009
(iv) Computational experiments testify the effectivenessof intervention measures
26 Mathematical Problems in Engineering
Conflict of Interests
The authors declare that there is no conflict of interestsregarding the publication of this paper
Acknowledgment
This work was supported in part by the National ScienceFoundation of China (nos 91024030 71303252 61403402and 91324013)
References
[1] W Mao and F-Y Wang Advances in Intelligence and SecurityInformatics Academic Press Oxford UK 2012
[2] E Bonabeau ldquoAgent-based modeling methods and techniquesfor simulating human systemsrdquo Proceedings of the NationalAcademy of Sciences of the United States of America vol 99 no3 pp 7280ndash7287 2002
[3] K M Carley D B Fridsma E Casman et al ldquoBioWar scalableagent-basedmodel of bioattacksrdquo IEEETransactions on SystemsMan and Cybernetics A Systems and Humans vol 36 no 2 pp252ndash265 2006
[4] J M Epstein ldquoModelling to contain pandemicsrdquo Nature vol460 no 7256 p 687 2009
[5] Los Alamos National Laboratory Reports about EPISIMS LosAlamos National Laboratory 2003
[6] C H Builder and S C Bankes ldquoArtificial societies a conceptfor basic research on the societal impacts of informationtechnologyrdquo RAND Report 149 RAND Corp 1991
[7] F-Y Wang ldquoComputational theory and method on complexsystemrdquo China Basic Science vol 6 no 5 pp 3ndash10 2004
[8] F-Y Wang ldquoToward a paradigm shift in social computing theACP approachrdquo IEEE Intelligent Systems vol 22 no 5 pp 65ndash67 2007
[9] F-Y Wang and S Tang ldquoArtificial societies for integratedand sustainable development of metropolitan systemsrdquo IEEEIntelligent Systems vol 19 no 4 pp 82ndash87 2004
[10] D Wen Y Yuan and X-R Li ldquoArtificial societies computa-tional experiments and parallel systems an investigation ona computational theory for complex socioeconomic systemsrdquoIEEE Transactions on Services Computing vol 6 no 2 pp 177ndash185 2013
[11] F Wang F Zeng and Y Yuan ldquoAn ACP-based approach forcomplexity analysis of E-commerce systemrdquo Complex Systemsand Complexity Science vol 5 no 3 pp 1ndash8 2008 (Chinese)
[12] F-Y Wang and P K Wong ldquoResearch commentary intelligentsystems and technology for integrative and predictivemedicinean ACP approachrdquoACMTransactions on Intelligent Systems andTechnology vol 4 no 2 pp 1ndash6 2013
[13] J Sifeng X Gang F Dong and H Chunpeng ldquoStudy on theemergency rescue decision support system of petrochemicalplant based on ACP theoryrdquo in Proceedings of the 6th Manage-ment Annual Meeting (MAM rsquo11) 2011 (Chinese)
[14] X Dong D Fan G Xiong F Zhu Z Zhang and Y Yao ldquoPar-allel Bus Rapid Transit (BRT) operation management systembased on ACP approachrdquo in Proceedings of the 6th ManagementAnnual Meeting (MAM rsquo11) Beijing China 2011
[15] B Ning F-Y Wang H-R Dong R-M Li D Wen and L LildquoParallel systems for Urban rail transportation based on ACP
approachrdquo Journal of Transportation Systems Engineering andInformation Technology vol 10 no 6 pp 23ndash28 2010 (Chinese)
[16] F-Y Wang J Zhao and S-X Lun ldquoArtificial power systemsfor the operation and management of complex power gridsrdquoSouthern Power System Technology vol 2 no 3 pp 1ndash11 2008(Chinese)
[17] M J North T R Howe N T Collier and J R Vos ldquoRepastsimphony development environmentrdquo in Proceedings of theAgent 2005 Conference on Generative Social Processes Modelsand Mechanisms 2005
[18] M J North T RHowe N T Collier and J R Vos ldquoRepast Sim-phony runtime systemrdquo in Proceedings of the Agent Conferenceon Generative Social Processes Models and Mechanisms 2005
[19] SWARM Development Group Swarm 211 Reference Guide2000 httpwwwswarmorg
[20] S Tisue and U Wilensky ldquoNetLogo a simple environment formodeling complexityrdquo in International Conference on ComplexSystems pp 16ndash21 2004
[21] S Luke C Cioffi-Revilla L Panait K Sullivan and G BalanldquoMASON a multiagent simulation environmentrdquo Simulationvol 81 no 7 pp 517ndash527 2005
[22] M Ichikawa H Tanuma Y Koyama andH Deguchi ldquoSOARSintroduction as a social microscope for simulations of socialinteractions and gamingrdquo in Organizing and Learning throughGaming and Simulation Proceedings of the 38th Conference ofthe International Simulation and Gaming Association (ISAGArsquo07) pp 149ndash158 2007
[23] D S Burke J M Epstein D A T Cummings et al ldquoIndividual-based computational modeling of smallpox epidemic controlstrategiesrdquo Academic Emergency Medicine vol 13 no 11 pp1142ndash1149 2006
[24] F-Y Wang ldquoParallel control and management for intelligenttransportation systems concepts architectures and applica-tionsrdquo IEEE Transactions on Intelligent Transportation Systemsvol 11 no 3 pp 630ndash638 2010
[25] Z Molnar D Balasubramanian and A Ledeczi ldquoAn introduc-tion to the genericmodeling environmentrdquo in Proceedings of theTOOLS Europe 2007 Workshop on Model-Driven DevelopmentTool Implementers Forum Zurich Switzerland 2007
[26] S Kelly and J-P Tolvanen Domain Framework in Domain-Specific Modeling Enabling Full Code Generation JohnWiley ampSons Hoboken NJ USA 2007
[27] B P Zeigler H Praehofer and T G Kim Theory of Modelingand Simulation Integrating Discrete Event and ContinuousComplex Dynamic Systems Academic Press San Diego CalifUSA 2000
[28] G Booch J Rumbaugh and I JacobsonThe Unified ModelingLanguage User Guide Addison-Wesley Reading Mass USA1999
[29] F Wagner R Schmuki T Wagne and P Wolstenholme Mod-eling Software with Finite State Machines A Practical ApproachCRC amp Taylor amp Francis Boca Raton Fla USA 2006
[30] T Espiner ldquoChina builds worldrsquos fastest supercomputerrdquo ZDNetUK October 2010
[31] B Chen and G Guo ldquoA two-tier parallel architecture for arti-ficial society simulationrdquo in Proceedings of the ACMIEEESCS26th Workshop on Principles of Advanced and DistributedSimulation (PADS rsquo12) pp 184ndash186 July 2012
[32] G Guo B Chen X G Qiu and Z Li ldquoParallel simulation oflarge-scale artificial society on CPUGPU mixed architecturerdquo
Mathematical Problems in Engineering 27
in Proceedings of the ACMIEEESCS 26th Workshop on Prin-ciples of Advanced and Distributed Simulation (PADS rsquo12) pp174ndash177 July 2012
[33] S-X Lun ldquoResearch on the classification of parallel executionmodes of ACP theoryrdquo Acta Automatica Sinica vol 38 no 10pp 1602ndash1608 2012
[34] F-YWang ldquoA framework for social signal processing and anal-ysis from social sensing networks to computational dialecticalanalyticsrdquo Science China vol 43 no 12 pp 1598ndash1611 2013
[35] G GuoOne Model Userrsquos Manual version 12 National Univer-sity of Defense University 2011
[36] Y Ge R Meng Z Cao X Qiu and K Huang ldquoVirtual citymdashanindividual-based digital environment for human mobility andinteractive behaviorrdquo Simulation Transactions of the Society forthe Modeling and Simulation International 2014
[37] M Kretzschmar and R T Mikolajczyk ldquoContact profiles ineight European countries and implications for modelling thespread of airborne infectious diseasesrdquo PLoS ONE vol 4 no6 Article ID e5931 2009
[38] W Duan and X Qiu ldquoFostering artificial societies using sociallearning and social control in parallel emergency managementsystemsrdquo Frontiers in Computer Science vol 6 no 5 pp 604ndash610 2012
[39] W J Edmunds C J OrsquoCallaghan and D J Nokes ldquoWhomixes with whom Amethod to determine the contact patternsof adults that may lead to the spread of airborne infectionsrdquoProceedings of the Royal Society B Biological Sciences vol 264no 1384 pp 949ndash957 1997
[40] J Banks J S Carson B L Neison and D M Nicol Discrete-Event System Simulation Prentice Hall Englewood Cliffs NJUSA 4th edition 2007
[41] H Yongxia ldquoInvestigation and analysis of college studentsrsquo workand restrdquo Value Engineering vol 31 no 3 2012 (Chinese)
[42] W Duan Z Cao Y Wang et al ldquoAn ACP approach to publichealth emergency management using a campus outbreak ofh1n1 influenza as a case studyrdquo IEEE Transactions on SystemsMan and Cybernetics Part A Systems and Humans vol 43 no5 pp 1028ndash1041 2013
[43] HWHethcote ldquoThemathematics of infectious diseasesrdquo SIAMReview vol 42 no 4 pp 599ndash653 2000
[44] G Chowell E Shim F Brauer and et al ldquoModelling thetransmission dynamics of acute haemorrhagic conjunctivitisapplication to the 2003 outbreak in Mexicordquo Statistics inMedicine vol 25 no 11 pp 1840ndash1857 2006
[45] Y Ge L Liu B Chen X Qiu and K Huang ldquoAgent-basedmodeling for InfluenzaH1N1 in an artificial classroomrdquo SystemsEngineering Procedia vol 2 pp 94ndash104 2011
[46] A R Tuite A L Greer MWhelan et al ldquoEstimated epidemio-logic parameters andmorbidity associatedwith pandemicH1N1influenzardquo CMAJ vol 182 no 2 pp 131ndash136 2010
[47] L Brouwers B Cakici M Camitz A Tegnell and MBoman ldquoEconomic consequences to society of pandemic H1N1influenza 2009mdashpreliminary results for Swedenrdquo Euro Surveil-lance vol 14 no 37 pp 1ndash7 2009
[48] Y Ge L Liu X Qiu H Song Y Wang and K Huang ldquoAframework of multilayer social networks for communicationbehavior with agent-based modelingrdquo Simulation Transactionsof the Society for the Modeling and Simulation International vol89 no 7 pp 810ndash828 2013
[49] W Duan X Qiu Z Cao X Zheng and K Cui ldquoHeterogeneousand stochastic agent-based models for analyzing infectious
diseasesrsquo super spreadersrdquo IEEE Intelligent Systems vol 28 no4 pp 18ndash25 2013
[50] The Preparation and Emergency Response Plans to PandemicPublished by Ministry of Health Ministry of Health 2005(Chinese)
[51] R M Fujimoto Parallel and Distributed Simulation SystemsWiley Interscience 2000
[52] F Cicirelli A Giordano and L Nigro ldquoDistributed simulationof situated multi-agent systemsrdquo in Proceedings of the 15thIEEEACM International Symposium on Distributed Simulationand Real Time Applications (DS-RT rsquo11) pp 28ndash35 September2011
[53] B Chen and G Guo ldquoA two-tier parallel architecturefor artificial society simulationrdquo in Proceedings of the 26thACMIEEESCS Workshop on Principles of Advanced and Dis-tributed Simulation (PADS rsquo12) pp 184ndash186 July 2012
[54] G Guo B Chen and X Qiu ldquoParallel simulation of large-scale artificial society with GPU as coprocessorrdquo InternationalJournal of Modeling Simulation and Scientific Computing vol4 no 2 Article ID 1350005 2013
[55] B Chen L-B Zhang X-C Liu and H Vangheluwe ldquoActivity-based simulation using DEVS increasing performance by anactivity model in parallel DEVS simulationrdquo Journal of ZhejiangUniversity Science C vol 15 no 1 pp 13ndash30 2014
[56] X-L Wang P Yang Z-D Cao et al ldquoQuantitative evaluationon the effectiveness of prevention and control measures againstpandemic influenza A (H1N1) in Beijing 2009rdquoChinese Journalof Epidemiology vol 31 no 12 pp 1374ndash1378 2010 (Chinese)
[57] S Mei A D van de Vijver L Xuan Y Zhu and P M ASloot ldquoQuantitatively evaluating interventions in the influenzaA (H1N1) epidemic on China campus grounded on individual-based simulationsrdquo Procedia Computer Science vol 1 no 1 pp1675ndash1682 2010
[58] X Qiu Z Cao and Z Li Mid-Term Reports of China NationalScience Foundation 91024030 2014
Submit your manuscripts athttpwwwhindawicom
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
MathematicsJournal of
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Mathematical Problems in Engineering
Hindawi Publishing Corporationhttpwwwhindawicom
Differential EquationsInternational Journal of
Volume 2014
Applied MathematicsJournal of
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Probability and StatisticsHindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Journal of
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Mathematical PhysicsAdvances in
Complex AnalysisJournal of
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
OptimizationJournal of
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
CombinatoricsHindawi Publishing Corporationhttpwwwhindawicom Volume 2014
International Journal of
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Operations ResearchAdvances in
Journal of
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Function Spaces
Abstract and Applied AnalysisHindawi Publishing Corporationhttpwwwhindawicom Volume 2014
International Journal of Mathematics and Mathematical Sciences
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
The Scientific World JournalHindawi Publishing Corporation httpwwwhindawicom Volume 2014
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Algebra
Discrete Dynamics in Nature and Society
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Decision SciencesAdvances in
Discrete MathematicsJournal of
Hindawi Publishing Corporationhttpwwwhindawicom
Volume 2014 Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Stochastic AnalysisInternational Journal of
Mathematical Problems in Engineering 5
Model DevelopmentTool (MDT)
Artificial SocietyEditor (ASE)
Population and GeospatialEnvironment Generation Tool (PGET)
Emergent EventsConfiguration Tool (EECT)
Intervention MeasuresConfiguration Tool (IMCT)
Experiments DesignTool (EDT)
Experiments ManagementTool (EMT)
Artificial SocietySituation Tool (ASST)
Open Source DataRegistration Tool (OsdRT)
Generic ModelingEnvironment (GME) Artificial
society (A)Computationalexperiments (c)
MDT EECT IMCT
ASST OsdRT
ASE PGET EDT EMT
Parallel execution (P)
Figure 3 The implementation of KD-ACP
Table 1 The working environment and programming languages of tools
Tools Working environment Programminglanguages
Developmentplatform User type
Generic ModelingEnvironment (GME)
General computer (desktopapplication) NULL NULL Domain experts
Model Development Tool(MDT)
General computer (desktopapplication) C++ Visual Studio Model developers
Artificial Society Editor(ASE)
General computer (desktopapplication) C Visual Studio Domain experts
Population and GeospatialEnvironment generationTool
General computer (desktopapplication) C Visual Studio Domain experts
Emergency EventsConfiguration Tool (EECT)
General computer (withInternet Explorer clientside)
C ASPNET Domain experts
Intervention MeasuresConfiguration Tool (IMCT)
General computer (withInternet Explorer clientside)
C ASPNET Domain experts
Experiments ManagementTool (EMT)
General computer (withInternet Explorer clientside)
C ASPNET
Users of computationexperimentServer of EMT
Nodes in supercomputer(Console Program serverside)
C++ Visual Studio
Runtime Infrastructure ofEMT
Nodes in supercomputer(Console Program serverside)
C++ Visual Studio
Artificial Society SituationTool (ASST)
General computer (desktopapplication) C++ Visual Studio Domain experts
Open Source DataRegistration Tool (OsdRT)
General computer (withInternet Explorer clientside)
Java JSP Domain experts
6 Mathematical Problems in Engineering
M2T transformation
Modeling in domain
GME Generic Modeling Environment
Metamodeling
Instance of
FSM
Metamodeling for artificial society
DEVS
Petri Net State charts
Agentmetamodel
Environmentmetamodel
Emergency eventmetamodel
Interventionmetamodel
DAE
Domain-specific metamodels Semantic well definedmetamodels
M2M transformation
Modeling of artificial society
Interventionmodel
Emergency eventmodel
Environmentmodel
Agentmodel
FSM based emergent andintervention models
FSM based environmentmodels
FSM based agent modelsDomain-specific models
Semantic well defined models
M2M transformation
Agent model code Environment model code
Agent behavior model
Agent mentality model
Disease model code
Basic population model Disease spreading model
Disease state transformmodel
Intervention model code
Agent social networkmodel
Basic environmentinformation model
Roadnet model
Vaccination
Infectors isolation
Contactors isolation
Close work officesAgent contact model
Domain-specific models relevant codes
MDT Model Development Tool
Implementation
Figure 4 From metamodelling and modelling by GME to implementation by MDT of artificial society
331 The Principle of the Modelling of Artificial SocietyAccording to the principle of domain-specific modellingthe modelling of artificial society contains the followingsteps first metamodelling the basic elements of artificialsociety second modelling the specific features in domain ofemergency management third implementing the models ofartificial society in codes The whole process is illustrated inFigure 4The first and second steps are implemented in GMEwhile the third step is implemented in MDT
The first step is metamodelling which mainly focuses onconstructing the metamodels of artificial society Metamod-elling tries to study the common patterns of artificial societyThe outputs of metamodelling are metamodels which repre-sents the abstraction of the whole systemThe basic elementsof artificial society are described in metamodel The processof metamodelling is divided into four phases The first is theconstruction of the domain-specific metamodels As shown
in Figure 4 agent metamodel environment metamodelemergent event metamodel and intervention metamodelcompose the metamodel of artificial society The second isthe construction of the metamodels described by typicalmodelling formalisms such as FSM DAE DEVS and PetriNet [27] These formalisms are all semantically well definedThe third is the definition of the model transformationfrom domain-specific metamodels to metamodels of typicalmodelling formalisms The transformation standardizes themetamodels of artificial society by typical modelling spec-ifications The fourth is the definition of the transforma-tion templates from metamodels to code framework Thetemplates list the basic abstract interfaces of metamodels ofartificial societyThese abstract interfaces are implemented inthe specific-domain modelling and code generations
The second step is modelling the models of artificialsociety such as agent model environment model emergent
Mathematical Problems in Engineering 7
Figure 5 The metamodels of artificial society in GME
event model and intervention model are built Actuallythe models are the instantiation of metamodels in the laststep Different from the general modelling environment likeUML [28] the domain-specific modelling provides a familiarmodelling environment for the domain experts in artificialsociety For example emergency response experts only con-cern emergent eventmodel and interventionmodel inheritedfrom metamodels After constructing the domain-specificmodels based on domain-specific metamodels domain usersexecute the model transformation defined in the first stepAll the models of artificial society are transformed into FSMmodels As a result the models are implemented in this uni-fied modelling formalism (FSM) The model transformationmakes the simulation of the models possible
The third step is the generation of executable codes ofmodels The executable code framework is generated bymapping template from metamodels to code frameworkdefined in the first step Moreover domain developers alsoadd necessary codes to the framework to integrate thedynamic semantics of the models The code frameworkoutputs the dynamic link libraries by compilingThe dynamiclink libraries are loaded in the large scale artificial societyruntime infrastructure in computational experiments
332 The Metamodelling and Modelling of Artificial Societyby GME GME is used to build metamodels and modelsin our work As mentioned before the abstraction andcommon patterns of society are represented in metamodelsAccording to the bottom-up modelling style metamodels ofagent environment and communications are described inGME Figure 5 shows part of metamodels of artificial societyThe features of an agent metamodel are extracted from thecensus figures and statistical data Environment metamodelsimulates the geospatial places for the behaviors of agentsThemetamodel of communications among agents ismodeledto simulate the interactions such as infection in epidemicsand rumor propagation in public opinion formation eventsIt is worth notifying that the metamodel of communications
includes both the emergent event metamodel and interven-tion metamodel
From the perspective of modelling the details from spe-cific domains are considered in the models by the instantia-tion frommetamodels of artificial society For example socialrelationships based on complex networks are added in agentmodel to support the communications Agent activity is alsoused to quantify the agent activity under different scenarioEnvironment models are linked with the help of transporta-tion services subways and roads are modeled while the pathsearch is encapsulated in the services Emergent event modeland intervention model are also the domain-specific modelsThe modelling of artificial Beijing in GME will be discussedin detail in next section
333 The Implementation of Models of Artificial Society byMDT As mentioned before MDT is used to implementmodels such as agent environment emergent event andintervention According to the template of code frameworkthe implementations of models are generated by MDT Theimplementations are classified into two categories FSMmodels and services FSM models such as agents andenvironments are built under the specification of FiniteState Machine (FSM) [29] in MDT while all the servicessuch as transportation are encapsulated under the PublicService Standard This standard provides a generic interfacespecification for modelers to encapsulate public commonservices in artificial society FSM models like agent are builtstatistically from the quantitatively analyzed characteristicssuch as demographic attributes social behaviors emergencybehaviors and social networks Social behaviors describe thedaily behaviors of individuals while emergency behaviorsdescribe the individual behaviors in emergent events Forexample infected individuals are all isolated in hospital inSARS Isolation ismodeled as a typical emergency behavior inour work Correspondingly services are used to simulate themacroactual society Take transportation service for instancethe path search is needed by almost every agent duringmoving from spot to spot
8 Mathematical Problems in Engineering
Artificial society description in emergency
Description ofemergency
ArtificialSociety Editor
(ASE)
Agent behavior set
Agent role setAgent status set
Environment setAgent relationship set
Statistical data of real society
Descriptionof artificial
societyframework
Terminal of agent modeling andartificial society describing
middot middot middotAgent
metamodeling
Customized agentmodeling
Relationship modelDailyEmergency
Mod
el D
evelo
pmen
t Too
l (M
DT)
Agent modeling
Agent model Encapsulate
Encapsulate
Behavior model
Intervention
Emergency
Response
Daily
Social relationshipmodeling
Socialbehaviormodeling
Socialbehaviormodel
Emergencybehavior
model
Emergencybehaviormodeling
Building
Geospatial statistical data
Population statistical data Datacollection
Actualsociety
Artificial societypopulation and
geospatial database
Organizationservice
Transportationservice+
MappingAgent model andservice repository+ emergency model
repository
Climateservice
Geographicalservice
+
Agent status in emergencyEmergency management
Emergent event set
Social relationship statistical data
Environment statistical data
Population behaviorstatistical data
Emergency statistical data
Emergency organization statistical data
Geospatial and SocialEnvironment Generation
Tool (GSeT)
Agent model repository andpopulation database
C++C++C++C++
Modeling
Statistical modelJob ageSex location
Public service modeling
Organizationmodel
Transportationmodel
Geographicalmodel
Climatemodel
Interventionservice
Publ
ic se
rvic
e sta
ndar
d
Figure 6 The editing and initialization of KD-ACP
MDT provides domain experts with a tool to obtain thecode implementations of models With the help of compileenvironment like visual studio MDT also supports thefurther programming development of the specific domaindetails which cannot be described in modelling step
Both the FSM models and services are developed by theMDT first and then stored in the agent model and servicerepository The repository manages the models according tothe requirements from emergency problems and provides themodels for EDT to make the experiment plans
34 The Editing and Initialization of Artificial Society byASE and PGET As shown in Figure 6 ASE is used to editthe scenarios of artificial society within emergent eventsThe editing is composed of two parts (1) the statisticalinformation of artificial society in daily life such as the rolesof agents the relationships of agents and the types of environ-ments and (2) the statistical information of artificial societyin emergency including the statistical data of emergent
events the emergency organization and emergency relatedbehaviors of agents
According to the requirements of the editing thesestatistical data are collected from actual society manually bythe domain experts Based on these statistical data PGETgenerates the artificial society population and geospatialenvironment database The database supports the instanti-ation of artificial society at individual level For examplethe attributes such as age and gender of each agent can befound in the database With the support of the databaseFSM models and service repository discussed before itis sufficient for domain experts to study the emergencyproblems by computational experiments
35The Computational Experiments and Parallel Execution inKD-ACP The tools of ldquoCrdquo part and ldquoPrdquo part in KD-ACP areused to support the process of computational experimentsand parallel execution The working process is shown inFigure 7 EECT and IMCT are both the starting and ending
Mathematical Problems in Engineering 9
Agent model andservice repository
repository
Emergency EventsConfigurationTool (EECT)
Open Source DataRegistration Tool (OsdRT)
Intervention MeasuresConfigurationTool (IMCT)
Load
Output Output
Internet
Emergencyresponse plans
Load
∙ Input models of emergencyevents∙ Output settings of agentbehavior in emergent events+ settings of emergent events
∙ Input models of interventionmeasures∙ Output settings of agentbehavior in interventionmeasures + settings ofintervention measures
+ emergency model
∙ Input data from internetnetworks∙ Output population data +emergency events data +intervention measures data
Artificial societypopulation and
geospatialenvironment database
Artificial societyruntime database
OutputThe most optimal
emergencyresponse plan
Actual society
OutputOutputOutput
Input
Load
Load
Load
Load
Output
Artificial SocietySituation Tool (ASST)
Emergency decisionorganization
ClusterRuntime infrastructure
Deploy
Deploy
Experimentsdesign tool (EDT)
Artificial societyinitial data file
Experimentplan
Artificial societymodels
Experiments Management
large scale artificial societyruntime infrastructure
Tool(EMT) +super computerTIANHE-1A
Load
∙ Input experiment plan +artificial society models +artificial society initial data file∙ Output artificial societyruntime state data
∙ Input description of artificialsociety + artificial society populationand geospatial environment society+ emergent events configuration +
intervention measures configuration∙ Output experiment plan + artificialsociety models + artificial societyinitial data file
∙ Input runtime statistical data∙ Output graphics charts ofartificial society statistical data +situation of artificial society
Figure 7 The computational experiments and parallel execution of KD-ACP
point The emergent events and intervention measures areconfigured by EECT and IMCT respectively The configu-rations of emergent events are used to simulate both thereal emergencies like SARS and H1N1 and the supposedemergencies for experiments Similarly the configurationsof intervention measures are also used to reproduce thereal one and simulate the supposed one The repeat of theemergency is used to verify the models while the supposedconfigurations are used to obtain the optimized decision planto the response of the possible emergencies
With the input of artificial society model and servicerepository artificial society population and geospatial envi-ronment database and the configurations discussed beforeEDT generates the experiment plans to meet the require-ments of research on emergency management The outputof EDT includes artificial society models artificial societyinitial data files and experiment plan The models aredownloaded from the repository while the data file is thecollection of data from the database to initialize the modelsWhen the models and data files are ready EMT loads theexperiment plan and deploys the models and data to thecluster or TIANHE-1A supercomputer [30] which was theworldrsquos fastest supercomputer built by National University of
Defense Technology (NUDT) in China in 2010 According tothe plan the experiments are done repeatedly on the largescale artificial society runtime infrastructure [31 32] by themultisample settingsThework process is the implementationof computational experiments in ACP approach
Traditionally emergency response plans are made byemergency management theories and experiences The onlyway to test the effective of plans is the feedback results of realworld ACP approach provides a novel method to supportemergency response plan making by parallel execution Asshown in Figure 7 the work process of KD-ACP is composedof two loops The inner loop composed by red arrowsdescribes the process of computational experiments while theouter loop of yellow arrows illustrates the process of parallelexecution During the runtime of computational experi-ments the statistical data of artificial society is collected andstored in the artificial society runtime database Based on thedatabase ASST outputs the customized situation of runningartificial society by graphics charts and situation mapsThe information is sent to the organizations of emergencydecision to support making the emergency response plansWith the help of computational experiments loop these plansare simulated repeatedly to find the most optimal one
10 Mathematical Problems in Engineering
Open Source Data Registration Tool (OsdRT)
Emergent EventsConfiguration Tool (EECT)
Intervention MeasuresConfiguration Tool (IMCT)
∙ Web information collecting
∙ Deep mining in social media
∙ Web information denoising
∙ Information filtering
Data acquisition
Actual society
Open source informationin social media
Internet
∙ Basic element extracting∙ Individuals and organizationsextraction∙ Sentiment analyzing andopinion mining∙ Social network analyzing
Data extraction
∙ Domain knowledge description∙ Information fusion and collisiondetection∙ Domain knowledge construction∙ Social media informationstandardization
Data standardization
Knowledge
Figure 8 The compositions of OsdRT [34]
Moreover the most optimal plan is used to the responseof emergency in actual society According to the idea ofparallel control in [24] the feedback of actual society ispartly collected from Internet networks by OsdRT As shownin Figure 8 OsdRT is composed of three components dataacquisition data extraction and data standardization Dataacquisition collects mines and filters information fromsocial sensing networks Data extraction includes basicelement extracting individual and organization extractingsentiment analyzing and social networks analyzing Datastandardization specifies the useful knowledge and sent it toconfiguration tools in KD-ACP
By processing in OsdRT the knowledge about emergentevents and intervention measures are analyzed first andregistered in the EECT and IMCT The registration updatesthe settings of emergent events and intervention measuresThis loop composed of yellow arrows implements the parallelexecution in ACP approach The implementation of parallelcontrol and management provides a data-driven approachthat considers both the engineering and social complexityfor modelling analysis and decision making in emergencymanagement
4 Modelling Beijing City with KD-ACP
41 How to Build the Artificial Beijing According to themodelling of artificial society in KD-ACP discussed beforethe Beijing city is modeled as follows
To meet the requirements of emergency managementsthe basic elements of artificial society are extended Asshown in Figure 9 six elements are required to simulate thecity agents environments transportation activity schedulecommunication and agent activity
411 Modelling Artificial Beijing Figure 10 shows the mainGUI of GME for the modelling of artificial modelling inpublic health events Metamodels are listed in the left area ofFigure 10 the list provides basic syntax elements for domainexperts to model artificial society Domain experts build
models based on the knowledge of their own MeanwhileGME supports hierarchy for building large scale systemsThesyntax symbol listed in GUI can be extended in new tab bydouble clicking Take agent for example the model of agentcan be detailed by edition in another tab page of agent
As shown in the center of Figure 10 the models ofartificial Beijing consist of five parts models of agent andenvironment domain models of public health events inter-ventionmodels controllermodels and services Agentmodeldescribes individuals in society it is composed of basicpopulation information action social relationships activityschedule and disease related information Activity schedulerepresents individualrsquos physical actionmodel focusing on thedaily action of agents Environment model includes physicalentities such as buildings playground transportations andagents contained in environment Domain models of publichealth events are composed of the propagation model of dis-ease disease state transition model and so on Interventionmodels include the settings of vaccination isolation and soon The models mentioned before are all FSM models Themechanisms of these models will be detailed in next sectionsController models and services are the public service mod-ules they are implemented in the development in MDT
412 Modelling Agents and Environments Under the speci-fication of FSM agent and environment models are imple-mented in two parts the state space and state transitionsThe state space is composed of the demographic attributesand behavior related attributes The transitions are triggeredwhen the conditions of states are satisfied As shown inFigure 11(a) the action of agent is changed when the ldquonexttimerdquo condition is satisfied in agent model while the agentslist is changed when the agent arrival condition is satisfied inthe environment model
413 Modelling Activities Agent activities come from theagent state transitions of actions such as movements andcommunications The actions of agents are instructed by theactivity schedule shown in Table 2 Activity schedule lists
Mathematical Problems in Engineering 11
Transportation EnvironmentsEnvironments
Agents
Artificial Beijing
Agent activityActivity schedule
Agent activity
Communication by socialrelationship networks
Communication by socialrelationship networks
Figure 9 The basic elements of artificial society
Syntax elements of DSM
Services
Models of agentenvironment andactivity schedule
Domain models of publichealth events
Intervention models
Modelingaspect ofdomainexperts
Controllermodels
Figure 10 The models of artificial Beijing in GME
all the actions with probability in one day for agents inboth normal and emergent situation [38] There are severaltypes of activity schedule in artificial Beijing student agentactivity schedule worker agent activity schedule emergentagent schedule and so on For example Table 2 gives an agentactivity schedule Upon the instruction of activity schedulestudent agent changes the actions by 119901
119894after state transitions
The119901119894in the tablemeans the action probability in the relevant
period In the duration from 0800 to 1200 a student agenteither goes to classroom to have class or goes to libraryto study The probability of class action is 119901
2while the
probability of study action is 1 minus 1199012
Agent behaviors are decided by the settings of activityschedule In addition to the daily activity schedulementionedbefore emergent activity schedules are also considered inour work Take public health events for instance an infectedagent changes schedule from a normal one to emergentone The workflow of a susceptible agent is illustratedin Figure 12 to show the change of behaviors After theinfection the agent is set in incubation phase Not all theincubation agents will become symptomatic Some of themturn back to being susceptible and some of them becomesymptomatic The symptomatic agents change their activityschedule from normal to emergent In the emergent case
12 Mathematical Problems in Engineering
Next actionNext environment (location)
Advance to next actiontime
Find next environment inactivity chain by action
Current actionCurrent environment (location)
Q
q0
120575 Find next action inactivity schedule by time
F
Σ
(a) The state transitions of agent
Q
q0
120575 F
Current agent list
Current pollution level
Next agents list
Next pollution level
Agents entered
Infected agents entered
Add agents to currentagents list
Compute pollution levelby infected agents
Σ
(b) The state transitions of environment
Figure 11 The state transitions of models
Table 2 The activity schedule of a student agent
Duration (Δ119905) Activity Location Probability 119879Location (minute)0000ndash0600 Sleep Dormitory 119875
0(100) 360
0600ndash0800 Breakfast Dormitoryrestaurant 1198751(068) 120
Sports-breakfast-travel Playground-dormitoryrestaurant 1 minus 1198751(032)
0800ndash1200 Class Classroom 1198752(077) 240
Study Library 1 minus 1198752(023)
1200ndash1400 Lunch Restaurant 1198753(090) 120
Lunch-shopping Restaurant-convenience store 1 minus 1198753(010)
1400ndash1800 Class Classroom 1198754(070) 240
Study Library 1 minus 1198754(030)
1800-1900Dinner Dormitoryrestaurant 119875
5(063)
60Sports-shopping Playgroundconvenience store 1198756(025)
Study Classroomlibrary 1 minus (1198755+ 1198756) (012)
1900ndash2200Rest Dormitory 119875
7(045)
180Dinner-rest Dormitoryrestaurant-dormitory 1198758(020)
Study Classroomlibrary 1 minus (1198757+ 1198758) (035)
2200ndash2400 Rest Dormitory 1198759(072) 120
Sleep Dormitory 1 minus 1198759(018)
infected agents go to hospital according to the treatmentschedule or stay in dormitory according to isolation scheduleAfter the treatment in hospital or self-immunoprocess agentsbecome healthy and immune of disease If the agents aretreated in the hospital they are not allowed to get out untilrecovered Moreover activity schedules are also influencedby the emergency response plans For example in the case ofisolation in emergency response plans the activity scheduleof agents who had contacted with infected is changed Onlythe locations including home and dormitory compose thisemergent schedule Likewise based on the statistical datafrom emergency response plans the additional possibilities
are added into activity schedulesThe behaviors of agents alsochange with activity schedules In the view of ACP approachthe injection of new emergency response plans implementsthe parallel execution of microagent behaviors
In another aspect contact frequency is also anothercrucial element to determine the infected rate in public healthevent It is important to model the contact frequency andcontact time of individuals Based on the studies on thecontact behavior of human being by questionnaire surveyEdmunds found that the contact frequency of individualcould be fitted approximately into a normal distributionand the mean and the standard deviation distributions are
Mathematical Problems in Engineering 13
Susceptible agent ImmuneYes
Healthy and immune
No
Choose schedule
Self-isolation Go to hospital
Staysymptomatic
No
Receive treatment
Contact withinfected agents
Symptomatic
No
Self-immunoprocess
Yes
Recovered
Yes
Incubation
Infection
No
Yes No
NoYes
Figure 12 The workflow of a susceptible agent
168 and 85 [39] So it is possible to apply normal randomvariable to model the contact frequency In our work Box-Muller method [40] is used to generate the random variableof contact frequency shown in the following
119865119862= 120583119865+ 120590119865(minus2 ln (120574
1))12 cos (2120587120574
2) (1)
in which 119865119862is the random variable of contact frequency 120583
119865
is the mean value of normal distribution 120590119865is the standard
deviation of normal distribution and 1205741and 120574
2are the
uniform random variables distributed in the interval [0 1]Based on (1) and survey data [41] the contact frequenciesof agent 119865contact(119860 119894) are discretized as in Table 3 within theconsideration of activity differences
Similarly the duration per contact between individualsis another key factor of AHC transmission The duration ofcontact could also bemodeled by the normal randomvariablein the following [42]
119879119862= 120583119879+ 120590119879(minus2 ln (120574
1))12 cos (2120587120574
2) (2)
in which 119879119862is the random variable of duration per contact
120583119879
is the mean value of normal distribution 120590119879
is thestandard deviation of normal distribution and 120574
1and 120574
2
are the uniform random variables distributed in the interval
[0 1] In addition because the time spent in specific location119879Location(119860 119894) also follows the random distribution like (3)in our work In order to make the duration of contact119879contact(119860 119894) shorter than the 119879Location(119860 119894) the mean andstandard deviation of (2) is set up in (3) and (4) [42]
120583119879=119879Location(120583119865+ 120590119865) (3)
120590119879=119879Location(10120583119865+ 10120590
119865) (4)
in which 119879Location is the time of an activity in a specificlocation set in Table 2 120583
119879and 120590
119879are the mean and standard
deviation of contact frequency As a result the duration ofagent 119879contact(119860 119894) in specific locations (listed in Table 2) isalso discretized in Table 3 within the consideration of activitydifferences
414 Modelling Transportation In artificial Beijing threekinds of travel models are considered walking inside thedistrict travelling by road networks and travelling by subwaynetworks According to the traveling models the agentmovements are implemented by the compositions of travelingmodels before So it is important to build the basic road
14 Mathematical Problems in Engineering
Table 3 The contact frequency and duration in different locations and activities by social relationships in case of student and teacher
Social relationship Activity Location 119879contact(119860 119894) (minute) 119865contact(119860 119894)
Student-student
Sleep Dormitory 0 0
Breakfast Restaurant 119873(52) 119873(32)Breakfast Dormitory 119873(52) 119873(21)Sports Playground 119873(21) 119873(53)Class Classroom 119873(21) 119873(21)Study Library 119873(21) 119873(21)Lunch Restaurant 119873(158) 119873(32)
Shopping Convenience Store 119873(203) 119873(21)Dinner Restaurant 119873(108) 119873(21)Dinner Dormitory 119873(158) 119873(32)Rest Dormitory 119873(63) 119873(52)
Teacher-student Class Classroom 119873(21) 119873(21)
Subway network of Beijing city (red lines) Road network of Beijing city (green lines)
Figure 13 The implementation of transportation for artificial Beijing
and subway networks With the help of Google map thekey points of road are sampled first and the road networksare generated by the links among point Similarly subwaynetworks are generated by the links between subway stationsmapped from the real stations in Google map Figure 13shows the road and subway networks in artificial Beijingwith the road networks in green and the subway networks inred Most of the districts in Beijing city are covered by thesetwo networks The algorithm of hierarchical route planningis proposed as follows
(i) Find the next position where agent will be located innext action
(ii) Obtain the travelingmodels from the statistical trans-portation data
(iii) Search the path from the starting position to thenearest road entry or subway stations
(iv) Search the optimal path from the starting station orroad entry to the nearest station or road exit of targetposition
(v) Search the path from the target subway station or roadexit to the target position
(vi) Attach the subway train number or the bus numberincluding the transfer information to the path accord-ing to the timetable of subway or bus
(vii) Connect all the paths obtained before Generate thepath from the starting position to the target position
Based on the algorithm for each agent in artificial Beijingthe transportation is simulated during the computationalexperiments It is worth noting that the activities insidethe buses or train cars are considered because the agentcommunication during travelling is also necessary to besimulated
Mathematical Problems in Engineering 15
(a) The implementation of agent in MDT (b) The code framework of agent
Figure 14 The agent model implementation in MDT [35]
415 The Implementation of Models by MDT All the modelsare developed in the MDT Figure 14 shows the GUI of thetool MDT is used to implement the state space and statetransitions of FSM models The code framework is shownin Figure 14(b) Using the C++ inheritance features and thetechniques of dynamic link library (dll) the models areencapsulated in the dll componentsThe flexible compositionmechanism is designed to support the models evolution incomputational experiments
416 The Description of Artificial Beijing by ASE As men-tioned in Section 34 ASE describes the artificial Beijingin macroview The statistical information of Beijing city iscollected by ASE Table 4 lists part of the description ofartificial Beijing generated by ASE
42 How to Obtain the Data of Beijing City
421 Population and Geospatial Environment Database ofBeijing City Based on the models developed for ArtificialBeijing it is necessary to generate the initial data to do thecomputational experiments Because the individual level dataare not available it is necessary to construct an individual-based population database for both accurate computationalexperiment and determining optimal decisions
According to the state spaces of models discussed inSection 412 the geospatial and population database of Arti-ficial Beijing is designed as shown in Figure 15 The kernelpart of database consists of the tables such as agent list tablegeospatial environment list table household list table andagent distribution table Agent list table is an individual leveltable used to store the data to initialize the state space of agentmodel such as id and gender age Geospatial environmentlist table is also an individual level table which stores thedata of environment attributes such as id street id andtype Household list table and agent distribution table storethe statistical data for the data generation for agent andenvironment models With the help of the geospatial andpopulation database the data generating of artificial Beijingis proposed in the next section
422 Generating Geospatial and Social Environment Datafor Artificial Beijing by PGET The algorithm of generatinggeospatial and social environment data for artificial Beijingis used to quantify the spatial distribution of population andformalize geospatial behavior of each agent The algorithmshown in Figure 16 is capable of generating a syntheticartificial society which allowsmultiresolution statistical datasocial interactive behavior and multilayer social networksto be integrated together The synthetic population canrepresent individual agents in the form of households andhousehold members and the synthetic population is statis-tically equivalent to a real population For each householdcharacteristics such as address household size family typesand relationships are generated Each person is described bycharacteristics such as age gender social role and correlatedlocations The algorithm provides an effective methodologyto reconstruct the computing environment in high resolutionby using statistical data in low resolution leading to betterprediction and management of emergencies The details ofalgorithm are illustrated in [36]
As shown in Figure 17 the implementation of the algo-rithm is embedded inside the GEPTThe collective input datasuch as the data files of population distribution is collected inthe tool first then the algorithm is activated to generate thedatabase The generation lasts almost twelve hours one timewith the specified parameters settings When the generationis finished the data of artificial Beijing is obtained in thedatabase with 19610000 agents and 16000 environments
5 H1N1 Experiments in ArtificialBeijing with KD-ACP
With the support of the artificial Beijing modeled before thecomputational experiments and parallel execution are alsoimplemented by KD-ACP The experiments are designed tofind the most optimal emergency decision response plansEpidemic in city is a typical scenario in emergency manage-ment H1N1 epidemic in Beijing in 2009 is used to be the usedcase to test KD-ACP According to Section 34 it is necessarytomodel emergencymanagement first in order to support thecomputational experiments
16 Mathematical Problems in Engineering
Table 4 The description of artificial Beijing generated by ASE
Statistical features ofartificial Beijing Descriptions of statistical features Data sourcedefault values
Agent roles Baby child primary student middle school student college student worker retired
Demographic data file ofBeijing
Environment types Super market store restaurant park stadium hospital school cliniccommercial building dormitory building residential building
Demographic data file ofBeijing
Demographic data Age distribution of population sex distribution of population householddistribution of population retired age distribution children distribution
Demographic data file ofBeijing
Environment data Distribution of all types of environments Environment distributiondata file of Beijing
Relationships Classmate relative college friendship family Demographic data file ofBeijing
Varying parameters
Age difference of couples119873(120583120590) 119873(03) isin [0 12]Minimummarriage age (male female) (22 20)Age difference of children119873(120583120590) 119873(13) isin [1 10]Enrollment rate of kindergarten 120572 08Enrollment rate of primary school 120573 1Enrollment rate of middle school 120574 1Enrollment rate of college 120575 08Entrance age of kindergarten 1205721015840 3Entrance age of primary school 1205731015840 7Entrance age of middle school 1205741015840 13Entrance age of middle school 1205751015840 19
Age distributioninformation table
Agent settable
Geospatialenvironment set table
Household settable
Figure 15 The geospatial and population database of Beijing city
51 Modelling Emergency Management The modelling ofemergency management consists of two parts the modellingof emergent events and the modelling of intervention mea-sures H1N1 and the intervention measures for H1N1 aremodeled in our case
511 Modelling H1N1 by MDT H1N1 Model in agent simu-lates the states transitions of health status and the relevantactions Referring to SIR (SusceptibleInfectiveRecovered)[43 44] models agent has three health statuses susceptibleinfected and healthy with immunity Only susceptible agentcan be infected by the contact with infected agents After
the three infected phases (incubation being symptomaticand recovery) of H1N1 model agent is healthy again It isworthy mentioning that the recovered agents can be infectediteratively when the immunity disappears gradually by timeThe state transitions are illustrated in Figure 18
According to [45] the latent period of H1N1 was in aWeibull distribution within people from one to seven days[46 47] The distribution was usually centered in the rangebetween one and three days AWeibull random variable [40]is used to model the latent period as described in
119879Lat = 120572 times [minus ln (1 minus 120574)]1120573+ 120592 (119879Lat ge 120592) (5)
Mathematical Problems in Engineering 17
Censusdata
Generate GUID
Assign individualhousehold
Generate familystructure
Assign family roleto every member
Endow individualwith age gender
social role
Distribute environmentlocations on the map
School Workplace Entertainmentlocations
Consumptionlocations
Assign school orworkplace to
individual
Correlatedindividual withother locations
Generate friendshipin school
Generate coworkerrelations inworkplace
Assign individualadministrative
region
Generate housebuildings
Locate householdto house building
Generate familyrelation
Generateneighborhood
relation
PopulationEnvironment
Environment listSocial network
Figure 16 The algorithm of generating geospatial and social environment data [36]
Edit
Edit
Edit
Edit
Edit
Edit
BrowseStatistical characteristic of population
Geospatial population data
Population data of street
Population types
Geospatial and social environment generation tool
Workspaces
Parameters distribution settings
Household types
Roles in household
Household structure
Social roles
BrowseBrowseBrowseBrowseBrowseBrowse
Population distribution data
File path
Retired distributionPop distributionHousehold numHousehold distributionSex distributionAge distributionData type
Figure 17 The GUI of Population and Geospatial Environment generation Tool (PGET)
18 Mathematical Problems in Engineering
Infected
SusceptibleHealth
withimmunity
Infected
by contac
t
with in
fected
agents
Recovery
Increase before beingsymptomatic symptomatic
Decrease after being
Decrease by time
Infectivity
Immunity
Figure 18 The transitions of agent health status
119879Lat denotes the duration of latent period 120592 120572 and 120573 arethe location parameter scale parameter and shape parameterof Weibull distribution respectively 120574 is a uniform randomnumber in the range [0 1] According to statistics [47] 120592 120572and 120573 are set as 0 18 and 121 respectively Then the mean(standard deviation) of latent period is calculated as 159 days(058 day2) Furthermore infectious period is set as 7 days
A susceptible studentrsquos probability of getting infecteddepended on the infectivity of infectious agents his ownimmunity level the duration of the contact action and soon The infectivity of an infected agent evolved with timeduring the infectious period According to the statistics ofinfectivity in the chart with red bars in Figure 19 an infectedstudent had the highest level of infectivity in the second dayafter he had the first symptom The infectivity levels in otherdays in comparison with the second day are in Table 5 Theday labeled as ldquominus1rdquo means the day before the starting of thesymptomatic period
Based on the algorithm discussed before theH1N1modelis developed with the help of MDT The settings of influenzamodel will be discussed in Section 521
512 Modelling Intervention Measures by MDT In corre-spondence intervention measures model is used to ceasethe outbreak of H1N1 As discussed in [42 48 49] if theappropriate measures are taken when the infectious diseaseemerges the transmission of the disease could be sloweddown and the damage could be decreased Therefore it isnecessary and important to design emergency interventionmeasures in the artificial Beijing According to the MinistryofHealth inChina the intervention policies [50] are designedto control the spread of influenza such as H1N1 The inter-ventionmeasures of these policies including are interventionsactivated time vaccination rate antibiotic delay to hospitalisolation duration close workspace duration and limitationof activities listed in Table 6 Similarly the models of theseintervention measures are also developed with the help ofMDT Using the reasonable data ranges listed in Table 6intervention models can be initialized to do the computa-tional experiments These experiments are used to show how
Symptomatic1 2 3 4 5 6 7minus1
Recovered
Tim
e (da
y)
Infe
ctiv
ity
Figure 19 The distribution of infectivity rate of an infected agent[37]
Table 5 The proportion of infectivity
Days (119894) minus1 1 2 3 4 5 6 6Proportion of infectivity 03 05 1 08 04 02 01 005
to restrain the H1N1 transmissionThe settings of these inter-vention measures in IMCT are introduced in Section 522
52 Experiment Settings
521 H1N1 Settings by EECT EECT is designed to initializethe emergent events models such as H1N1The settings comefrom the specified emergent event models As shown inFigure 20 in theH1N1 used case the settings are composed ofthe distribution of infection source infection rate infectionperiod and so on The infection rate changes with envi-ronments and infection periods The settings are initializedfrom both the research of medicine on H1N1 and the data ofH1N1 outbreak in Beijing in 2009The process of influenza inartificial Beijing is simulated to repeat the possible scenariosof H1N1 break During the computational experiments ofartificial Beijing theH1N1 settings such as the period of beingsymptomatic are updated by the dynamic data registrationsfrom OsdRT The updating corrects the H1N1 model duringthe experiments With the help of OsdRT artificial Beijing isable to approach actual society It is also an implementationof parallel execution
522 Intervention Measures Settings by IMCT In the H1N1case the settings of intervention measures are shown inFigure 21 in IMCT These measures are designed under theinstruction of Emergency Decision Organizations like ChinaDisease Center (CDC) The measures such as vaccinationrate the rate of closedworkspace and the starting time can beinitialized in the tool Emergency response plans are imple-mented in these settings As illustrated in Figure 7 all the pos-sible compositions within value settings in reasonable inter-vals are tested to find the most optimal emergency responseplan For example different compositions of measures suchas vaccination rates and rates of workspace closed are usedto design the plans of experiments By the analysis of theresults the best composition of interventionmeasures can beobtainedThe analysis will be discussed in case study in detail
Mathematical Problems in Engineering 19
Table 6 The intervention measures of influenza
Intervention measures ValuesInterventions activated Time 119879Intervention isin [0 80th day]Vaccination rate 119877Vaccination isin [0 1]Antibiotic rate 119877Antibiotic isin [0 1]Delay to hospital 119867ospital(119860 119894) 119879infected lt 119905 lt 119879recovered + 119879HDelay 119879HDelay isin (0 3th day]Isolation duration 119868solation(119860 119894) 119879Intervention lt 119905 lt 119879Intervention + 119879IDuration 119879IDuration isin (0 14th day]Close workspace duration 119862lose(119864119894) 119879Intervention lt 119905 lt 119879Intervention + 119879CWDuration 119864119894 isinWorkspace 119879CWDuration isin (0 7th day]Limitation of agent activity duration 119871 imitation(119860 119894) 119879Intervention lt 119905 lt 119879Intervention + 119879LAADuration 119879LAADuration isin (0 7th day]119877Vaccination means the vaccination rate of artificial Beijing while 119877Antibiotic means the antibiotic rate used for infected agents in hospital119867ospital (119860119894) means that the infected agent 119860119894 is sent to hospital he or she will receive the retreatment immediately119868solation (119860119894) means that the infected agent 119860119894 is isolated at home he or she can only contact with the other agents until isolation duration is over119862lose (119864119894) means that the environment 119864119894 typed of workspace is closed until the close workspace duration is over119871 imitation (119860119894) means that the activities of agent 119860119894 are limited only the workspace and home are allowed until limitation duration is over
Incubation Weibulldistribution
Symbolic LogNormaldistribution
Recovering LogNormaldistribution
H1N1influenza
Settings of H1N1 influenza model
Disease model
Transmissionmodel
Figure 20 The settings of H1N1 in EECT
It is worth notifying that the settings of interventionmeasures are also influenced by OsdRT For example theH1N1 outbreak areas are notified by OsdRTThe interventionmeasures such as close workspace are used only in theoutbreak areas Therefore the unnecessary computation ofintervention measures is avoided So the data injection byOsdRT not only makes artificial society approach actualsociety but also increases the performance of computationalexperiments
53 Experiments by EMT When the initialization of arti-ficial Beijing is ready EMT is used to do the computa-tional experiments EMT is composed of two parts thecentral controller in the central node and the residen-tial service deployed in the computational nodes Centralcontroller loads the experiment plans and controls the whole
experiments process the GUI of controller is shown inFigure 22 EMT manages the computational experimentsin four steps cluster configuration load experiment plansmodels deployment and running experiments
531 Cluster Configuration Before the configuration clusterinformation is collected from the residential services insidethe cluster The information includes nodes amount nodenames node IPs and CPU occupations It is used to quantifythe computing power of the clusterThe cluster configurationis based on this information The nodes are selected forexperiments Then the LP (local process) [51] number foreach node is set according to the number of the CPU kernelsAfter the configuration the experiment plans are loaded tocustomize the models and data
20 Mathematical Problems in Engineering
Streets of close workspaces
Limitations ofactivities
IsolationsRespirator
Closeworkspace
Antibiotic
VaccinationHospital
Surveillance
Figure 21 The settings of intervention measures in IMCT
Information ofexperiment plan
List of experiment plans
Start | Pause | Reset | Stop | Speedup | Slowdown | Back
NodeName | CPU occupation | Memory occupation |Network | Models | Status
Figure 22 The deployment and running of experiments in EMT
532 Load Experiment Plan As discussed in Section 35 thedescription of experiment plans is composed of the infor-mation in three aspects models and the initial data for theexperiments the deployment mapping tables from the mod-els to the nodes and the plan of experiment executionWhenthe experiment plan is loaded by the EMT the file namesand model descriptions of models and data are listed Withthe support of cluster configuration models can be mappedto the nodes in the GUI according to the computationrequirements EMT also provides the setting of experimentsitself including the running times start time and end time
533 Models and Data Deployment Models are stored inthe agent population and services repository According tothe settings in experiment plan the customized models aredownloaded from the repository first then the models suchas agent model environment model emergent events modeland intervention measures model are integrated to be thecompositionmodels of artificial BeijingThemodels and datalocations in the nodes are configured in the ldquomodel pathrdquoand ldquodata pathrdquo in the GUI of central controller Finallythe models are uploaded to the nodes in cluster with theconsideration of partition The partition of artificial society
Mathematical Problems in Engineering 21
models is different from the multiagent system in [52]Models in artificial society are divided into three levels agentsand environments grids and city Agents and environmentsare encapsulated in grids which compose the whole cityThus the partition of artificial society becomes the partitionof grids The parallelism degrees are found to support thepartition of models agent independence and grid indepen-dence According to the parallelisms agents are partitioned toLPs according to the population distribution in grids Agentmovements among grids are simulated by the movementevents among grids in different LPs As a result a two-tierparallel architecture is proposed to support the simulation ofartificial society The architecture is detailed in [53]
534 Running of Experiments When the preparation isready the experiments can be done by the central controllerof EMT The running and control commands of the experi-ments are grouped into three aspects start pause or stop theexperiments get the running information of the nodes suchas the snapshot of the desktop of node With the help of thesecommands users can run the computational experimentsto do the research on the emergency management It isworth notifying that the simulation engine is optimized inorder to support the ten million agent simulation UsingGPU as coprocessor a two-tier parallel simulation engine isdesigned with support of MPI and OpenCL through phasedsynchronization [54] One-sided communication is used forreflection of remote simulation objects and message passingbetween processes A general kernel function prototype iselaborately designed and conditionally compiled for execu-tion on both CPU and GPU Moreover the optimizationoperations like load balance is developed under the instruc-tion of activity-based simulation [55] The densely populatedgrids are given more CPU and memory resource accordingto the activity predictions
54 Results of H1N1 Experiments with KD-ACP On thebasis of artificial society models and the intervention mea-sures mentioned in Section 512 a series of computationalexperiments are performed to study the H1N1 influenza inBeijing In our case of artificial Beijing 19610000 agentsare simulated in the cluster within 48 CPU cores and 128Gmemories the nodes are connected by kilomega networks Ittakes 18 hours to simulate a 250-day disease spreading
In order to illustrate how the optimal plans are obtainedbyKD-ACP the experiments are divided into four groups theexperiments ofmodel validation the experiments of sensitiv-ity analysis of vaccination rate the experiments of sensitivityanalysis of isolation and the experiments of the combinationof interventionmeasuresThese four groups show the generalprocess of the research on emergency management in H1N1influenza Firstly the models of H1N1 are validated withthe support of historical data Secondly the traditionalmedical interventions of pandemic such as vaccination areanalyzed to find the optimal plan Thirdly the traditionalnonmedical interventions of pandemic such as isolation areanalyzed to find the optimal plan Finally the combination ofinterventions is tested to show the combined effect
It is worth mentioning that all the experiments in ourwork are performed 100 times These experiments are ini-tialized with different settings of interventions in IMCT Theresults shown in the figures are the mean values
541 The Experiments of Model Validation In our artifi-cial Beijing the models of H1N1 influenza are built basedon the existing researches Wang and her colleagues hadstudied the H1N1 influenza by SEIR (SusceptibleExposedInfectiveRecovered) models [56] With the support of theirwork we build the models of artificial Beijing in the mannerof multiagents
In Figure 23 the control group shown in red line isdrawn based on the realistic statistical data collected in H1N1influenza in Beijing in 2009 The influenza lasts more thansix months and more than 170000 people are infectedComparing with the control group the simulated resultswithout interventions are drawn in blue It is obvious that thesimulated case fits the control group well before 70 days Butthe ascending of control group is slowed after 70 days andthe peak value (17883) is much less than the simulated results(25135) It is because the vaccination is activated at day 67 incontrol group according to the intervention measures usedin 2009 The restrain by vaccination becomes effective eightdays later The change of infected number after vaccinationis shown by red spotted line The trend of control group isdifferent from the simulation data after 75 days
However the simulated results (blue line) in compu-tational experiments fit the control group (as shown inFigure 23) in the first 70 days when no intervention mea-sures are executed The key features such as effective basicreproduction number (119877
0) are also almost the same in both
cases Therefore it can be concluded that the models ofartificial society built in microview are validated by therealistic statistical data in influenza in Beijing in 2009
In this group of experiments no interventions are exe-cuted in order to validate the models Therefore the simu-lated results of this extreme case which cannot happen todaycould be obtained As shown by blue line the number ofinfected agents without interventions reaches peak value atday 91 with 25135 consisting of 8052 in symptomatic phaseand 28950 in recovery phase It is obvious that the totalnumber of infected agents grows slowly at the beginningtime Then the number increases quickly when the influenzabreaks and it reaches its maximum value with 383870 in theend This phenomenon can be explained as in the followingAt the beginning of influenza due to the limited numberof infected agents the propagation of the disease remainsslow but many agents become susceptible When the numberof infected agents increases rapidly the spreading becomessignificant However the number of infected agents decreasesafter day 91 It is because the disease spreading relies on thesocial networks and spatial contact networks [49] It meansthat the infected agents can only infect agents either in thesocial relationship networks or in the spatially contacted
542 The Experiments of Sensitivity Analysis of VaccinationRate According to Section 512 vaccination is a typical
22 Mathematical Problems in Engineering
0
Infe
cted
agen
ts
0 50 100 150 200 250
Statistical dataSimulation data
25000
20000
15000
10000
5000
Day d
Figure 23 The results with no intervention executed and the com-parison with realistic statistical data in Beijing in 2009
medical intervention in influenza Traditionally the effective-ness of vaccination can only be evaluated after the influenzaand the key settings such as vaccination rate are determinedby the experience It is always difficult for emergency man-agement organizations to find the most optimal vaccinationrate in advanceTherefore the sensitivity of vaccination rate isanalyzed in this group of experiments The settings are listedin Table 7 average results are shown in Figure 24
Vaccination rates are set linearly (10 30 50 and70) to test the changes of influenza The peak value ofinfected agents decreases from 25135 22392 15787 6907to 638 respectively It shows that the more the agents arevaccinated the smaller the peak value of the infected agentswill be In the case of 70 vaccination the total infected agentis only 13799 It means that the influenza in this case does nothappen due to the high vaccination rate
The change of peak values also shows an interestingphenomenon Vaccination rates increase linearly the peakvalues decrease slower than the ascending of vaccination ratesat beginning but the descending becomes more and morefaster with the linear increase of vaccination rate It meansthat the small scale of vaccination (less than 30) does notmake senseWith the increase of vaccination rate the numberof susceptible agents decreases significantly In sequencethe possibility of infections is lowered greatly Additionallyit could be found that the decrease of peak values comeslater than the increase of the vaccination rates This is quitereasonable according to Mei et alrsquos work in [57]
As a result it is concluded that high vaccination rate iseffective in the control of H1N1 influenza though costly
543 The Experiments of Sensitivity Analysis of IsolationAccording to Section 512 isolation is a typical nonmedicalintervention It is the most used intervention measure in
Table 7 The initial settings of experiments of vaccination
Parameters ValueAgent count 19610000Initial infected count 40Interventions activated time(119879Intervention)
0
Vaccination rate (119877Vaccination) 0 10 30 50 70
0
Infe
cted
agen
ts
Loading intervention of vaccination
No vaccination10 vaccination30 vaccination
50 vaccination70 vaccination
0 50 100 150 200 250
25000
20000
15000
10000
5000
Day d
Figure 24 The results of loading intervention of vaccination
the emergency management in influenza As mentionedbefore a problem also exists that the measure cannot bevalidated before the intervention Therefore the sensitivityof isolation is analyzed in this group of experiments It isworth mentioning that the isolation is usually combinedwith the hospital policy It means the infected individual issent to hospital meanwhile the people who contact withthe infected are all isolated As a result parameters delay tohospital and isolation duration are designed together in thesettings of experiment Delay to hospital stands for the periodfrom infection time to hospital time while Isolation durationrepresents the period when an agent cannot contact others ifhe has contact with an infected agent The settings of theseparameters are listed in Table 8 average results are shown inFigure 25
Firstly the parameter delay to hospital is simulated aloneto test the sensitivity The simulated results are shown insolid line in Figure 25 The peak value decreases from 6708to 2218 when the delay to hospital decreases from 15 daysto 1 day The half day change brings the 67 reductionMoreover the peak value almost approximates to zero inthe case of 05 days The influenza does not happen in thiscase It can be concluded that the earlier the infected agents
Mathematical Problems in Engineering 23
Table 8 The initial settings of experiments of isolation
Parameters ValueAgent count 19610000Initial infected count 40Interventions activated time (119879Intervention) 80Delay to hospital (119879HDelay) isolation duration (119879IDuration) (15 0) (1 0) (05 0) (15 3) (1 3) (05 3)
0 50 100 150 200 2500
1000
2000
3000
4000
5000
6000
7000
Infe
cted
agen
ts
Loading intervention of isolation
Delay to hospital = 15
Delay to hospital = 10
Delay to hospital = 05
Delay to hospital = 15 isolation duration = 3
Delay to hospital = 10 isolation duration = 3
Delay to hospital = 05 isolation duration = 3
Day d
Figure 25 The results of loading intervention of isolation
are sent to hospital the less the agents will be infectedSo the construction of effective mechanism to find infectedindividuals is important in the emergency management ofinfluenza
Secondly the parameter isolation duration is added tosimulate the isolation of interventionObviously the isolationmeasure decreases the infected agents greatly Compared tothe parameter delay to hospital of 15-day case the peak valuedecreases from 6708 to 362when the isolation is added in theinterventionThe reduction of 6346 agents comes from the 3-day isolation of agents who had contacted with the infectedagents However the decrease of parameter delay to hospitalcannot bring the similar reduction in the 3-day isolationcases According to the figure the peak value decreases from362 to 51 from 15-day case to 05-day case only 311 agents aresaved from infection As a result the sensitivity of simulatedresults does not change obviously with delay to hospital inthe case of isolation It can be concluded that isolation is amore effective measure Though it is not easy for emergencymanagement organizations to decrease the delay to hospitaldue to the current monitoring mechanism the influenza canstill be controlled by isolation
544 The Experiments of Combination of Intervention Mea-sures Isolation is not the only nonmedical intervention ininfluenza close workspace is another common measureWhen close workspace is activated the workspaces are allclosed agents can only stay home at the closed time Luckilyisolation and close workspace are independent from oneanother It is possible for emergency management organi-zations to use these two measures together So this groupof experiments is designed to obtain the effectiveness of thecombination of intervention measures The parameters areinitialized with different settings listed in Table 9
Parameters delay to hospital and isolation duration is setin the middle values (1 day 3 days) according to the lastsection According to the experience close duration is set asseven days It is known that close workspace is a really costintervention measure the activated time is a key parameterWith the fixed parameters mentioned before the changeof interventions activated time is analyzed in this group ofcomputational experimentsThe average results are shown inFigure 26
Compared with the simulated results of experimentsdiscussed above Figure 26 strongly illustrates that influenzais ceased when the intervention measures are combinedtogether The peak value of infected agents is 141 in the ldquo80thdays of close-workspace and isolation with 1 day delay tohospitalrdquo case while 25135 in ldquono interventionrdquo case 2218 inldquo1 day delay to hospitalrdquo case 203 in ldquo1 day delay to hospitaland isolationrdquo case The interventions bring a great reductionof infection the peak values decrease from ten thousandslevel to hundred level
On the other hand the peak value even decreases to 108when the execution time of close workspace is 10 days earlierThe date of maximum value of infected agents is advancedfrom the 135th day to the 108th day The whole circle of H1N1influenza is reducedThe results of this group of experimentsshow that intervention measures should be activated as soonas possible and the compositions of interventions are moreeffective in emergency management of H1N1 influenza
In summary based on the general process lots of exper-iments could be done by KD-ACP to study emergency man-agement in H1N1 influenza Many conclusions are obtainedbased on the analysis of the experiments in this sectionAccording to these conclusions themost optimal response inour case study is the case of ldquo70th days of close-workspace andisolation with 1 day delay to hospitalrdquo It is because this casehas the fewest infected agents relatively But the modelling ofH1N1 influenza is not sufficient Economic and social factorsof interventions such as the money cost by vaccination andisolation the influence of the close-workspace in society
24 Mathematical Problems in Engineering
Table 9 The initial settings of experiments of isolation and closeworkspace
Parameters ValueAgent count 19610000Initial infected count 40Interventions activated time (119879Intervention) 70th day 80th dayDelay to hospital (119879HDelay) 1 dayIsolation duration (119879IDuration) 3 daysClose workspace duration (119879CWDuration) 7 days
50 100 150 200 2500
50
100
150
200
Infe
cted
agen
ts
Loading intervention of isolation and close workspace
Delay to hospital = 1 isolation duration = 3Start close time = 70 close duration time = 7Delay to hospital = 1 isolation duration = 3Start close time = 80 close duration time = 7
Day d
Figure 26The results of loading interventions of isolation and closeworkspace
and even opinion transmission of H1N1 in internet are notconsidered in our case study Without the analysis of thesefactors it does not make sense to talk about the most optimalintervention measures However KD-ACP had provided asoftware framework and general process of the research onemergency management Supported by KD-ACP the expertsof emergency management are able to study emergent eventssuch as H1N1 influenza as detailed as possible Based on theexperiments and analysis again and again the most optimalintervention measures will be found ultimately
55 Parallel Execution by ASST and OsdRT As discussed inSection 35 ASST andOsdRT are used to support the parallelexecution of artificial Beijing The statistical data is collectedand stored in artificial society runtime database at runtimeASST downloads the infection data from the database anddisplays it in the map As shown in Figure 27 the infectedagents are distributed in the districts in BeijingThe snapshotgives the geospatial distribution of the H1N1 epidemic Thelegends of the agents in different status are listed in the left
right corner in Figure 27 Five statuses are shown in the figurehealth without vaccination health with vaccination incuba-tion being symptomatic and convalescence ASST providesa very direct feeling for the emergency experts Partly withthe situation from ASST Emergency Decision Organizationmakes the emergency response plans to the response of theemergency scenario With the help of plans the epidemicin actual society can be restrained In the meantime theinfluenza information is reflected in the Internet Accordingto Figure 8 the data acquisition data extraction and datastandardization of H1N1 influenza are processed in OsdRTOsdRT is implemented by Dr Caorsquos team from Institute ofAutomation from Chinese Academy of Sciences [58] Thetool provides not only the reports of H1N1 SARS hand-foot-mouth disease and intestinal diseases are all monitored Theweb page is shown in Figure 28 H1N1 reports from Internetare collected first Based on the data extraction of the reportsH1N1 case of distribution is obtained and shown in theincidencemap As a result H1N1 incidence data is regarded asthe knowledge which is sent to EECT and IMCTThe settingsof emergent events and intervention measures are modifiedin order to make artificial society approach actual society innext turn of computational experiments With the help ofOsdRT the loop of parallel execution process integrates ACPapproach
6 Conclusion
This work provides an integrated software framework forsocial Computing in emergencymanagement From a systemperspective KD-ACP provides a reliable flexible low costand effective platform for the scientists to do the research onemergency response problems The analysis and predictionof these problems with inherent complexity can be solvedby the repetition of possible alternative experiments onthe software framework Therefore KD-ACP which is anattempt to implement ACP approach seemed as an effectivecomputational framework to support the decision making inemergency response
Currently KD-ACP is used to study the H1N1 epidemicin Beijing in 2009 The most optimal compositions of inter-vention measures are testified to support the response tonext influenza However it is still a long way ahead beforethe automation systematization and practicalization of KD-ACP A lot of work will be carried out along several directionsas in the following
(i) The parallel execution process in KD-ACP is actuallynot strongly connected with the emergency responseorganization Traditional experience and theory stillplay a dominant role in the decision of emergencymanagement It is necessary to propose a mechanismof hall for workshop to facilitate the parallel executionloop in actual society
(ii) Section 33 introduces the principle and process ofmodelling artificial society Metamodels and modelsof artificial Beijing are built by GME Model transfor-mation and code generation are used to implementcodes from FSM based models But many details
Mathematical Problems in Engineering 25
Fangshan district Daxing district Fengtai district Xicheng district
Mentougou district Haidian districtShijingshan district Dongcheng district
Incubation
Symptomatic
Convalescence
Health withoutvaccination
Health withvaccinationLegend of agent
health status
Figure 27 The situation of artificial Beijing by ASST
H1N1
Districts of Beijing
H1N1 case ofdistribution
Incidence map
Distribution ofH1N1 reportsfrom internet
Incidence of H1N1
Fangshan districtLiangxiang town
Population 19000Incidence 135789
(1100000)
Figure 28 The implementation of OsdRT by Institute of Automation from Chinese Academy of Sciences
are still not illustrated due to the topic of thispaper The key techniques such as template of codeframework and model transformations from specificdomains to FSM will be focused and introduced indetail in our next paper
(iii) The performance of computational experiments inKD-ACP is still slower than expectation The opti-mization of simulation engine and models structureis necessary to improve the performance So it ispossible to run artificial society with the real worldin parallel
In summary KD-ACP paves a new way for scientists inboth emergencymanagement and computation simulation to
collaborate with each other in solving the emergency man-agement problems by social computing
Research Highlights
(i) KD-ACP is implemented based on ACP approach
(ii) Artificial Beijing is built with the help of KD-ACP
(iii) H1N1 is modeled to simulate the influenza in Beijingin 2009
(iv) Computational experiments testify the effectivenessof intervention measures
26 Mathematical Problems in Engineering
Conflict of Interests
The authors declare that there is no conflict of interestsregarding the publication of this paper
Acknowledgment
This work was supported in part by the National ScienceFoundation of China (nos 91024030 71303252 61403402and 91324013)
References
[1] W Mao and F-Y Wang Advances in Intelligence and SecurityInformatics Academic Press Oxford UK 2012
[2] E Bonabeau ldquoAgent-based modeling methods and techniquesfor simulating human systemsrdquo Proceedings of the NationalAcademy of Sciences of the United States of America vol 99 no3 pp 7280ndash7287 2002
[3] K M Carley D B Fridsma E Casman et al ldquoBioWar scalableagent-basedmodel of bioattacksrdquo IEEETransactions on SystemsMan and Cybernetics A Systems and Humans vol 36 no 2 pp252ndash265 2006
[4] J M Epstein ldquoModelling to contain pandemicsrdquo Nature vol460 no 7256 p 687 2009
[5] Los Alamos National Laboratory Reports about EPISIMS LosAlamos National Laboratory 2003
[6] C H Builder and S C Bankes ldquoArtificial societies a conceptfor basic research on the societal impacts of informationtechnologyrdquo RAND Report 149 RAND Corp 1991
[7] F-Y Wang ldquoComputational theory and method on complexsystemrdquo China Basic Science vol 6 no 5 pp 3ndash10 2004
[8] F-Y Wang ldquoToward a paradigm shift in social computing theACP approachrdquo IEEE Intelligent Systems vol 22 no 5 pp 65ndash67 2007
[9] F-Y Wang and S Tang ldquoArtificial societies for integratedand sustainable development of metropolitan systemsrdquo IEEEIntelligent Systems vol 19 no 4 pp 82ndash87 2004
[10] D Wen Y Yuan and X-R Li ldquoArtificial societies computa-tional experiments and parallel systems an investigation ona computational theory for complex socioeconomic systemsrdquoIEEE Transactions on Services Computing vol 6 no 2 pp 177ndash185 2013
[11] F Wang F Zeng and Y Yuan ldquoAn ACP-based approach forcomplexity analysis of E-commerce systemrdquo Complex Systemsand Complexity Science vol 5 no 3 pp 1ndash8 2008 (Chinese)
[12] F-Y Wang and P K Wong ldquoResearch commentary intelligentsystems and technology for integrative and predictivemedicinean ACP approachrdquoACMTransactions on Intelligent Systems andTechnology vol 4 no 2 pp 1ndash6 2013
[13] J Sifeng X Gang F Dong and H Chunpeng ldquoStudy on theemergency rescue decision support system of petrochemicalplant based on ACP theoryrdquo in Proceedings of the 6th Manage-ment Annual Meeting (MAM rsquo11) 2011 (Chinese)
[14] X Dong D Fan G Xiong F Zhu Z Zhang and Y Yao ldquoPar-allel Bus Rapid Transit (BRT) operation management systembased on ACP approachrdquo in Proceedings of the 6th ManagementAnnual Meeting (MAM rsquo11) Beijing China 2011
[15] B Ning F-Y Wang H-R Dong R-M Li D Wen and L LildquoParallel systems for Urban rail transportation based on ACP
approachrdquo Journal of Transportation Systems Engineering andInformation Technology vol 10 no 6 pp 23ndash28 2010 (Chinese)
[16] F-Y Wang J Zhao and S-X Lun ldquoArtificial power systemsfor the operation and management of complex power gridsrdquoSouthern Power System Technology vol 2 no 3 pp 1ndash11 2008(Chinese)
[17] M J North T R Howe N T Collier and J R Vos ldquoRepastsimphony development environmentrdquo in Proceedings of theAgent 2005 Conference on Generative Social Processes Modelsand Mechanisms 2005
[18] M J North T RHowe N T Collier and J R Vos ldquoRepast Sim-phony runtime systemrdquo in Proceedings of the Agent Conferenceon Generative Social Processes Models and Mechanisms 2005
[19] SWARM Development Group Swarm 211 Reference Guide2000 httpwwwswarmorg
[20] S Tisue and U Wilensky ldquoNetLogo a simple environment formodeling complexityrdquo in International Conference on ComplexSystems pp 16ndash21 2004
[21] S Luke C Cioffi-Revilla L Panait K Sullivan and G BalanldquoMASON a multiagent simulation environmentrdquo Simulationvol 81 no 7 pp 517ndash527 2005
[22] M Ichikawa H Tanuma Y Koyama andH Deguchi ldquoSOARSintroduction as a social microscope for simulations of socialinteractions and gamingrdquo in Organizing and Learning throughGaming and Simulation Proceedings of the 38th Conference ofthe International Simulation and Gaming Association (ISAGArsquo07) pp 149ndash158 2007
[23] D S Burke J M Epstein D A T Cummings et al ldquoIndividual-based computational modeling of smallpox epidemic controlstrategiesrdquo Academic Emergency Medicine vol 13 no 11 pp1142ndash1149 2006
[24] F-Y Wang ldquoParallel control and management for intelligenttransportation systems concepts architectures and applica-tionsrdquo IEEE Transactions on Intelligent Transportation Systemsvol 11 no 3 pp 630ndash638 2010
[25] Z Molnar D Balasubramanian and A Ledeczi ldquoAn introduc-tion to the genericmodeling environmentrdquo in Proceedings of theTOOLS Europe 2007 Workshop on Model-Driven DevelopmentTool Implementers Forum Zurich Switzerland 2007
[26] S Kelly and J-P Tolvanen Domain Framework in Domain-Specific Modeling Enabling Full Code Generation JohnWiley ampSons Hoboken NJ USA 2007
[27] B P Zeigler H Praehofer and T G Kim Theory of Modelingand Simulation Integrating Discrete Event and ContinuousComplex Dynamic Systems Academic Press San Diego CalifUSA 2000
[28] G Booch J Rumbaugh and I JacobsonThe Unified ModelingLanguage User Guide Addison-Wesley Reading Mass USA1999
[29] F Wagner R Schmuki T Wagne and P Wolstenholme Mod-eling Software with Finite State Machines A Practical ApproachCRC amp Taylor amp Francis Boca Raton Fla USA 2006
[30] T Espiner ldquoChina builds worldrsquos fastest supercomputerrdquo ZDNetUK October 2010
[31] B Chen and G Guo ldquoA two-tier parallel architecture for arti-ficial society simulationrdquo in Proceedings of the ACMIEEESCS26th Workshop on Principles of Advanced and DistributedSimulation (PADS rsquo12) pp 184ndash186 July 2012
[32] G Guo B Chen X G Qiu and Z Li ldquoParallel simulation oflarge-scale artificial society on CPUGPU mixed architecturerdquo
Mathematical Problems in Engineering 27
in Proceedings of the ACMIEEESCS 26th Workshop on Prin-ciples of Advanced and Distributed Simulation (PADS rsquo12) pp174ndash177 July 2012
[33] S-X Lun ldquoResearch on the classification of parallel executionmodes of ACP theoryrdquo Acta Automatica Sinica vol 38 no 10pp 1602ndash1608 2012
[34] F-YWang ldquoA framework for social signal processing and anal-ysis from social sensing networks to computational dialecticalanalyticsrdquo Science China vol 43 no 12 pp 1598ndash1611 2013
[35] G GuoOne Model Userrsquos Manual version 12 National Univer-sity of Defense University 2011
[36] Y Ge R Meng Z Cao X Qiu and K Huang ldquoVirtual citymdashanindividual-based digital environment for human mobility andinteractive behaviorrdquo Simulation Transactions of the Society forthe Modeling and Simulation International 2014
[37] M Kretzschmar and R T Mikolajczyk ldquoContact profiles ineight European countries and implications for modelling thespread of airborne infectious diseasesrdquo PLoS ONE vol 4 no6 Article ID e5931 2009
[38] W Duan and X Qiu ldquoFostering artificial societies using sociallearning and social control in parallel emergency managementsystemsrdquo Frontiers in Computer Science vol 6 no 5 pp 604ndash610 2012
[39] W J Edmunds C J OrsquoCallaghan and D J Nokes ldquoWhomixes with whom Amethod to determine the contact patternsof adults that may lead to the spread of airborne infectionsrdquoProceedings of the Royal Society B Biological Sciences vol 264no 1384 pp 949ndash957 1997
[40] J Banks J S Carson B L Neison and D M Nicol Discrete-Event System Simulation Prentice Hall Englewood Cliffs NJUSA 4th edition 2007
[41] H Yongxia ldquoInvestigation and analysis of college studentsrsquo workand restrdquo Value Engineering vol 31 no 3 2012 (Chinese)
[42] W Duan Z Cao Y Wang et al ldquoAn ACP approach to publichealth emergency management using a campus outbreak ofh1n1 influenza as a case studyrdquo IEEE Transactions on SystemsMan and Cybernetics Part A Systems and Humans vol 43 no5 pp 1028ndash1041 2013
[43] HWHethcote ldquoThemathematics of infectious diseasesrdquo SIAMReview vol 42 no 4 pp 599ndash653 2000
[44] G Chowell E Shim F Brauer and et al ldquoModelling thetransmission dynamics of acute haemorrhagic conjunctivitisapplication to the 2003 outbreak in Mexicordquo Statistics inMedicine vol 25 no 11 pp 1840ndash1857 2006
[45] Y Ge L Liu B Chen X Qiu and K Huang ldquoAgent-basedmodeling for InfluenzaH1N1 in an artificial classroomrdquo SystemsEngineering Procedia vol 2 pp 94ndash104 2011
[46] A R Tuite A L Greer MWhelan et al ldquoEstimated epidemio-logic parameters andmorbidity associatedwith pandemicH1N1influenzardquo CMAJ vol 182 no 2 pp 131ndash136 2010
[47] L Brouwers B Cakici M Camitz A Tegnell and MBoman ldquoEconomic consequences to society of pandemic H1N1influenza 2009mdashpreliminary results for Swedenrdquo Euro Surveil-lance vol 14 no 37 pp 1ndash7 2009
[48] Y Ge L Liu X Qiu H Song Y Wang and K Huang ldquoAframework of multilayer social networks for communicationbehavior with agent-based modelingrdquo Simulation Transactionsof the Society for the Modeling and Simulation International vol89 no 7 pp 810ndash828 2013
[49] W Duan X Qiu Z Cao X Zheng and K Cui ldquoHeterogeneousand stochastic agent-based models for analyzing infectious
diseasesrsquo super spreadersrdquo IEEE Intelligent Systems vol 28 no4 pp 18ndash25 2013
[50] The Preparation and Emergency Response Plans to PandemicPublished by Ministry of Health Ministry of Health 2005(Chinese)
[51] R M Fujimoto Parallel and Distributed Simulation SystemsWiley Interscience 2000
[52] F Cicirelli A Giordano and L Nigro ldquoDistributed simulationof situated multi-agent systemsrdquo in Proceedings of the 15thIEEEACM International Symposium on Distributed Simulationand Real Time Applications (DS-RT rsquo11) pp 28ndash35 September2011
[53] B Chen and G Guo ldquoA two-tier parallel architecturefor artificial society simulationrdquo in Proceedings of the 26thACMIEEESCS Workshop on Principles of Advanced and Dis-tributed Simulation (PADS rsquo12) pp 184ndash186 July 2012
[54] G Guo B Chen and X Qiu ldquoParallel simulation of large-scale artificial society with GPU as coprocessorrdquo InternationalJournal of Modeling Simulation and Scientific Computing vol4 no 2 Article ID 1350005 2013
[55] B Chen L-B Zhang X-C Liu and H Vangheluwe ldquoActivity-based simulation using DEVS increasing performance by anactivity model in parallel DEVS simulationrdquo Journal of ZhejiangUniversity Science C vol 15 no 1 pp 13ndash30 2014
[56] X-L Wang P Yang Z-D Cao et al ldquoQuantitative evaluationon the effectiveness of prevention and control measures againstpandemic influenza A (H1N1) in Beijing 2009rdquoChinese Journalof Epidemiology vol 31 no 12 pp 1374ndash1378 2010 (Chinese)
[57] S Mei A D van de Vijver L Xuan Y Zhu and P M ASloot ldquoQuantitatively evaluating interventions in the influenzaA (H1N1) epidemic on China campus grounded on individual-based simulationsrdquo Procedia Computer Science vol 1 no 1 pp1675ndash1682 2010
[58] X Qiu Z Cao and Z Li Mid-Term Reports of China NationalScience Foundation 91024030 2014
Submit your manuscripts athttpwwwhindawicom
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
MathematicsJournal of
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Mathematical Problems in Engineering
Hindawi Publishing Corporationhttpwwwhindawicom
Differential EquationsInternational Journal of
Volume 2014
Applied MathematicsJournal of
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Probability and StatisticsHindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Journal of
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Mathematical PhysicsAdvances in
Complex AnalysisJournal of
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
OptimizationJournal of
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
CombinatoricsHindawi Publishing Corporationhttpwwwhindawicom Volume 2014
International Journal of
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Operations ResearchAdvances in
Journal of
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Function Spaces
Abstract and Applied AnalysisHindawi Publishing Corporationhttpwwwhindawicom Volume 2014
International Journal of Mathematics and Mathematical Sciences
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
The Scientific World JournalHindawi Publishing Corporation httpwwwhindawicom Volume 2014
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Algebra
Discrete Dynamics in Nature and Society
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Decision SciencesAdvances in
Discrete MathematicsJournal of
Hindawi Publishing Corporationhttpwwwhindawicom
Volume 2014 Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Stochastic AnalysisInternational Journal of
6 Mathematical Problems in Engineering
M2T transformation
Modeling in domain
GME Generic Modeling Environment
Metamodeling
Instance of
FSM
Metamodeling for artificial society
DEVS
Petri Net State charts
Agentmetamodel
Environmentmetamodel
Emergency eventmetamodel
Interventionmetamodel
DAE
Domain-specific metamodels Semantic well definedmetamodels
M2M transformation
Modeling of artificial society
Interventionmodel
Emergency eventmodel
Environmentmodel
Agentmodel
FSM based emergent andintervention models
FSM based environmentmodels
FSM based agent modelsDomain-specific models
Semantic well defined models
M2M transformation
Agent model code Environment model code
Agent behavior model
Agent mentality model
Disease model code
Basic population model Disease spreading model
Disease state transformmodel
Intervention model code
Agent social networkmodel
Basic environmentinformation model
Roadnet model
Vaccination
Infectors isolation
Contactors isolation
Close work officesAgent contact model
Domain-specific models relevant codes
MDT Model Development Tool
Implementation
Figure 4 From metamodelling and modelling by GME to implementation by MDT of artificial society
331 The Principle of the Modelling of Artificial SocietyAccording to the principle of domain-specific modellingthe modelling of artificial society contains the followingsteps first metamodelling the basic elements of artificialsociety second modelling the specific features in domain ofemergency management third implementing the models ofartificial society in codes The whole process is illustrated inFigure 4The first and second steps are implemented in GMEwhile the third step is implemented in MDT
The first step is metamodelling which mainly focuses onconstructing the metamodels of artificial society Metamod-elling tries to study the common patterns of artificial societyThe outputs of metamodelling are metamodels which repre-sents the abstraction of the whole systemThe basic elementsof artificial society are described in metamodel The processof metamodelling is divided into four phases The first is theconstruction of the domain-specific metamodels As shown
in Figure 4 agent metamodel environment metamodelemergent event metamodel and intervention metamodelcompose the metamodel of artificial society The second isthe construction of the metamodels described by typicalmodelling formalisms such as FSM DAE DEVS and PetriNet [27] These formalisms are all semantically well definedThe third is the definition of the model transformationfrom domain-specific metamodels to metamodels of typicalmodelling formalisms The transformation standardizes themetamodels of artificial society by typical modelling spec-ifications The fourth is the definition of the transforma-tion templates from metamodels to code framework Thetemplates list the basic abstract interfaces of metamodels ofartificial societyThese abstract interfaces are implemented inthe specific-domain modelling and code generations
The second step is modelling the models of artificialsociety such as agent model environment model emergent
Mathematical Problems in Engineering 7
Figure 5 The metamodels of artificial society in GME
event model and intervention model are built Actuallythe models are the instantiation of metamodels in the laststep Different from the general modelling environment likeUML [28] the domain-specific modelling provides a familiarmodelling environment for the domain experts in artificialsociety For example emergency response experts only con-cern emergent eventmodel and interventionmodel inheritedfrom metamodels After constructing the domain-specificmodels based on domain-specific metamodels domain usersexecute the model transformation defined in the first stepAll the models of artificial society are transformed into FSMmodels As a result the models are implemented in this uni-fied modelling formalism (FSM) The model transformationmakes the simulation of the models possible
The third step is the generation of executable codes ofmodels The executable code framework is generated bymapping template from metamodels to code frameworkdefined in the first step Moreover domain developers alsoadd necessary codes to the framework to integrate thedynamic semantics of the models The code frameworkoutputs the dynamic link libraries by compilingThe dynamiclink libraries are loaded in the large scale artificial societyruntime infrastructure in computational experiments
332 The Metamodelling and Modelling of Artificial Societyby GME GME is used to build metamodels and modelsin our work As mentioned before the abstraction andcommon patterns of society are represented in metamodelsAccording to the bottom-up modelling style metamodels ofagent environment and communications are described inGME Figure 5 shows part of metamodels of artificial societyThe features of an agent metamodel are extracted from thecensus figures and statistical data Environment metamodelsimulates the geospatial places for the behaviors of agentsThemetamodel of communications among agents ismodeledto simulate the interactions such as infection in epidemicsand rumor propagation in public opinion formation eventsIt is worth notifying that the metamodel of communications
includes both the emergent event metamodel and interven-tion metamodel
From the perspective of modelling the details from spe-cific domains are considered in the models by the instantia-tion frommetamodels of artificial society For example socialrelationships based on complex networks are added in agentmodel to support the communications Agent activity is alsoused to quantify the agent activity under different scenarioEnvironment models are linked with the help of transporta-tion services subways and roads are modeled while the pathsearch is encapsulated in the services Emergent event modeland intervention model are also the domain-specific modelsThe modelling of artificial Beijing in GME will be discussedin detail in next section
333 The Implementation of Models of Artificial Society byMDT As mentioned before MDT is used to implementmodels such as agent environment emergent event andintervention According to the template of code frameworkthe implementations of models are generated by MDT Theimplementations are classified into two categories FSMmodels and services FSM models such as agents andenvironments are built under the specification of FiniteState Machine (FSM) [29] in MDT while all the servicessuch as transportation are encapsulated under the PublicService Standard This standard provides a generic interfacespecification for modelers to encapsulate public commonservices in artificial society FSM models like agent are builtstatistically from the quantitatively analyzed characteristicssuch as demographic attributes social behaviors emergencybehaviors and social networks Social behaviors describe thedaily behaviors of individuals while emergency behaviorsdescribe the individual behaviors in emergent events Forexample infected individuals are all isolated in hospital inSARS Isolation ismodeled as a typical emergency behavior inour work Correspondingly services are used to simulate themacroactual society Take transportation service for instancethe path search is needed by almost every agent duringmoving from spot to spot
8 Mathematical Problems in Engineering
Artificial society description in emergency
Description ofemergency
ArtificialSociety Editor
(ASE)
Agent behavior set
Agent role setAgent status set
Environment setAgent relationship set
Statistical data of real society
Descriptionof artificial
societyframework
Terminal of agent modeling andartificial society describing
middot middot middotAgent
metamodeling
Customized agentmodeling
Relationship modelDailyEmergency
Mod
el D
evelo
pmen
t Too
l (M
DT)
Agent modeling
Agent model Encapsulate
Encapsulate
Behavior model
Intervention
Emergency
Response
Daily
Social relationshipmodeling
Socialbehaviormodeling
Socialbehaviormodel
Emergencybehavior
model
Emergencybehaviormodeling
Building
Geospatial statistical data
Population statistical data Datacollection
Actualsociety
Artificial societypopulation and
geospatial database
Organizationservice
Transportationservice+
MappingAgent model andservice repository+ emergency model
repository
Climateservice
Geographicalservice
+
Agent status in emergencyEmergency management
Emergent event set
Social relationship statistical data
Environment statistical data
Population behaviorstatistical data
Emergency statistical data
Emergency organization statistical data
Geospatial and SocialEnvironment Generation
Tool (GSeT)
Agent model repository andpopulation database
C++C++C++C++
Modeling
Statistical modelJob ageSex location
Public service modeling
Organizationmodel
Transportationmodel
Geographicalmodel
Climatemodel
Interventionservice
Publ
ic se
rvic
e sta
ndar
d
Figure 6 The editing and initialization of KD-ACP
MDT provides domain experts with a tool to obtain thecode implementations of models With the help of compileenvironment like visual studio MDT also supports thefurther programming development of the specific domaindetails which cannot be described in modelling step
Both the FSM models and services are developed by theMDT first and then stored in the agent model and servicerepository The repository manages the models according tothe requirements from emergency problems and provides themodels for EDT to make the experiment plans
34 The Editing and Initialization of Artificial Society byASE and PGET As shown in Figure 6 ASE is used to editthe scenarios of artificial society within emergent eventsThe editing is composed of two parts (1) the statisticalinformation of artificial society in daily life such as the rolesof agents the relationships of agents and the types of environ-ments and (2) the statistical information of artificial societyin emergency including the statistical data of emergent
events the emergency organization and emergency relatedbehaviors of agents
According to the requirements of the editing thesestatistical data are collected from actual society manually bythe domain experts Based on these statistical data PGETgenerates the artificial society population and geospatialenvironment database The database supports the instanti-ation of artificial society at individual level For examplethe attributes such as age and gender of each agent can befound in the database With the support of the databaseFSM models and service repository discussed before itis sufficient for domain experts to study the emergencyproblems by computational experiments
35The Computational Experiments and Parallel Execution inKD-ACP The tools of ldquoCrdquo part and ldquoPrdquo part in KD-ACP areused to support the process of computational experimentsand parallel execution The working process is shown inFigure 7 EECT and IMCT are both the starting and ending
Mathematical Problems in Engineering 9
Agent model andservice repository
repository
Emergency EventsConfigurationTool (EECT)
Open Source DataRegistration Tool (OsdRT)
Intervention MeasuresConfigurationTool (IMCT)
Load
Output Output
Internet
Emergencyresponse plans
Load
∙ Input models of emergencyevents∙ Output settings of agentbehavior in emergent events+ settings of emergent events
∙ Input models of interventionmeasures∙ Output settings of agentbehavior in interventionmeasures + settings ofintervention measures
+ emergency model
∙ Input data from internetnetworks∙ Output population data +emergency events data +intervention measures data
Artificial societypopulation and
geospatialenvironment database
Artificial societyruntime database
OutputThe most optimal
emergencyresponse plan
Actual society
OutputOutputOutput
Input
Load
Load
Load
Load
Output
Artificial SocietySituation Tool (ASST)
Emergency decisionorganization
ClusterRuntime infrastructure
Deploy
Deploy
Experimentsdesign tool (EDT)
Artificial societyinitial data file
Experimentplan
Artificial societymodels
Experiments Management
large scale artificial societyruntime infrastructure
Tool(EMT) +super computerTIANHE-1A
Load
∙ Input experiment plan +artificial society models +artificial society initial data file∙ Output artificial societyruntime state data
∙ Input description of artificialsociety + artificial society populationand geospatial environment society+ emergent events configuration +
intervention measures configuration∙ Output experiment plan + artificialsociety models + artificial societyinitial data file
∙ Input runtime statistical data∙ Output graphics charts ofartificial society statistical data +situation of artificial society
Figure 7 The computational experiments and parallel execution of KD-ACP
point The emergent events and intervention measures areconfigured by EECT and IMCT respectively The configu-rations of emergent events are used to simulate both thereal emergencies like SARS and H1N1 and the supposedemergencies for experiments Similarly the configurationsof intervention measures are also used to reproduce thereal one and simulate the supposed one The repeat of theemergency is used to verify the models while the supposedconfigurations are used to obtain the optimized decision planto the response of the possible emergencies
With the input of artificial society model and servicerepository artificial society population and geospatial envi-ronment database and the configurations discussed beforeEDT generates the experiment plans to meet the require-ments of research on emergency management The outputof EDT includes artificial society models artificial societyinitial data files and experiment plan The models aredownloaded from the repository while the data file is thecollection of data from the database to initialize the modelsWhen the models and data files are ready EMT loads theexperiment plan and deploys the models and data to thecluster or TIANHE-1A supercomputer [30] which was theworldrsquos fastest supercomputer built by National University of
Defense Technology (NUDT) in China in 2010 According tothe plan the experiments are done repeatedly on the largescale artificial society runtime infrastructure [31 32] by themultisample settingsThework process is the implementationof computational experiments in ACP approach
Traditionally emergency response plans are made byemergency management theories and experiences The onlyway to test the effective of plans is the feedback results of realworld ACP approach provides a novel method to supportemergency response plan making by parallel execution Asshown in Figure 7 the work process of KD-ACP is composedof two loops The inner loop composed by red arrowsdescribes the process of computational experiments while theouter loop of yellow arrows illustrates the process of parallelexecution During the runtime of computational experi-ments the statistical data of artificial society is collected andstored in the artificial society runtime database Based on thedatabase ASST outputs the customized situation of runningartificial society by graphics charts and situation mapsThe information is sent to the organizations of emergencydecision to support making the emergency response plansWith the help of computational experiments loop these plansare simulated repeatedly to find the most optimal one
10 Mathematical Problems in Engineering
Open Source Data Registration Tool (OsdRT)
Emergent EventsConfiguration Tool (EECT)
Intervention MeasuresConfiguration Tool (IMCT)
∙ Web information collecting
∙ Deep mining in social media
∙ Web information denoising
∙ Information filtering
Data acquisition
Actual society
Open source informationin social media
Internet
∙ Basic element extracting∙ Individuals and organizationsextraction∙ Sentiment analyzing andopinion mining∙ Social network analyzing
Data extraction
∙ Domain knowledge description∙ Information fusion and collisiondetection∙ Domain knowledge construction∙ Social media informationstandardization
Data standardization
Knowledge
Figure 8 The compositions of OsdRT [34]
Moreover the most optimal plan is used to the responseof emergency in actual society According to the idea ofparallel control in [24] the feedback of actual society ispartly collected from Internet networks by OsdRT As shownin Figure 8 OsdRT is composed of three components dataacquisition data extraction and data standardization Dataacquisition collects mines and filters information fromsocial sensing networks Data extraction includes basicelement extracting individual and organization extractingsentiment analyzing and social networks analyzing Datastandardization specifies the useful knowledge and sent it toconfiguration tools in KD-ACP
By processing in OsdRT the knowledge about emergentevents and intervention measures are analyzed first andregistered in the EECT and IMCT The registration updatesthe settings of emergent events and intervention measuresThis loop composed of yellow arrows implements the parallelexecution in ACP approach The implementation of parallelcontrol and management provides a data-driven approachthat considers both the engineering and social complexityfor modelling analysis and decision making in emergencymanagement
4 Modelling Beijing City with KD-ACP
41 How to Build the Artificial Beijing According to themodelling of artificial society in KD-ACP discussed beforethe Beijing city is modeled as follows
To meet the requirements of emergency managementsthe basic elements of artificial society are extended Asshown in Figure 9 six elements are required to simulate thecity agents environments transportation activity schedulecommunication and agent activity
411 Modelling Artificial Beijing Figure 10 shows the mainGUI of GME for the modelling of artificial modelling inpublic health events Metamodels are listed in the left area ofFigure 10 the list provides basic syntax elements for domainexperts to model artificial society Domain experts build
models based on the knowledge of their own MeanwhileGME supports hierarchy for building large scale systemsThesyntax symbol listed in GUI can be extended in new tab bydouble clicking Take agent for example the model of agentcan be detailed by edition in another tab page of agent
As shown in the center of Figure 10 the models ofartificial Beijing consist of five parts models of agent andenvironment domain models of public health events inter-ventionmodels controllermodels and services Agentmodeldescribes individuals in society it is composed of basicpopulation information action social relationships activityschedule and disease related information Activity schedulerepresents individualrsquos physical actionmodel focusing on thedaily action of agents Environment model includes physicalentities such as buildings playground transportations andagents contained in environment Domain models of publichealth events are composed of the propagation model of dis-ease disease state transition model and so on Interventionmodels include the settings of vaccination isolation and soon The models mentioned before are all FSM models Themechanisms of these models will be detailed in next sectionsController models and services are the public service mod-ules they are implemented in the development in MDT
412 Modelling Agents and Environments Under the speci-fication of FSM agent and environment models are imple-mented in two parts the state space and state transitionsThe state space is composed of the demographic attributesand behavior related attributes The transitions are triggeredwhen the conditions of states are satisfied As shown inFigure 11(a) the action of agent is changed when the ldquonexttimerdquo condition is satisfied in agent model while the agentslist is changed when the agent arrival condition is satisfied inthe environment model
413 Modelling Activities Agent activities come from theagent state transitions of actions such as movements andcommunications The actions of agents are instructed by theactivity schedule shown in Table 2 Activity schedule lists
Mathematical Problems in Engineering 11
Transportation EnvironmentsEnvironments
Agents
Artificial Beijing
Agent activityActivity schedule
Agent activity
Communication by socialrelationship networks
Communication by socialrelationship networks
Figure 9 The basic elements of artificial society
Syntax elements of DSM
Services
Models of agentenvironment andactivity schedule
Domain models of publichealth events
Intervention models
Modelingaspect ofdomainexperts
Controllermodels
Figure 10 The models of artificial Beijing in GME
all the actions with probability in one day for agents inboth normal and emergent situation [38] There are severaltypes of activity schedule in artificial Beijing student agentactivity schedule worker agent activity schedule emergentagent schedule and so on For example Table 2 gives an agentactivity schedule Upon the instruction of activity schedulestudent agent changes the actions by 119901
119894after state transitions
The119901119894in the tablemeans the action probability in the relevant
period In the duration from 0800 to 1200 a student agenteither goes to classroom to have class or goes to libraryto study The probability of class action is 119901
2while the
probability of study action is 1 minus 1199012
Agent behaviors are decided by the settings of activityschedule In addition to the daily activity schedulementionedbefore emergent activity schedules are also considered inour work Take public health events for instance an infectedagent changes schedule from a normal one to emergentone The workflow of a susceptible agent is illustratedin Figure 12 to show the change of behaviors After theinfection the agent is set in incubation phase Not all theincubation agents will become symptomatic Some of themturn back to being susceptible and some of them becomesymptomatic The symptomatic agents change their activityschedule from normal to emergent In the emergent case
12 Mathematical Problems in Engineering
Next actionNext environment (location)
Advance to next actiontime
Find next environment inactivity chain by action
Current actionCurrent environment (location)
Q
q0
120575 Find next action inactivity schedule by time
F
Σ
(a) The state transitions of agent
Q
q0
120575 F
Current agent list
Current pollution level
Next agents list
Next pollution level
Agents entered
Infected agents entered
Add agents to currentagents list
Compute pollution levelby infected agents
Σ
(b) The state transitions of environment
Figure 11 The state transitions of models
Table 2 The activity schedule of a student agent
Duration (Δ119905) Activity Location Probability 119879Location (minute)0000ndash0600 Sleep Dormitory 119875
0(100) 360
0600ndash0800 Breakfast Dormitoryrestaurant 1198751(068) 120
Sports-breakfast-travel Playground-dormitoryrestaurant 1 minus 1198751(032)
0800ndash1200 Class Classroom 1198752(077) 240
Study Library 1 minus 1198752(023)
1200ndash1400 Lunch Restaurant 1198753(090) 120
Lunch-shopping Restaurant-convenience store 1 minus 1198753(010)
1400ndash1800 Class Classroom 1198754(070) 240
Study Library 1 minus 1198754(030)
1800-1900Dinner Dormitoryrestaurant 119875
5(063)
60Sports-shopping Playgroundconvenience store 1198756(025)
Study Classroomlibrary 1 minus (1198755+ 1198756) (012)
1900ndash2200Rest Dormitory 119875
7(045)
180Dinner-rest Dormitoryrestaurant-dormitory 1198758(020)
Study Classroomlibrary 1 minus (1198757+ 1198758) (035)
2200ndash2400 Rest Dormitory 1198759(072) 120
Sleep Dormitory 1 minus 1198759(018)
infected agents go to hospital according to the treatmentschedule or stay in dormitory according to isolation scheduleAfter the treatment in hospital or self-immunoprocess agentsbecome healthy and immune of disease If the agents aretreated in the hospital they are not allowed to get out untilrecovered Moreover activity schedules are also influencedby the emergency response plans For example in the case ofisolation in emergency response plans the activity scheduleof agents who had contacted with infected is changed Onlythe locations including home and dormitory compose thisemergent schedule Likewise based on the statistical datafrom emergency response plans the additional possibilities
are added into activity schedulesThe behaviors of agents alsochange with activity schedules In the view of ACP approachthe injection of new emergency response plans implementsthe parallel execution of microagent behaviors
In another aspect contact frequency is also anothercrucial element to determine the infected rate in public healthevent It is important to model the contact frequency andcontact time of individuals Based on the studies on thecontact behavior of human being by questionnaire surveyEdmunds found that the contact frequency of individualcould be fitted approximately into a normal distributionand the mean and the standard deviation distributions are
Mathematical Problems in Engineering 13
Susceptible agent ImmuneYes
Healthy and immune
No
Choose schedule
Self-isolation Go to hospital
Staysymptomatic
No
Receive treatment
Contact withinfected agents
Symptomatic
No
Self-immunoprocess
Yes
Recovered
Yes
Incubation
Infection
No
Yes No
NoYes
Figure 12 The workflow of a susceptible agent
168 and 85 [39] So it is possible to apply normal randomvariable to model the contact frequency In our work Box-Muller method [40] is used to generate the random variableof contact frequency shown in the following
119865119862= 120583119865+ 120590119865(minus2 ln (120574
1))12 cos (2120587120574
2) (1)
in which 119865119862is the random variable of contact frequency 120583
119865
is the mean value of normal distribution 120590119865is the standard
deviation of normal distribution and 1205741and 120574
2are the
uniform random variables distributed in the interval [0 1]Based on (1) and survey data [41] the contact frequenciesof agent 119865contact(119860 119894) are discretized as in Table 3 within theconsideration of activity differences
Similarly the duration per contact between individualsis another key factor of AHC transmission The duration ofcontact could also bemodeled by the normal randomvariablein the following [42]
119879119862= 120583119879+ 120590119879(minus2 ln (120574
1))12 cos (2120587120574
2) (2)
in which 119879119862is the random variable of duration per contact
120583119879
is the mean value of normal distribution 120590119879
is thestandard deviation of normal distribution and 120574
1and 120574
2
are the uniform random variables distributed in the interval
[0 1] In addition because the time spent in specific location119879Location(119860 119894) also follows the random distribution like (3)in our work In order to make the duration of contact119879contact(119860 119894) shorter than the 119879Location(119860 119894) the mean andstandard deviation of (2) is set up in (3) and (4) [42]
120583119879=119879Location(120583119865+ 120590119865) (3)
120590119879=119879Location(10120583119865+ 10120590
119865) (4)
in which 119879Location is the time of an activity in a specificlocation set in Table 2 120583
119879and 120590
119879are the mean and standard
deviation of contact frequency As a result the duration ofagent 119879contact(119860 119894) in specific locations (listed in Table 2) isalso discretized in Table 3 within the consideration of activitydifferences
414 Modelling Transportation In artificial Beijing threekinds of travel models are considered walking inside thedistrict travelling by road networks and travelling by subwaynetworks According to the traveling models the agentmovements are implemented by the compositions of travelingmodels before So it is important to build the basic road
14 Mathematical Problems in Engineering
Table 3 The contact frequency and duration in different locations and activities by social relationships in case of student and teacher
Social relationship Activity Location 119879contact(119860 119894) (minute) 119865contact(119860 119894)
Student-student
Sleep Dormitory 0 0
Breakfast Restaurant 119873(52) 119873(32)Breakfast Dormitory 119873(52) 119873(21)Sports Playground 119873(21) 119873(53)Class Classroom 119873(21) 119873(21)Study Library 119873(21) 119873(21)Lunch Restaurant 119873(158) 119873(32)
Shopping Convenience Store 119873(203) 119873(21)Dinner Restaurant 119873(108) 119873(21)Dinner Dormitory 119873(158) 119873(32)Rest Dormitory 119873(63) 119873(52)
Teacher-student Class Classroom 119873(21) 119873(21)
Subway network of Beijing city (red lines) Road network of Beijing city (green lines)
Figure 13 The implementation of transportation for artificial Beijing
and subway networks With the help of Google map thekey points of road are sampled first and the road networksare generated by the links among point Similarly subwaynetworks are generated by the links between subway stationsmapped from the real stations in Google map Figure 13shows the road and subway networks in artificial Beijingwith the road networks in green and the subway networks inred Most of the districts in Beijing city are covered by thesetwo networks The algorithm of hierarchical route planningis proposed as follows
(i) Find the next position where agent will be located innext action
(ii) Obtain the travelingmodels from the statistical trans-portation data
(iii) Search the path from the starting position to thenearest road entry or subway stations
(iv) Search the optimal path from the starting station orroad entry to the nearest station or road exit of targetposition
(v) Search the path from the target subway station or roadexit to the target position
(vi) Attach the subway train number or the bus numberincluding the transfer information to the path accord-ing to the timetable of subway or bus
(vii) Connect all the paths obtained before Generate thepath from the starting position to the target position
Based on the algorithm for each agent in artificial Beijingthe transportation is simulated during the computationalexperiments It is worth noting that the activities insidethe buses or train cars are considered because the agentcommunication during travelling is also necessary to besimulated
Mathematical Problems in Engineering 15
(a) The implementation of agent in MDT (b) The code framework of agent
Figure 14 The agent model implementation in MDT [35]
415 The Implementation of Models by MDT All the modelsare developed in the MDT Figure 14 shows the GUI of thetool MDT is used to implement the state space and statetransitions of FSM models The code framework is shownin Figure 14(b) Using the C++ inheritance features and thetechniques of dynamic link library (dll) the models areencapsulated in the dll componentsThe flexible compositionmechanism is designed to support the models evolution incomputational experiments
416 The Description of Artificial Beijing by ASE As men-tioned in Section 34 ASE describes the artificial Beijingin macroview The statistical information of Beijing city iscollected by ASE Table 4 lists part of the description ofartificial Beijing generated by ASE
42 How to Obtain the Data of Beijing City
421 Population and Geospatial Environment Database ofBeijing City Based on the models developed for ArtificialBeijing it is necessary to generate the initial data to do thecomputational experiments Because the individual level dataare not available it is necessary to construct an individual-based population database for both accurate computationalexperiment and determining optimal decisions
According to the state spaces of models discussed inSection 412 the geospatial and population database of Arti-ficial Beijing is designed as shown in Figure 15 The kernelpart of database consists of the tables such as agent list tablegeospatial environment list table household list table andagent distribution table Agent list table is an individual leveltable used to store the data to initialize the state space of agentmodel such as id and gender age Geospatial environmentlist table is also an individual level table which stores thedata of environment attributes such as id street id andtype Household list table and agent distribution table storethe statistical data for the data generation for agent andenvironment models With the help of the geospatial andpopulation database the data generating of artificial Beijingis proposed in the next section
422 Generating Geospatial and Social Environment Datafor Artificial Beijing by PGET The algorithm of generatinggeospatial and social environment data for artificial Beijingis used to quantify the spatial distribution of population andformalize geospatial behavior of each agent The algorithmshown in Figure 16 is capable of generating a syntheticartificial society which allowsmultiresolution statistical datasocial interactive behavior and multilayer social networksto be integrated together The synthetic population canrepresent individual agents in the form of households andhousehold members and the synthetic population is statis-tically equivalent to a real population For each householdcharacteristics such as address household size family typesand relationships are generated Each person is described bycharacteristics such as age gender social role and correlatedlocations The algorithm provides an effective methodologyto reconstruct the computing environment in high resolutionby using statistical data in low resolution leading to betterprediction and management of emergencies The details ofalgorithm are illustrated in [36]
As shown in Figure 17 the implementation of the algo-rithm is embedded inside the GEPTThe collective input datasuch as the data files of population distribution is collected inthe tool first then the algorithm is activated to generate thedatabase The generation lasts almost twelve hours one timewith the specified parameters settings When the generationis finished the data of artificial Beijing is obtained in thedatabase with 19610000 agents and 16000 environments
5 H1N1 Experiments in ArtificialBeijing with KD-ACP
With the support of the artificial Beijing modeled before thecomputational experiments and parallel execution are alsoimplemented by KD-ACP The experiments are designed tofind the most optimal emergency decision response plansEpidemic in city is a typical scenario in emergency manage-ment H1N1 epidemic in Beijing in 2009 is used to be the usedcase to test KD-ACP According to Section 34 it is necessarytomodel emergencymanagement first in order to support thecomputational experiments
16 Mathematical Problems in Engineering
Table 4 The description of artificial Beijing generated by ASE
Statistical features ofartificial Beijing Descriptions of statistical features Data sourcedefault values
Agent roles Baby child primary student middle school student college student worker retired
Demographic data file ofBeijing
Environment types Super market store restaurant park stadium hospital school cliniccommercial building dormitory building residential building
Demographic data file ofBeijing
Demographic data Age distribution of population sex distribution of population householddistribution of population retired age distribution children distribution
Demographic data file ofBeijing
Environment data Distribution of all types of environments Environment distributiondata file of Beijing
Relationships Classmate relative college friendship family Demographic data file ofBeijing
Varying parameters
Age difference of couples119873(120583120590) 119873(03) isin [0 12]Minimummarriage age (male female) (22 20)Age difference of children119873(120583120590) 119873(13) isin [1 10]Enrollment rate of kindergarten 120572 08Enrollment rate of primary school 120573 1Enrollment rate of middle school 120574 1Enrollment rate of college 120575 08Entrance age of kindergarten 1205721015840 3Entrance age of primary school 1205731015840 7Entrance age of middle school 1205741015840 13Entrance age of middle school 1205751015840 19
Age distributioninformation table
Agent settable
Geospatialenvironment set table
Household settable
Figure 15 The geospatial and population database of Beijing city
51 Modelling Emergency Management The modelling ofemergency management consists of two parts the modellingof emergent events and the modelling of intervention mea-sures H1N1 and the intervention measures for H1N1 aremodeled in our case
511 Modelling H1N1 by MDT H1N1 Model in agent simu-lates the states transitions of health status and the relevantactions Referring to SIR (SusceptibleInfectiveRecovered)[43 44] models agent has three health statuses susceptibleinfected and healthy with immunity Only susceptible agentcan be infected by the contact with infected agents After
the three infected phases (incubation being symptomaticand recovery) of H1N1 model agent is healthy again It isworthy mentioning that the recovered agents can be infectediteratively when the immunity disappears gradually by timeThe state transitions are illustrated in Figure 18
According to [45] the latent period of H1N1 was in aWeibull distribution within people from one to seven days[46 47] The distribution was usually centered in the rangebetween one and three days AWeibull random variable [40]is used to model the latent period as described in
119879Lat = 120572 times [minus ln (1 minus 120574)]1120573+ 120592 (119879Lat ge 120592) (5)
Mathematical Problems in Engineering 17
Censusdata
Generate GUID
Assign individualhousehold
Generate familystructure
Assign family roleto every member
Endow individualwith age gender
social role
Distribute environmentlocations on the map
School Workplace Entertainmentlocations
Consumptionlocations
Assign school orworkplace to
individual
Correlatedindividual withother locations
Generate friendshipin school
Generate coworkerrelations inworkplace
Assign individualadministrative
region
Generate housebuildings
Locate householdto house building
Generate familyrelation
Generateneighborhood
relation
PopulationEnvironment
Environment listSocial network
Figure 16 The algorithm of generating geospatial and social environment data [36]
Edit
Edit
Edit
Edit
Edit
Edit
BrowseStatistical characteristic of population
Geospatial population data
Population data of street
Population types
Geospatial and social environment generation tool
Workspaces
Parameters distribution settings
Household types
Roles in household
Household structure
Social roles
BrowseBrowseBrowseBrowseBrowseBrowse
Population distribution data
File path
Retired distributionPop distributionHousehold numHousehold distributionSex distributionAge distributionData type
Figure 17 The GUI of Population and Geospatial Environment generation Tool (PGET)
18 Mathematical Problems in Engineering
Infected
SusceptibleHealth
withimmunity
Infected
by contac
t
with in
fected
agents
Recovery
Increase before beingsymptomatic symptomatic
Decrease after being
Decrease by time
Infectivity
Immunity
Figure 18 The transitions of agent health status
119879Lat denotes the duration of latent period 120592 120572 and 120573 arethe location parameter scale parameter and shape parameterof Weibull distribution respectively 120574 is a uniform randomnumber in the range [0 1] According to statistics [47] 120592 120572and 120573 are set as 0 18 and 121 respectively Then the mean(standard deviation) of latent period is calculated as 159 days(058 day2) Furthermore infectious period is set as 7 days
A susceptible studentrsquos probability of getting infecteddepended on the infectivity of infectious agents his ownimmunity level the duration of the contact action and soon The infectivity of an infected agent evolved with timeduring the infectious period According to the statistics ofinfectivity in the chart with red bars in Figure 19 an infectedstudent had the highest level of infectivity in the second dayafter he had the first symptom The infectivity levels in otherdays in comparison with the second day are in Table 5 Theday labeled as ldquominus1rdquo means the day before the starting of thesymptomatic period
Based on the algorithm discussed before theH1N1modelis developed with the help of MDT The settings of influenzamodel will be discussed in Section 521
512 Modelling Intervention Measures by MDT In corre-spondence intervention measures model is used to ceasethe outbreak of H1N1 As discussed in [42 48 49] if theappropriate measures are taken when the infectious diseaseemerges the transmission of the disease could be sloweddown and the damage could be decreased Therefore it isnecessary and important to design emergency interventionmeasures in the artificial Beijing According to the MinistryofHealth inChina the intervention policies [50] are designedto control the spread of influenza such as H1N1 The inter-ventionmeasures of these policies including are interventionsactivated time vaccination rate antibiotic delay to hospitalisolation duration close workspace duration and limitationof activities listed in Table 6 Similarly the models of theseintervention measures are also developed with the help ofMDT Using the reasonable data ranges listed in Table 6intervention models can be initialized to do the computa-tional experiments These experiments are used to show how
Symptomatic1 2 3 4 5 6 7minus1
Recovered
Tim
e (da
y)
Infe
ctiv
ity
Figure 19 The distribution of infectivity rate of an infected agent[37]
Table 5 The proportion of infectivity
Days (119894) minus1 1 2 3 4 5 6 6Proportion of infectivity 03 05 1 08 04 02 01 005
to restrain the H1N1 transmissionThe settings of these inter-vention measures in IMCT are introduced in Section 522
52 Experiment Settings
521 H1N1 Settings by EECT EECT is designed to initializethe emergent events models such as H1N1The settings comefrom the specified emergent event models As shown inFigure 20 in theH1N1 used case the settings are composed ofthe distribution of infection source infection rate infectionperiod and so on The infection rate changes with envi-ronments and infection periods The settings are initializedfrom both the research of medicine on H1N1 and the data ofH1N1 outbreak in Beijing in 2009The process of influenza inartificial Beijing is simulated to repeat the possible scenariosof H1N1 break During the computational experiments ofartificial Beijing theH1N1 settings such as the period of beingsymptomatic are updated by the dynamic data registrationsfrom OsdRT The updating corrects the H1N1 model duringthe experiments With the help of OsdRT artificial Beijing isable to approach actual society It is also an implementationof parallel execution
522 Intervention Measures Settings by IMCT In the H1N1case the settings of intervention measures are shown inFigure 21 in IMCT These measures are designed under theinstruction of Emergency Decision Organizations like ChinaDisease Center (CDC) The measures such as vaccinationrate the rate of closedworkspace and the starting time can beinitialized in the tool Emergency response plans are imple-mented in these settings As illustrated in Figure 7 all the pos-sible compositions within value settings in reasonable inter-vals are tested to find the most optimal emergency responseplan For example different compositions of measures suchas vaccination rates and rates of workspace closed are usedto design the plans of experiments By the analysis of theresults the best composition of interventionmeasures can beobtainedThe analysis will be discussed in case study in detail
Mathematical Problems in Engineering 19
Table 6 The intervention measures of influenza
Intervention measures ValuesInterventions activated Time 119879Intervention isin [0 80th day]Vaccination rate 119877Vaccination isin [0 1]Antibiotic rate 119877Antibiotic isin [0 1]Delay to hospital 119867ospital(119860 119894) 119879infected lt 119905 lt 119879recovered + 119879HDelay 119879HDelay isin (0 3th day]Isolation duration 119868solation(119860 119894) 119879Intervention lt 119905 lt 119879Intervention + 119879IDuration 119879IDuration isin (0 14th day]Close workspace duration 119862lose(119864119894) 119879Intervention lt 119905 lt 119879Intervention + 119879CWDuration 119864119894 isinWorkspace 119879CWDuration isin (0 7th day]Limitation of agent activity duration 119871 imitation(119860 119894) 119879Intervention lt 119905 lt 119879Intervention + 119879LAADuration 119879LAADuration isin (0 7th day]119877Vaccination means the vaccination rate of artificial Beijing while 119877Antibiotic means the antibiotic rate used for infected agents in hospital119867ospital (119860119894) means that the infected agent 119860119894 is sent to hospital he or she will receive the retreatment immediately119868solation (119860119894) means that the infected agent 119860119894 is isolated at home he or she can only contact with the other agents until isolation duration is over119862lose (119864119894) means that the environment 119864119894 typed of workspace is closed until the close workspace duration is over119871 imitation (119860119894) means that the activities of agent 119860119894 are limited only the workspace and home are allowed until limitation duration is over
Incubation Weibulldistribution
Symbolic LogNormaldistribution
Recovering LogNormaldistribution
H1N1influenza
Settings of H1N1 influenza model
Disease model
Transmissionmodel
Figure 20 The settings of H1N1 in EECT
It is worth notifying that the settings of interventionmeasures are also influenced by OsdRT For example theH1N1 outbreak areas are notified by OsdRTThe interventionmeasures such as close workspace are used only in theoutbreak areas Therefore the unnecessary computation ofintervention measures is avoided So the data injection byOsdRT not only makes artificial society approach actualsociety but also increases the performance of computationalexperiments
53 Experiments by EMT When the initialization of arti-ficial Beijing is ready EMT is used to do the computa-tional experiments EMT is composed of two parts thecentral controller in the central node and the residen-tial service deployed in the computational nodes Centralcontroller loads the experiment plans and controls the whole
experiments process the GUI of controller is shown inFigure 22 EMT manages the computational experimentsin four steps cluster configuration load experiment plansmodels deployment and running experiments
531 Cluster Configuration Before the configuration clusterinformation is collected from the residential services insidethe cluster The information includes nodes amount nodenames node IPs and CPU occupations It is used to quantifythe computing power of the clusterThe cluster configurationis based on this information The nodes are selected forexperiments Then the LP (local process) [51] number foreach node is set according to the number of the CPU kernelsAfter the configuration the experiment plans are loaded tocustomize the models and data
20 Mathematical Problems in Engineering
Streets of close workspaces
Limitations ofactivities
IsolationsRespirator
Closeworkspace
Antibiotic
VaccinationHospital
Surveillance
Figure 21 The settings of intervention measures in IMCT
Information ofexperiment plan
List of experiment plans
Start | Pause | Reset | Stop | Speedup | Slowdown | Back
NodeName | CPU occupation | Memory occupation |Network | Models | Status
Figure 22 The deployment and running of experiments in EMT
532 Load Experiment Plan As discussed in Section 35 thedescription of experiment plans is composed of the infor-mation in three aspects models and the initial data for theexperiments the deployment mapping tables from the mod-els to the nodes and the plan of experiment executionWhenthe experiment plan is loaded by the EMT the file namesand model descriptions of models and data are listed Withthe support of cluster configuration models can be mappedto the nodes in the GUI according to the computationrequirements EMT also provides the setting of experimentsitself including the running times start time and end time
533 Models and Data Deployment Models are stored inthe agent population and services repository According tothe settings in experiment plan the customized models aredownloaded from the repository first then the models suchas agent model environment model emergent events modeland intervention measures model are integrated to be thecompositionmodels of artificial BeijingThemodels and datalocations in the nodes are configured in the ldquomodel pathrdquoand ldquodata pathrdquo in the GUI of central controller Finallythe models are uploaded to the nodes in cluster with theconsideration of partition The partition of artificial society
Mathematical Problems in Engineering 21
models is different from the multiagent system in [52]Models in artificial society are divided into three levels agentsand environments grids and city Agents and environmentsare encapsulated in grids which compose the whole cityThus the partition of artificial society becomes the partitionof grids The parallelism degrees are found to support thepartition of models agent independence and grid indepen-dence According to the parallelisms agents are partitioned toLPs according to the population distribution in grids Agentmovements among grids are simulated by the movementevents among grids in different LPs As a result a two-tierparallel architecture is proposed to support the simulation ofartificial society The architecture is detailed in [53]
534 Running of Experiments When the preparation isready the experiments can be done by the central controllerof EMT The running and control commands of the experi-ments are grouped into three aspects start pause or stop theexperiments get the running information of the nodes suchas the snapshot of the desktop of node With the help of thesecommands users can run the computational experimentsto do the research on the emergency management It isworth notifying that the simulation engine is optimized inorder to support the ten million agent simulation UsingGPU as coprocessor a two-tier parallel simulation engine isdesigned with support of MPI and OpenCL through phasedsynchronization [54] One-sided communication is used forreflection of remote simulation objects and message passingbetween processes A general kernel function prototype iselaborately designed and conditionally compiled for execu-tion on both CPU and GPU Moreover the optimizationoperations like load balance is developed under the instruc-tion of activity-based simulation [55] The densely populatedgrids are given more CPU and memory resource accordingto the activity predictions
54 Results of H1N1 Experiments with KD-ACP On thebasis of artificial society models and the intervention mea-sures mentioned in Section 512 a series of computationalexperiments are performed to study the H1N1 influenza inBeijing In our case of artificial Beijing 19610000 agentsare simulated in the cluster within 48 CPU cores and 128Gmemories the nodes are connected by kilomega networks Ittakes 18 hours to simulate a 250-day disease spreading
In order to illustrate how the optimal plans are obtainedbyKD-ACP the experiments are divided into four groups theexperiments ofmodel validation the experiments of sensitiv-ity analysis of vaccination rate the experiments of sensitivityanalysis of isolation and the experiments of the combinationof interventionmeasuresThese four groups show the generalprocess of the research on emergency management in H1N1influenza Firstly the models of H1N1 are validated withthe support of historical data Secondly the traditionalmedical interventions of pandemic such as vaccination areanalyzed to find the optimal plan Thirdly the traditionalnonmedical interventions of pandemic such as isolation areanalyzed to find the optimal plan Finally the combination ofinterventions is tested to show the combined effect
It is worth mentioning that all the experiments in ourwork are performed 100 times These experiments are ini-tialized with different settings of interventions in IMCT Theresults shown in the figures are the mean values
541 The Experiments of Model Validation In our artifi-cial Beijing the models of H1N1 influenza are built basedon the existing researches Wang and her colleagues hadstudied the H1N1 influenza by SEIR (SusceptibleExposedInfectiveRecovered) models [56] With the support of theirwork we build the models of artificial Beijing in the mannerof multiagents
In Figure 23 the control group shown in red line isdrawn based on the realistic statistical data collected in H1N1influenza in Beijing in 2009 The influenza lasts more thansix months and more than 170000 people are infectedComparing with the control group the simulated resultswithout interventions are drawn in blue It is obvious that thesimulated case fits the control group well before 70 days Butthe ascending of control group is slowed after 70 days andthe peak value (17883) is much less than the simulated results(25135) It is because the vaccination is activated at day 67 incontrol group according to the intervention measures usedin 2009 The restrain by vaccination becomes effective eightdays later The change of infected number after vaccinationis shown by red spotted line The trend of control group isdifferent from the simulation data after 75 days
However the simulated results (blue line) in compu-tational experiments fit the control group (as shown inFigure 23) in the first 70 days when no intervention mea-sures are executed The key features such as effective basicreproduction number (119877
0) are also almost the same in both
cases Therefore it can be concluded that the models ofartificial society built in microview are validated by therealistic statistical data in influenza in Beijing in 2009
In this group of experiments no interventions are exe-cuted in order to validate the models Therefore the simu-lated results of this extreme case which cannot happen todaycould be obtained As shown by blue line the number ofinfected agents without interventions reaches peak value atday 91 with 25135 consisting of 8052 in symptomatic phaseand 28950 in recovery phase It is obvious that the totalnumber of infected agents grows slowly at the beginningtime Then the number increases quickly when the influenzabreaks and it reaches its maximum value with 383870 in theend This phenomenon can be explained as in the followingAt the beginning of influenza due to the limited numberof infected agents the propagation of the disease remainsslow but many agents become susceptible When the numberof infected agents increases rapidly the spreading becomessignificant However the number of infected agents decreasesafter day 91 It is because the disease spreading relies on thesocial networks and spatial contact networks [49] It meansthat the infected agents can only infect agents either in thesocial relationship networks or in the spatially contacted
542 The Experiments of Sensitivity Analysis of VaccinationRate According to Section 512 vaccination is a typical
22 Mathematical Problems in Engineering
0
Infe
cted
agen
ts
0 50 100 150 200 250
Statistical dataSimulation data
25000
20000
15000
10000
5000
Day d
Figure 23 The results with no intervention executed and the com-parison with realistic statistical data in Beijing in 2009
medical intervention in influenza Traditionally the effective-ness of vaccination can only be evaluated after the influenzaand the key settings such as vaccination rate are determinedby the experience It is always difficult for emergency man-agement organizations to find the most optimal vaccinationrate in advanceTherefore the sensitivity of vaccination rate isanalyzed in this group of experiments The settings are listedin Table 7 average results are shown in Figure 24
Vaccination rates are set linearly (10 30 50 and70) to test the changes of influenza The peak value ofinfected agents decreases from 25135 22392 15787 6907to 638 respectively It shows that the more the agents arevaccinated the smaller the peak value of the infected agentswill be In the case of 70 vaccination the total infected agentis only 13799 It means that the influenza in this case does nothappen due to the high vaccination rate
The change of peak values also shows an interestingphenomenon Vaccination rates increase linearly the peakvalues decrease slower than the ascending of vaccination ratesat beginning but the descending becomes more and morefaster with the linear increase of vaccination rate It meansthat the small scale of vaccination (less than 30) does notmake senseWith the increase of vaccination rate the numberof susceptible agents decreases significantly In sequencethe possibility of infections is lowered greatly Additionallyit could be found that the decrease of peak values comeslater than the increase of the vaccination rates This is quitereasonable according to Mei et alrsquos work in [57]
As a result it is concluded that high vaccination rate iseffective in the control of H1N1 influenza though costly
543 The Experiments of Sensitivity Analysis of IsolationAccording to Section 512 isolation is a typical nonmedicalintervention It is the most used intervention measure in
Table 7 The initial settings of experiments of vaccination
Parameters ValueAgent count 19610000Initial infected count 40Interventions activated time(119879Intervention)
0
Vaccination rate (119877Vaccination) 0 10 30 50 70
0
Infe
cted
agen
ts
Loading intervention of vaccination
No vaccination10 vaccination30 vaccination
50 vaccination70 vaccination
0 50 100 150 200 250
25000
20000
15000
10000
5000
Day d
Figure 24 The results of loading intervention of vaccination
the emergency management in influenza As mentionedbefore a problem also exists that the measure cannot bevalidated before the intervention Therefore the sensitivityof isolation is analyzed in this group of experiments It isworth mentioning that the isolation is usually combinedwith the hospital policy It means the infected individual issent to hospital meanwhile the people who contact withthe infected are all isolated As a result parameters delay tohospital and isolation duration are designed together in thesettings of experiment Delay to hospital stands for the periodfrom infection time to hospital time while Isolation durationrepresents the period when an agent cannot contact others ifhe has contact with an infected agent The settings of theseparameters are listed in Table 8 average results are shown inFigure 25
Firstly the parameter delay to hospital is simulated aloneto test the sensitivity The simulated results are shown insolid line in Figure 25 The peak value decreases from 6708to 2218 when the delay to hospital decreases from 15 daysto 1 day The half day change brings the 67 reductionMoreover the peak value almost approximates to zero inthe case of 05 days The influenza does not happen in thiscase It can be concluded that the earlier the infected agents
Mathematical Problems in Engineering 23
Table 8 The initial settings of experiments of isolation
Parameters ValueAgent count 19610000Initial infected count 40Interventions activated time (119879Intervention) 80Delay to hospital (119879HDelay) isolation duration (119879IDuration) (15 0) (1 0) (05 0) (15 3) (1 3) (05 3)
0 50 100 150 200 2500
1000
2000
3000
4000
5000
6000
7000
Infe
cted
agen
ts
Loading intervention of isolation
Delay to hospital = 15
Delay to hospital = 10
Delay to hospital = 05
Delay to hospital = 15 isolation duration = 3
Delay to hospital = 10 isolation duration = 3
Delay to hospital = 05 isolation duration = 3
Day d
Figure 25 The results of loading intervention of isolation
are sent to hospital the less the agents will be infectedSo the construction of effective mechanism to find infectedindividuals is important in the emergency management ofinfluenza
Secondly the parameter isolation duration is added tosimulate the isolation of interventionObviously the isolationmeasure decreases the infected agents greatly Compared tothe parameter delay to hospital of 15-day case the peak valuedecreases from 6708 to 362when the isolation is added in theinterventionThe reduction of 6346 agents comes from the 3-day isolation of agents who had contacted with the infectedagents However the decrease of parameter delay to hospitalcannot bring the similar reduction in the 3-day isolationcases According to the figure the peak value decreases from362 to 51 from 15-day case to 05-day case only 311 agents aresaved from infection As a result the sensitivity of simulatedresults does not change obviously with delay to hospital inthe case of isolation It can be concluded that isolation is amore effective measure Though it is not easy for emergencymanagement organizations to decrease the delay to hospitaldue to the current monitoring mechanism the influenza canstill be controlled by isolation
544 The Experiments of Combination of Intervention Mea-sures Isolation is not the only nonmedical intervention ininfluenza close workspace is another common measureWhen close workspace is activated the workspaces are allclosed agents can only stay home at the closed time Luckilyisolation and close workspace are independent from oneanother It is possible for emergency management organi-zations to use these two measures together So this groupof experiments is designed to obtain the effectiveness of thecombination of intervention measures The parameters areinitialized with different settings listed in Table 9
Parameters delay to hospital and isolation duration is setin the middle values (1 day 3 days) according to the lastsection According to the experience close duration is set asseven days It is known that close workspace is a really costintervention measure the activated time is a key parameterWith the fixed parameters mentioned before the changeof interventions activated time is analyzed in this group ofcomputational experimentsThe average results are shown inFigure 26
Compared with the simulated results of experimentsdiscussed above Figure 26 strongly illustrates that influenzais ceased when the intervention measures are combinedtogether The peak value of infected agents is 141 in the ldquo80thdays of close-workspace and isolation with 1 day delay tohospitalrdquo case while 25135 in ldquono interventionrdquo case 2218 inldquo1 day delay to hospitalrdquo case 203 in ldquo1 day delay to hospitaland isolationrdquo case The interventions bring a great reductionof infection the peak values decrease from ten thousandslevel to hundred level
On the other hand the peak value even decreases to 108when the execution time of close workspace is 10 days earlierThe date of maximum value of infected agents is advancedfrom the 135th day to the 108th day The whole circle of H1N1influenza is reducedThe results of this group of experimentsshow that intervention measures should be activated as soonas possible and the compositions of interventions are moreeffective in emergency management of H1N1 influenza
In summary based on the general process lots of exper-iments could be done by KD-ACP to study emergency man-agement in H1N1 influenza Many conclusions are obtainedbased on the analysis of the experiments in this sectionAccording to these conclusions themost optimal response inour case study is the case of ldquo70th days of close-workspace andisolation with 1 day delay to hospitalrdquo It is because this casehas the fewest infected agents relatively But the modelling ofH1N1 influenza is not sufficient Economic and social factorsof interventions such as the money cost by vaccination andisolation the influence of the close-workspace in society
24 Mathematical Problems in Engineering
Table 9 The initial settings of experiments of isolation and closeworkspace
Parameters ValueAgent count 19610000Initial infected count 40Interventions activated time (119879Intervention) 70th day 80th dayDelay to hospital (119879HDelay) 1 dayIsolation duration (119879IDuration) 3 daysClose workspace duration (119879CWDuration) 7 days
50 100 150 200 2500
50
100
150
200
Infe
cted
agen
ts
Loading intervention of isolation and close workspace
Delay to hospital = 1 isolation duration = 3Start close time = 70 close duration time = 7Delay to hospital = 1 isolation duration = 3Start close time = 80 close duration time = 7
Day d
Figure 26The results of loading interventions of isolation and closeworkspace
and even opinion transmission of H1N1 in internet are notconsidered in our case study Without the analysis of thesefactors it does not make sense to talk about the most optimalintervention measures However KD-ACP had provided asoftware framework and general process of the research onemergency management Supported by KD-ACP the expertsof emergency management are able to study emergent eventssuch as H1N1 influenza as detailed as possible Based on theexperiments and analysis again and again the most optimalintervention measures will be found ultimately
55 Parallel Execution by ASST and OsdRT As discussed inSection 35 ASST andOsdRT are used to support the parallelexecution of artificial Beijing The statistical data is collectedand stored in artificial society runtime database at runtimeASST downloads the infection data from the database anddisplays it in the map As shown in Figure 27 the infectedagents are distributed in the districts in BeijingThe snapshotgives the geospatial distribution of the H1N1 epidemic Thelegends of the agents in different status are listed in the left
right corner in Figure 27 Five statuses are shown in the figurehealth without vaccination health with vaccination incuba-tion being symptomatic and convalescence ASST providesa very direct feeling for the emergency experts Partly withthe situation from ASST Emergency Decision Organizationmakes the emergency response plans to the response of theemergency scenario With the help of plans the epidemicin actual society can be restrained In the meantime theinfluenza information is reflected in the Internet Accordingto Figure 8 the data acquisition data extraction and datastandardization of H1N1 influenza are processed in OsdRTOsdRT is implemented by Dr Caorsquos team from Institute ofAutomation from Chinese Academy of Sciences [58] Thetool provides not only the reports of H1N1 SARS hand-foot-mouth disease and intestinal diseases are all monitored Theweb page is shown in Figure 28 H1N1 reports from Internetare collected first Based on the data extraction of the reportsH1N1 case of distribution is obtained and shown in theincidencemap As a result H1N1 incidence data is regarded asthe knowledge which is sent to EECT and IMCTThe settingsof emergent events and intervention measures are modifiedin order to make artificial society approach actual society innext turn of computational experiments With the help ofOsdRT the loop of parallel execution process integrates ACPapproach
6 Conclusion
This work provides an integrated software framework forsocial Computing in emergencymanagement From a systemperspective KD-ACP provides a reliable flexible low costand effective platform for the scientists to do the research onemergency response problems The analysis and predictionof these problems with inherent complexity can be solvedby the repetition of possible alternative experiments onthe software framework Therefore KD-ACP which is anattempt to implement ACP approach seemed as an effectivecomputational framework to support the decision making inemergency response
Currently KD-ACP is used to study the H1N1 epidemicin Beijing in 2009 The most optimal compositions of inter-vention measures are testified to support the response tonext influenza However it is still a long way ahead beforethe automation systematization and practicalization of KD-ACP A lot of work will be carried out along several directionsas in the following
(i) The parallel execution process in KD-ACP is actuallynot strongly connected with the emergency responseorganization Traditional experience and theory stillplay a dominant role in the decision of emergencymanagement It is necessary to propose a mechanismof hall for workshop to facilitate the parallel executionloop in actual society
(ii) Section 33 introduces the principle and process ofmodelling artificial society Metamodels and modelsof artificial Beijing are built by GME Model transfor-mation and code generation are used to implementcodes from FSM based models But many details
Mathematical Problems in Engineering 25
Fangshan district Daxing district Fengtai district Xicheng district
Mentougou district Haidian districtShijingshan district Dongcheng district
Incubation
Symptomatic
Convalescence
Health withoutvaccination
Health withvaccinationLegend of agent
health status
Figure 27 The situation of artificial Beijing by ASST
H1N1
Districts of Beijing
H1N1 case ofdistribution
Incidence map
Distribution ofH1N1 reportsfrom internet
Incidence of H1N1
Fangshan districtLiangxiang town
Population 19000Incidence 135789
(1100000)
Figure 28 The implementation of OsdRT by Institute of Automation from Chinese Academy of Sciences
are still not illustrated due to the topic of thispaper The key techniques such as template of codeframework and model transformations from specificdomains to FSM will be focused and introduced indetail in our next paper
(iii) The performance of computational experiments inKD-ACP is still slower than expectation The opti-mization of simulation engine and models structureis necessary to improve the performance So it ispossible to run artificial society with the real worldin parallel
In summary KD-ACP paves a new way for scientists inboth emergencymanagement and computation simulation to
collaborate with each other in solving the emergency man-agement problems by social computing
Research Highlights
(i) KD-ACP is implemented based on ACP approach
(ii) Artificial Beijing is built with the help of KD-ACP
(iii) H1N1 is modeled to simulate the influenza in Beijingin 2009
(iv) Computational experiments testify the effectivenessof intervention measures
26 Mathematical Problems in Engineering
Conflict of Interests
The authors declare that there is no conflict of interestsregarding the publication of this paper
Acknowledgment
This work was supported in part by the National ScienceFoundation of China (nos 91024030 71303252 61403402and 91324013)
References
[1] W Mao and F-Y Wang Advances in Intelligence and SecurityInformatics Academic Press Oxford UK 2012
[2] E Bonabeau ldquoAgent-based modeling methods and techniquesfor simulating human systemsrdquo Proceedings of the NationalAcademy of Sciences of the United States of America vol 99 no3 pp 7280ndash7287 2002
[3] K M Carley D B Fridsma E Casman et al ldquoBioWar scalableagent-basedmodel of bioattacksrdquo IEEETransactions on SystemsMan and Cybernetics A Systems and Humans vol 36 no 2 pp252ndash265 2006
[4] J M Epstein ldquoModelling to contain pandemicsrdquo Nature vol460 no 7256 p 687 2009
[5] Los Alamos National Laboratory Reports about EPISIMS LosAlamos National Laboratory 2003
[6] C H Builder and S C Bankes ldquoArtificial societies a conceptfor basic research on the societal impacts of informationtechnologyrdquo RAND Report 149 RAND Corp 1991
[7] F-Y Wang ldquoComputational theory and method on complexsystemrdquo China Basic Science vol 6 no 5 pp 3ndash10 2004
[8] F-Y Wang ldquoToward a paradigm shift in social computing theACP approachrdquo IEEE Intelligent Systems vol 22 no 5 pp 65ndash67 2007
[9] F-Y Wang and S Tang ldquoArtificial societies for integratedand sustainable development of metropolitan systemsrdquo IEEEIntelligent Systems vol 19 no 4 pp 82ndash87 2004
[10] D Wen Y Yuan and X-R Li ldquoArtificial societies computa-tional experiments and parallel systems an investigation ona computational theory for complex socioeconomic systemsrdquoIEEE Transactions on Services Computing vol 6 no 2 pp 177ndash185 2013
[11] F Wang F Zeng and Y Yuan ldquoAn ACP-based approach forcomplexity analysis of E-commerce systemrdquo Complex Systemsand Complexity Science vol 5 no 3 pp 1ndash8 2008 (Chinese)
[12] F-Y Wang and P K Wong ldquoResearch commentary intelligentsystems and technology for integrative and predictivemedicinean ACP approachrdquoACMTransactions on Intelligent Systems andTechnology vol 4 no 2 pp 1ndash6 2013
[13] J Sifeng X Gang F Dong and H Chunpeng ldquoStudy on theemergency rescue decision support system of petrochemicalplant based on ACP theoryrdquo in Proceedings of the 6th Manage-ment Annual Meeting (MAM rsquo11) 2011 (Chinese)
[14] X Dong D Fan G Xiong F Zhu Z Zhang and Y Yao ldquoPar-allel Bus Rapid Transit (BRT) operation management systembased on ACP approachrdquo in Proceedings of the 6th ManagementAnnual Meeting (MAM rsquo11) Beijing China 2011
[15] B Ning F-Y Wang H-R Dong R-M Li D Wen and L LildquoParallel systems for Urban rail transportation based on ACP
approachrdquo Journal of Transportation Systems Engineering andInformation Technology vol 10 no 6 pp 23ndash28 2010 (Chinese)
[16] F-Y Wang J Zhao and S-X Lun ldquoArtificial power systemsfor the operation and management of complex power gridsrdquoSouthern Power System Technology vol 2 no 3 pp 1ndash11 2008(Chinese)
[17] M J North T R Howe N T Collier and J R Vos ldquoRepastsimphony development environmentrdquo in Proceedings of theAgent 2005 Conference on Generative Social Processes Modelsand Mechanisms 2005
[18] M J North T RHowe N T Collier and J R Vos ldquoRepast Sim-phony runtime systemrdquo in Proceedings of the Agent Conferenceon Generative Social Processes Models and Mechanisms 2005
[19] SWARM Development Group Swarm 211 Reference Guide2000 httpwwwswarmorg
[20] S Tisue and U Wilensky ldquoNetLogo a simple environment formodeling complexityrdquo in International Conference on ComplexSystems pp 16ndash21 2004
[21] S Luke C Cioffi-Revilla L Panait K Sullivan and G BalanldquoMASON a multiagent simulation environmentrdquo Simulationvol 81 no 7 pp 517ndash527 2005
[22] M Ichikawa H Tanuma Y Koyama andH Deguchi ldquoSOARSintroduction as a social microscope for simulations of socialinteractions and gamingrdquo in Organizing and Learning throughGaming and Simulation Proceedings of the 38th Conference ofthe International Simulation and Gaming Association (ISAGArsquo07) pp 149ndash158 2007
[23] D S Burke J M Epstein D A T Cummings et al ldquoIndividual-based computational modeling of smallpox epidemic controlstrategiesrdquo Academic Emergency Medicine vol 13 no 11 pp1142ndash1149 2006
[24] F-Y Wang ldquoParallel control and management for intelligenttransportation systems concepts architectures and applica-tionsrdquo IEEE Transactions on Intelligent Transportation Systemsvol 11 no 3 pp 630ndash638 2010
[25] Z Molnar D Balasubramanian and A Ledeczi ldquoAn introduc-tion to the genericmodeling environmentrdquo in Proceedings of theTOOLS Europe 2007 Workshop on Model-Driven DevelopmentTool Implementers Forum Zurich Switzerland 2007
[26] S Kelly and J-P Tolvanen Domain Framework in Domain-Specific Modeling Enabling Full Code Generation JohnWiley ampSons Hoboken NJ USA 2007
[27] B P Zeigler H Praehofer and T G Kim Theory of Modelingand Simulation Integrating Discrete Event and ContinuousComplex Dynamic Systems Academic Press San Diego CalifUSA 2000
[28] G Booch J Rumbaugh and I JacobsonThe Unified ModelingLanguage User Guide Addison-Wesley Reading Mass USA1999
[29] F Wagner R Schmuki T Wagne and P Wolstenholme Mod-eling Software with Finite State Machines A Practical ApproachCRC amp Taylor amp Francis Boca Raton Fla USA 2006
[30] T Espiner ldquoChina builds worldrsquos fastest supercomputerrdquo ZDNetUK October 2010
[31] B Chen and G Guo ldquoA two-tier parallel architecture for arti-ficial society simulationrdquo in Proceedings of the ACMIEEESCS26th Workshop on Principles of Advanced and DistributedSimulation (PADS rsquo12) pp 184ndash186 July 2012
[32] G Guo B Chen X G Qiu and Z Li ldquoParallel simulation oflarge-scale artificial society on CPUGPU mixed architecturerdquo
Mathematical Problems in Engineering 27
in Proceedings of the ACMIEEESCS 26th Workshop on Prin-ciples of Advanced and Distributed Simulation (PADS rsquo12) pp174ndash177 July 2012
[33] S-X Lun ldquoResearch on the classification of parallel executionmodes of ACP theoryrdquo Acta Automatica Sinica vol 38 no 10pp 1602ndash1608 2012
[34] F-YWang ldquoA framework for social signal processing and anal-ysis from social sensing networks to computational dialecticalanalyticsrdquo Science China vol 43 no 12 pp 1598ndash1611 2013
[35] G GuoOne Model Userrsquos Manual version 12 National Univer-sity of Defense University 2011
[36] Y Ge R Meng Z Cao X Qiu and K Huang ldquoVirtual citymdashanindividual-based digital environment for human mobility andinteractive behaviorrdquo Simulation Transactions of the Society forthe Modeling and Simulation International 2014
[37] M Kretzschmar and R T Mikolajczyk ldquoContact profiles ineight European countries and implications for modelling thespread of airborne infectious diseasesrdquo PLoS ONE vol 4 no6 Article ID e5931 2009
[38] W Duan and X Qiu ldquoFostering artificial societies using sociallearning and social control in parallel emergency managementsystemsrdquo Frontiers in Computer Science vol 6 no 5 pp 604ndash610 2012
[39] W J Edmunds C J OrsquoCallaghan and D J Nokes ldquoWhomixes with whom Amethod to determine the contact patternsof adults that may lead to the spread of airborne infectionsrdquoProceedings of the Royal Society B Biological Sciences vol 264no 1384 pp 949ndash957 1997
[40] J Banks J S Carson B L Neison and D M Nicol Discrete-Event System Simulation Prentice Hall Englewood Cliffs NJUSA 4th edition 2007
[41] H Yongxia ldquoInvestigation and analysis of college studentsrsquo workand restrdquo Value Engineering vol 31 no 3 2012 (Chinese)
[42] W Duan Z Cao Y Wang et al ldquoAn ACP approach to publichealth emergency management using a campus outbreak ofh1n1 influenza as a case studyrdquo IEEE Transactions on SystemsMan and Cybernetics Part A Systems and Humans vol 43 no5 pp 1028ndash1041 2013
[43] HWHethcote ldquoThemathematics of infectious diseasesrdquo SIAMReview vol 42 no 4 pp 599ndash653 2000
[44] G Chowell E Shim F Brauer and et al ldquoModelling thetransmission dynamics of acute haemorrhagic conjunctivitisapplication to the 2003 outbreak in Mexicordquo Statistics inMedicine vol 25 no 11 pp 1840ndash1857 2006
[45] Y Ge L Liu B Chen X Qiu and K Huang ldquoAgent-basedmodeling for InfluenzaH1N1 in an artificial classroomrdquo SystemsEngineering Procedia vol 2 pp 94ndash104 2011
[46] A R Tuite A L Greer MWhelan et al ldquoEstimated epidemio-logic parameters andmorbidity associatedwith pandemicH1N1influenzardquo CMAJ vol 182 no 2 pp 131ndash136 2010
[47] L Brouwers B Cakici M Camitz A Tegnell and MBoman ldquoEconomic consequences to society of pandemic H1N1influenza 2009mdashpreliminary results for Swedenrdquo Euro Surveil-lance vol 14 no 37 pp 1ndash7 2009
[48] Y Ge L Liu X Qiu H Song Y Wang and K Huang ldquoAframework of multilayer social networks for communicationbehavior with agent-based modelingrdquo Simulation Transactionsof the Society for the Modeling and Simulation International vol89 no 7 pp 810ndash828 2013
[49] W Duan X Qiu Z Cao X Zheng and K Cui ldquoHeterogeneousand stochastic agent-based models for analyzing infectious
diseasesrsquo super spreadersrdquo IEEE Intelligent Systems vol 28 no4 pp 18ndash25 2013
[50] The Preparation and Emergency Response Plans to PandemicPublished by Ministry of Health Ministry of Health 2005(Chinese)
[51] R M Fujimoto Parallel and Distributed Simulation SystemsWiley Interscience 2000
[52] F Cicirelli A Giordano and L Nigro ldquoDistributed simulationof situated multi-agent systemsrdquo in Proceedings of the 15thIEEEACM International Symposium on Distributed Simulationand Real Time Applications (DS-RT rsquo11) pp 28ndash35 September2011
[53] B Chen and G Guo ldquoA two-tier parallel architecturefor artificial society simulationrdquo in Proceedings of the 26thACMIEEESCS Workshop on Principles of Advanced and Dis-tributed Simulation (PADS rsquo12) pp 184ndash186 July 2012
[54] G Guo B Chen and X Qiu ldquoParallel simulation of large-scale artificial society with GPU as coprocessorrdquo InternationalJournal of Modeling Simulation and Scientific Computing vol4 no 2 Article ID 1350005 2013
[55] B Chen L-B Zhang X-C Liu and H Vangheluwe ldquoActivity-based simulation using DEVS increasing performance by anactivity model in parallel DEVS simulationrdquo Journal of ZhejiangUniversity Science C vol 15 no 1 pp 13ndash30 2014
[56] X-L Wang P Yang Z-D Cao et al ldquoQuantitative evaluationon the effectiveness of prevention and control measures againstpandemic influenza A (H1N1) in Beijing 2009rdquoChinese Journalof Epidemiology vol 31 no 12 pp 1374ndash1378 2010 (Chinese)
[57] S Mei A D van de Vijver L Xuan Y Zhu and P M ASloot ldquoQuantitatively evaluating interventions in the influenzaA (H1N1) epidemic on China campus grounded on individual-based simulationsrdquo Procedia Computer Science vol 1 no 1 pp1675ndash1682 2010
[58] X Qiu Z Cao and Z Li Mid-Term Reports of China NationalScience Foundation 91024030 2014
Submit your manuscripts athttpwwwhindawicom
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
MathematicsJournal of
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Mathematical Problems in Engineering
Hindawi Publishing Corporationhttpwwwhindawicom
Differential EquationsInternational Journal of
Volume 2014
Applied MathematicsJournal of
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Probability and StatisticsHindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Journal of
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Mathematical PhysicsAdvances in
Complex AnalysisJournal of
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
OptimizationJournal of
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
CombinatoricsHindawi Publishing Corporationhttpwwwhindawicom Volume 2014
International Journal of
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Operations ResearchAdvances in
Journal of
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Function Spaces
Abstract and Applied AnalysisHindawi Publishing Corporationhttpwwwhindawicom Volume 2014
International Journal of Mathematics and Mathematical Sciences
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
The Scientific World JournalHindawi Publishing Corporation httpwwwhindawicom Volume 2014
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Algebra
Discrete Dynamics in Nature and Society
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Decision SciencesAdvances in
Discrete MathematicsJournal of
Hindawi Publishing Corporationhttpwwwhindawicom
Volume 2014 Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Stochastic AnalysisInternational Journal of
Mathematical Problems in Engineering 7
Figure 5 The metamodels of artificial society in GME
event model and intervention model are built Actuallythe models are the instantiation of metamodels in the laststep Different from the general modelling environment likeUML [28] the domain-specific modelling provides a familiarmodelling environment for the domain experts in artificialsociety For example emergency response experts only con-cern emergent eventmodel and interventionmodel inheritedfrom metamodels After constructing the domain-specificmodels based on domain-specific metamodels domain usersexecute the model transformation defined in the first stepAll the models of artificial society are transformed into FSMmodels As a result the models are implemented in this uni-fied modelling formalism (FSM) The model transformationmakes the simulation of the models possible
The third step is the generation of executable codes ofmodels The executable code framework is generated bymapping template from metamodels to code frameworkdefined in the first step Moreover domain developers alsoadd necessary codes to the framework to integrate thedynamic semantics of the models The code frameworkoutputs the dynamic link libraries by compilingThe dynamiclink libraries are loaded in the large scale artificial societyruntime infrastructure in computational experiments
332 The Metamodelling and Modelling of Artificial Societyby GME GME is used to build metamodels and modelsin our work As mentioned before the abstraction andcommon patterns of society are represented in metamodelsAccording to the bottom-up modelling style metamodels ofagent environment and communications are described inGME Figure 5 shows part of metamodels of artificial societyThe features of an agent metamodel are extracted from thecensus figures and statistical data Environment metamodelsimulates the geospatial places for the behaviors of agentsThemetamodel of communications among agents ismodeledto simulate the interactions such as infection in epidemicsand rumor propagation in public opinion formation eventsIt is worth notifying that the metamodel of communications
includes both the emergent event metamodel and interven-tion metamodel
From the perspective of modelling the details from spe-cific domains are considered in the models by the instantia-tion frommetamodels of artificial society For example socialrelationships based on complex networks are added in agentmodel to support the communications Agent activity is alsoused to quantify the agent activity under different scenarioEnvironment models are linked with the help of transporta-tion services subways and roads are modeled while the pathsearch is encapsulated in the services Emergent event modeland intervention model are also the domain-specific modelsThe modelling of artificial Beijing in GME will be discussedin detail in next section
333 The Implementation of Models of Artificial Society byMDT As mentioned before MDT is used to implementmodels such as agent environment emergent event andintervention According to the template of code frameworkthe implementations of models are generated by MDT Theimplementations are classified into two categories FSMmodels and services FSM models such as agents andenvironments are built under the specification of FiniteState Machine (FSM) [29] in MDT while all the servicessuch as transportation are encapsulated under the PublicService Standard This standard provides a generic interfacespecification for modelers to encapsulate public commonservices in artificial society FSM models like agent are builtstatistically from the quantitatively analyzed characteristicssuch as demographic attributes social behaviors emergencybehaviors and social networks Social behaviors describe thedaily behaviors of individuals while emergency behaviorsdescribe the individual behaviors in emergent events Forexample infected individuals are all isolated in hospital inSARS Isolation ismodeled as a typical emergency behavior inour work Correspondingly services are used to simulate themacroactual society Take transportation service for instancethe path search is needed by almost every agent duringmoving from spot to spot
8 Mathematical Problems in Engineering
Artificial society description in emergency
Description ofemergency
ArtificialSociety Editor
(ASE)
Agent behavior set
Agent role setAgent status set
Environment setAgent relationship set
Statistical data of real society
Descriptionof artificial
societyframework
Terminal of agent modeling andartificial society describing
middot middot middotAgent
metamodeling
Customized agentmodeling
Relationship modelDailyEmergency
Mod
el D
evelo
pmen
t Too
l (M
DT)
Agent modeling
Agent model Encapsulate
Encapsulate
Behavior model
Intervention
Emergency
Response
Daily
Social relationshipmodeling
Socialbehaviormodeling
Socialbehaviormodel
Emergencybehavior
model
Emergencybehaviormodeling
Building
Geospatial statistical data
Population statistical data Datacollection
Actualsociety
Artificial societypopulation and
geospatial database
Organizationservice
Transportationservice+
MappingAgent model andservice repository+ emergency model
repository
Climateservice
Geographicalservice
+
Agent status in emergencyEmergency management
Emergent event set
Social relationship statistical data
Environment statistical data
Population behaviorstatistical data
Emergency statistical data
Emergency organization statistical data
Geospatial and SocialEnvironment Generation
Tool (GSeT)
Agent model repository andpopulation database
C++C++C++C++
Modeling
Statistical modelJob ageSex location
Public service modeling
Organizationmodel
Transportationmodel
Geographicalmodel
Climatemodel
Interventionservice
Publ
ic se
rvic
e sta
ndar
d
Figure 6 The editing and initialization of KD-ACP
MDT provides domain experts with a tool to obtain thecode implementations of models With the help of compileenvironment like visual studio MDT also supports thefurther programming development of the specific domaindetails which cannot be described in modelling step
Both the FSM models and services are developed by theMDT first and then stored in the agent model and servicerepository The repository manages the models according tothe requirements from emergency problems and provides themodels for EDT to make the experiment plans
34 The Editing and Initialization of Artificial Society byASE and PGET As shown in Figure 6 ASE is used to editthe scenarios of artificial society within emergent eventsThe editing is composed of two parts (1) the statisticalinformation of artificial society in daily life such as the rolesof agents the relationships of agents and the types of environ-ments and (2) the statistical information of artificial societyin emergency including the statistical data of emergent
events the emergency organization and emergency relatedbehaviors of agents
According to the requirements of the editing thesestatistical data are collected from actual society manually bythe domain experts Based on these statistical data PGETgenerates the artificial society population and geospatialenvironment database The database supports the instanti-ation of artificial society at individual level For examplethe attributes such as age and gender of each agent can befound in the database With the support of the databaseFSM models and service repository discussed before itis sufficient for domain experts to study the emergencyproblems by computational experiments
35The Computational Experiments and Parallel Execution inKD-ACP The tools of ldquoCrdquo part and ldquoPrdquo part in KD-ACP areused to support the process of computational experimentsand parallel execution The working process is shown inFigure 7 EECT and IMCT are both the starting and ending
Mathematical Problems in Engineering 9
Agent model andservice repository
repository
Emergency EventsConfigurationTool (EECT)
Open Source DataRegistration Tool (OsdRT)
Intervention MeasuresConfigurationTool (IMCT)
Load
Output Output
Internet
Emergencyresponse plans
Load
∙ Input models of emergencyevents∙ Output settings of agentbehavior in emergent events+ settings of emergent events
∙ Input models of interventionmeasures∙ Output settings of agentbehavior in interventionmeasures + settings ofintervention measures
+ emergency model
∙ Input data from internetnetworks∙ Output population data +emergency events data +intervention measures data
Artificial societypopulation and
geospatialenvironment database
Artificial societyruntime database
OutputThe most optimal
emergencyresponse plan
Actual society
OutputOutputOutput
Input
Load
Load
Load
Load
Output
Artificial SocietySituation Tool (ASST)
Emergency decisionorganization
ClusterRuntime infrastructure
Deploy
Deploy
Experimentsdesign tool (EDT)
Artificial societyinitial data file
Experimentplan
Artificial societymodels
Experiments Management
large scale artificial societyruntime infrastructure
Tool(EMT) +super computerTIANHE-1A
Load
∙ Input experiment plan +artificial society models +artificial society initial data file∙ Output artificial societyruntime state data
∙ Input description of artificialsociety + artificial society populationand geospatial environment society+ emergent events configuration +
intervention measures configuration∙ Output experiment plan + artificialsociety models + artificial societyinitial data file
∙ Input runtime statistical data∙ Output graphics charts ofartificial society statistical data +situation of artificial society
Figure 7 The computational experiments and parallel execution of KD-ACP
point The emergent events and intervention measures areconfigured by EECT and IMCT respectively The configu-rations of emergent events are used to simulate both thereal emergencies like SARS and H1N1 and the supposedemergencies for experiments Similarly the configurationsof intervention measures are also used to reproduce thereal one and simulate the supposed one The repeat of theemergency is used to verify the models while the supposedconfigurations are used to obtain the optimized decision planto the response of the possible emergencies
With the input of artificial society model and servicerepository artificial society population and geospatial envi-ronment database and the configurations discussed beforeEDT generates the experiment plans to meet the require-ments of research on emergency management The outputof EDT includes artificial society models artificial societyinitial data files and experiment plan The models aredownloaded from the repository while the data file is thecollection of data from the database to initialize the modelsWhen the models and data files are ready EMT loads theexperiment plan and deploys the models and data to thecluster or TIANHE-1A supercomputer [30] which was theworldrsquos fastest supercomputer built by National University of
Defense Technology (NUDT) in China in 2010 According tothe plan the experiments are done repeatedly on the largescale artificial society runtime infrastructure [31 32] by themultisample settingsThework process is the implementationof computational experiments in ACP approach
Traditionally emergency response plans are made byemergency management theories and experiences The onlyway to test the effective of plans is the feedback results of realworld ACP approach provides a novel method to supportemergency response plan making by parallel execution Asshown in Figure 7 the work process of KD-ACP is composedof two loops The inner loop composed by red arrowsdescribes the process of computational experiments while theouter loop of yellow arrows illustrates the process of parallelexecution During the runtime of computational experi-ments the statistical data of artificial society is collected andstored in the artificial society runtime database Based on thedatabase ASST outputs the customized situation of runningartificial society by graphics charts and situation mapsThe information is sent to the organizations of emergencydecision to support making the emergency response plansWith the help of computational experiments loop these plansare simulated repeatedly to find the most optimal one
10 Mathematical Problems in Engineering
Open Source Data Registration Tool (OsdRT)
Emergent EventsConfiguration Tool (EECT)
Intervention MeasuresConfiguration Tool (IMCT)
∙ Web information collecting
∙ Deep mining in social media
∙ Web information denoising
∙ Information filtering
Data acquisition
Actual society
Open source informationin social media
Internet
∙ Basic element extracting∙ Individuals and organizationsextraction∙ Sentiment analyzing andopinion mining∙ Social network analyzing
Data extraction
∙ Domain knowledge description∙ Information fusion and collisiondetection∙ Domain knowledge construction∙ Social media informationstandardization
Data standardization
Knowledge
Figure 8 The compositions of OsdRT [34]
Moreover the most optimal plan is used to the responseof emergency in actual society According to the idea ofparallel control in [24] the feedback of actual society ispartly collected from Internet networks by OsdRT As shownin Figure 8 OsdRT is composed of three components dataacquisition data extraction and data standardization Dataacquisition collects mines and filters information fromsocial sensing networks Data extraction includes basicelement extracting individual and organization extractingsentiment analyzing and social networks analyzing Datastandardization specifies the useful knowledge and sent it toconfiguration tools in KD-ACP
By processing in OsdRT the knowledge about emergentevents and intervention measures are analyzed first andregistered in the EECT and IMCT The registration updatesthe settings of emergent events and intervention measuresThis loop composed of yellow arrows implements the parallelexecution in ACP approach The implementation of parallelcontrol and management provides a data-driven approachthat considers both the engineering and social complexityfor modelling analysis and decision making in emergencymanagement
4 Modelling Beijing City with KD-ACP
41 How to Build the Artificial Beijing According to themodelling of artificial society in KD-ACP discussed beforethe Beijing city is modeled as follows
To meet the requirements of emergency managementsthe basic elements of artificial society are extended Asshown in Figure 9 six elements are required to simulate thecity agents environments transportation activity schedulecommunication and agent activity
411 Modelling Artificial Beijing Figure 10 shows the mainGUI of GME for the modelling of artificial modelling inpublic health events Metamodels are listed in the left area ofFigure 10 the list provides basic syntax elements for domainexperts to model artificial society Domain experts build
models based on the knowledge of their own MeanwhileGME supports hierarchy for building large scale systemsThesyntax symbol listed in GUI can be extended in new tab bydouble clicking Take agent for example the model of agentcan be detailed by edition in another tab page of agent
As shown in the center of Figure 10 the models ofartificial Beijing consist of five parts models of agent andenvironment domain models of public health events inter-ventionmodels controllermodels and services Agentmodeldescribes individuals in society it is composed of basicpopulation information action social relationships activityschedule and disease related information Activity schedulerepresents individualrsquos physical actionmodel focusing on thedaily action of agents Environment model includes physicalentities such as buildings playground transportations andagents contained in environment Domain models of publichealth events are composed of the propagation model of dis-ease disease state transition model and so on Interventionmodels include the settings of vaccination isolation and soon The models mentioned before are all FSM models Themechanisms of these models will be detailed in next sectionsController models and services are the public service mod-ules they are implemented in the development in MDT
412 Modelling Agents and Environments Under the speci-fication of FSM agent and environment models are imple-mented in two parts the state space and state transitionsThe state space is composed of the demographic attributesand behavior related attributes The transitions are triggeredwhen the conditions of states are satisfied As shown inFigure 11(a) the action of agent is changed when the ldquonexttimerdquo condition is satisfied in agent model while the agentslist is changed when the agent arrival condition is satisfied inthe environment model
413 Modelling Activities Agent activities come from theagent state transitions of actions such as movements andcommunications The actions of agents are instructed by theactivity schedule shown in Table 2 Activity schedule lists
Mathematical Problems in Engineering 11
Transportation EnvironmentsEnvironments
Agents
Artificial Beijing
Agent activityActivity schedule
Agent activity
Communication by socialrelationship networks
Communication by socialrelationship networks
Figure 9 The basic elements of artificial society
Syntax elements of DSM
Services
Models of agentenvironment andactivity schedule
Domain models of publichealth events
Intervention models
Modelingaspect ofdomainexperts
Controllermodels
Figure 10 The models of artificial Beijing in GME
all the actions with probability in one day for agents inboth normal and emergent situation [38] There are severaltypes of activity schedule in artificial Beijing student agentactivity schedule worker agent activity schedule emergentagent schedule and so on For example Table 2 gives an agentactivity schedule Upon the instruction of activity schedulestudent agent changes the actions by 119901
119894after state transitions
The119901119894in the tablemeans the action probability in the relevant
period In the duration from 0800 to 1200 a student agenteither goes to classroom to have class or goes to libraryto study The probability of class action is 119901
2while the
probability of study action is 1 minus 1199012
Agent behaviors are decided by the settings of activityschedule In addition to the daily activity schedulementionedbefore emergent activity schedules are also considered inour work Take public health events for instance an infectedagent changes schedule from a normal one to emergentone The workflow of a susceptible agent is illustratedin Figure 12 to show the change of behaviors After theinfection the agent is set in incubation phase Not all theincubation agents will become symptomatic Some of themturn back to being susceptible and some of them becomesymptomatic The symptomatic agents change their activityschedule from normal to emergent In the emergent case
12 Mathematical Problems in Engineering
Next actionNext environment (location)
Advance to next actiontime
Find next environment inactivity chain by action
Current actionCurrent environment (location)
Q
q0
120575 Find next action inactivity schedule by time
F
Σ
(a) The state transitions of agent
Q
q0
120575 F
Current agent list
Current pollution level
Next agents list
Next pollution level
Agents entered
Infected agents entered
Add agents to currentagents list
Compute pollution levelby infected agents
Σ
(b) The state transitions of environment
Figure 11 The state transitions of models
Table 2 The activity schedule of a student agent
Duration (Δ119905) Activity Location Probability 119879Location (minute)0000ndash0600 Sleep Dormitory 119875
0(100) 360
0600ndash0800 Breakfast Dormitoryrestaurant 1198751(068) 120
Sports-breakfast-travel Playground-dormitoryrestaurant 1 minus 1198751(032)
0800ndash1200 Class Classroom 1198752(077) 240
Study Library 1 minus 1198752(023)
1200ndash1400 Lunch Restaurant 1198753(090) 120
Lunch-shopping Restaurant-convenience store 1 minus 1198753(010)
1400ndash1800 Class Classroom 1198754(070) 240
Study Library 1 minus 1198754(030)
1800-1900Dinner Dormitoryrestaurant 119875
5(063)
60Sports-shopping Playgroundconvenience store 1198756(025)
Study Classroomlibrary 1 minus (1198755+ 1198756) (012)
1900ndash2200Rest Dormitory 119875
7(045)
180Dinner-rest Dormitoryrestaurant-dormitory 1198758(020)
Study Classroomlibrary 1 minus (1198757+ 1198758) (035)
2200ndash2400 Rest Dormitory 1198759(072) 120
Sleep Dormitory 1 minus 1198759(018)
infected agents go to hospital according to the treatmentschedule or stay in dormitory according to isolation scheduleAfter the treatment in hospital or self-immunoprocess agentsbecome healthy and immune of disease If the agents aretreated in the hospital they are not allowed to get out untilrecovered Moreover activity schedules are also influencedby the emergency response plans For example in the case ofisolation in emergency response plans the activity scheduleof agents who had contacted with infected is changed Onlythe locations including home and dormitory compose thisemergent schedule Likewise based on the statistical datafrom emergency response plans the additional possibilities
are added into activity schedulesThe behaviors of agents alsochange with activity schedules In the view of ACP approachthe injection of new emergency response plans implementsthe parallel execution of microagent behaviors
In another aspect contact frequency is also anothercrucial element to determine the infected rate in public healthevent It is important to model the contact frequency andcontact time of individuals Based on the studies on thecontact behavior of human being by questionnaire surveyEdmunds found that the contact frequency of individualcould be fitted approximately into a normal distributionand the mean and the standard deviation distributions are
Mathematical Problems in Engineering 13
Susceptible agent ImmuneYes
Healthy and immune
No
Choose schedule
Self-isolation Go to hospital
Staysymptomatic
No
Receive treatment
Contact withinfected agents
Symptomatic
No
Self-immunoprocess
Yes
Recovered
Yes
Incubation
Infection
No
Yes No
NoYes
Figure 12 The workflow of a susceptible agent
168 and 85 [39] So it is possible to apply normal randomvariable to model the contact frequency In our work Box-Muller method [40] is used to generate the random variableof contact frequency shown in the following
119865119862= 120583119865+ 120590119865(minus2 ln (120574
1))12 cos (2120587120574
2) (1)
in which 119865119862is the random variable of contact frequency 120583
119865
is the mean value of normal distribution 120590119865is the standard
deviation of normal distribution and 1205741and 120574
2are the
uniform random variables distributed in the interval [0 1]Based on (1) and survey data [41] the contact frequenciesof agent 119865contact(119860 119894) are discretized as in Table 3 within theconsideration of activity differences
Similarly the duration per contact between individualsis another key factor of AHC transmission The duration ofcontact could also bemodeled by the normal randomvariablein the following [42]
119879119862= 120583119879+ 120590119879(minus2 ln (120574
1))12 cos (2120587120574
2) (2)
in which 119879119862is the random variable of duration per contact
120583119879
is the mean value of normal distribution 120590119879
is thestandard deviation of normal distribution and 120574
1and 120574
2
are the uniform random variables distributed in the interval
[0 1] In addition because the time spent in specific location119879Location(119860 119894) also follows the random distribution like (3)in our work In order to make the duration of contact119879contact(119860 119894) shorter than the 119879Location(119860 119894) the mean andstandard deviation of (2) is set up in (3) and (4) [42]
120583119879=119879Location(120583119865+ 120590119865) (3)
120590119879=119879Location(10120583119865+ 10120590
119865) (4)
in which 119879Location is the time of an activity in a specificlocation set in Table 2 120583
119879and 120590
119879are the mean and standard
deviation of contact frequency As a result the duration ofagent 119879contact(119860 119894) in specific locations (listed in Table 2) isalso discretized in Table 3 within the consideration of activitydifferences
414 Modelling Transportation In artificial Beijing threekinds of travel models are considered walking inside thedistrict travelling by road networks and travelling by subwaynetworks According to the traveling models the agentmovements are implemented by the compositions of travelingmodels before So it is important to build the basic road
14 Mathematical Problems in Engineering
Table 3 The contact frequency and duration in different locations and activities by social relationships in case of student and teacher
Social relationship Activity Location 119879contact(119860 119894) (minute) 119865contact(119860 119894)
Student-student
Sleep Dormitory 0 0
Breakfast Restaurant 119873(52) 119873(32)Breakfast Dormitory 119873(52) 119873(21)Sports Playground 119873(21) 119873(53)Class Classroom 119873(21) 119873(21)Study Library 119873(21) 119873(21)Lunch Restaurant 119873(158) 119873(32)
Shopping Convenience Store 119873(203) 119873(21)Dinner Restaurant 119873(108) 119873(21)Dinner Dormitory 119873(158) 119873(32)Rest Dormitory 119873(63) 119873(52)
Teacher-student Class Classroom 119873(21) 119873(21)
Subway network of Beijing city (red lines) Road network of Beijing city (green lines)
Figure 13 The implementation of transportation for artificial Beijing
and subway networks With the help of Google map thekey points of road are sampled first and the road networksare generated by the links among point Similarly subwaynetworks are generated by the links between subway stationsmapped from the real stations in Google map Figure 13shows the road and subway networks in artificial Beijingwith the road networks in green and the subway networks inred Most of the districts in Beijing city are covered by thesetwo networks The algorithm of hierarchical route planningis proposed as follows
(i) Find the next position where agent will be located innext action
(ii) Obtain the travelingmodels from the statistical trans-portation data
(iii) Search the path from the starting position to thenearest road entry or subway stations
(iv) Search the optimal path from the starting station orroad entry to the nearest station or road exit of targetposition
(v) Search the path from the target subway station or roadexit to the target position
(vi) Attach the subway train number or the bus numberincluding the transfer information to the path accord-ing to the timetable of subway or bus
(vii) Connect all the paths obtained before Generate thepath from the starting position to the target position
Based on the algorithm for each agent in artificial Beijingthe transportation is simulated during the computationalexperiments It is worth noting that the activities insidethe buses or train cars are considered because the agentcommunication during travelling is also necessary to besimulated
Mathematical Problems in Engineering 15
(a) The implementation of agent in MDT (b) The code framework of agent
Figure 14 The agent model implementation in MDT [35]
415 The Implementation of Models by MDT All the modelsare developed in the MDT Figure 14 shows the GUI of thetool MDT is used to implement the state space and statetransitions of FSM models The code framework is shownin Figure 14(b) Using the C++ inheritance features and thetechniques of dynamic link library (dll) the models areencapsulated in the dll componentsThe flexible compositionmechanism is designed to support the models evolution incomputational experiments
416 The Description of Artificial Beijing by ASE As men-tioned in Section 34 ASE describes the artificial Beijingin macroview The statistical information of Beijing city iscollected by ASE Table 4 lists part of the description ofartificial Beijing generated by ASE
42 How to Obtain the Data of Beijing City
421 Population and Geospatial Environment Database ofBeijing City Based on the models developed for ArtificialBeijing it is necessary to generate the initial data to do thecomputational experiments Because the individual level dataare not available it is necessary to construct an individual-based population database for both accurate computationalexperiment and determining optimal decisions
According to the state spaces of models discussed inSection 412 the geospatial and population database of Arti-ficial Beijing is designed as shown in Figure 15 The kernelpart of database consists of the tables such as agent list tablegeospatial environment list table household list table andagent distribution table Agent list table is an individual leveltable used to store the data to initialize the state space of agentmodel such as id and gender age Geospatial environmentlist table is also an individual level table which stores thedata of environment attributes such as id street id andtype Household list table and agent distribution table storethe statistical data for the data generation for agent andenvironment models With the help of the geospatial andpopulation database the data generating of artificial Beijingis proposed in the next section
422 Generating Geospatial and Social Environment Datafor Artificial Beijing by PGET The algorithm of generatinggeospatial and social environment data for artificial Beijingis used to quantify the spatial distribution of population andformalize geospatial behavior of each agent The algorithmshown in Figure 16 is capable of generating a syntheticartificial society which allowsmultiresolution statistical datasocial interactive behavior and multilayer social networksto be integrated together The synthetic population canrepresent individual agents in the form of households andhousehold members and the synthetic population is statis-tically equivalent to a real population For each householdcharacteristics such as address household size family typesand relationships are generated Each person is described bycharacteristics such as age gender social role and correlatedlocations The algorithm provides an effective methodologyto reconstruct the computing environment in high resolutionby using statistical data in low resolution leading to betterprediction and management of emergencies The details ofalgorithm are illustrated in [36]
As shown in Figure 17 the implementation of the algo-rithm is embedded inside the GEPTThe collective input datasuch as the data files of population distribution is collected inthe tool first then the algorithm is activated to generate thedatabase The generation lasts almost twelve hours one timewith the specified parameters settings When the generationis finished the data of artificial Beijing is obtained in thedatabase with 19610000 agents and 16000 environments
5 H1N1 Experiments in ArtificialBeijing with KD-ACP
With the support of the artificial Beijing modeled before thecomputational experiments and parallel execution are alsoimplemented by KD-ACP The experiments are designed tofind the most optimal emergency decision response plansEpidemic in city is a typical scenario in emergency manage-ment H1N1 epidemic in Beijing in 2009 is used to be the usedcase to test KD-ACP According to Section 34 it is necessarytomodel emergencymanagement first in order to support thecomputational experiments
16 Mathematical Problems in Engineering
Table 4 The description of artificial Beijing generated by ASE
Statistical features ofartificial Beijing Descriptions of statistical features Data sourcedefault values
Agent roles Baby child primary student middle school student college student worker retired
Demographic data file ofBeijing
Environment types Super market store restaurant park stadium hospital school cliniccommercial building dormitory building residential building
Demographic data file ofBeijing
Demographic data Age distribution of population sex distribution of population householddistribution of population retired age distribution children distribution
Demographic data file ofBeijing
Environment data Distribution of all types of environments Environment distributiondata file of Beijing
Relationships Classmate relative college friendship family Demographic data file ofBeijing
Varying parameters
Age difference of couples119873(120583120590) 119873(03) isin [0 12]Minimummarriage age (male female) (22 20)Age difference of children119873(120583120590) 119873(13) isin [1 10]Enrollment rate of kindergarten 120572 08Enrollment rate of primary school 120573 1Enrollment rate of middle school 120574 1Enrollment rate of college 120575 08Entrance age of kindergarten 1205721015840 3Entrance age of primary school 1205731015840 7Entrance age of middle school 1205741015840 13Entrance age of middle school 1205751015840 19
Age distributioninformation table
Agent settable
Geospatialenvironment set table
Household settable
Figure 15 The geospatial and population database of Beijing city
51 Modelling Emergency Management The modelling ofemergency management consists of two parts the modellingof emergent events and the modelling of intervention mea-sures H1N1 and the intervention measures for H1N1 aremodeled in our case
511 Modelling H1N1 by MDT H1N1 Model in agent simu-lates the states transitions of health status and the relevantactions Referring to SIR (SusceptibleInfectiveRecovered)[43 44] models agent has three health statuses susceptibleinfected and healthy with immunity Only susceptible agentcan be infected by the contact with infected agents After
the three infected phases (incubation being symptomaticand recovery) of H1N1 model agent is healthy again It isworthy mentioning that the recovered agents can be infectediteratively when the immunity disappears gradually by timeThe state transitions are illustrated in Figure 18
According to [45] the latent period of H1N1 was in aWeibull distribution within people from one to seven days[46 47] The distribution was usually centered in the rangebetween one and three days AWeibull random variable [40]is used to model the latent period as described in
119879Lat = 120572 times [minus ln (1 minus 120574)]1120573+ 120592 (119879Lat ge 120592) (5)
Mathematical Problems in Engineering 17
Censusdata
Generate GUID
Assign individualhousehold
Generate familystructure
Assign family roleto every member
Endow individualwith age gender
social role
Distribute environmentlocations on the map
School Workplace Entertainmentlocations
Consumptionlocations
Assign school orworkplace to
individual
Correlatedindividual withother locations
Generate friendshipin school
Generate coworkerrelations inworkplace
Assign individualadministrative
region
Generate housebuildings
Locate householdto house building
Generate familyrelation
Generateneighborhood
relation
PopulationEnvironment
Environment listSocial network
Figure 16 The algorithm of generating geospatial and social environment data [36]
Edit
Edit
Edit
Edit
Edit
Edit
BrowseStatistical characteristic of population
Geospatial population data
Population data of street
Population types
Geospatial and social environment generation tool
Workspaces
Parameters distribution settings
Household types
Roles in household
Household structure
Social roles
BrowseBrowseBrowseBrowseBrowseBrowse
Population distribution data
File path
Retired distributionPop distributionHousehold numHousehold distributionSex distributionAge distributionData type
Figure 17 The GUI of Population and Geospatial Environment generation Tool (PGET)
18 Mathematical Problems in Engineering
Infected
SusceptibleHealth
withimmunity
Infected
by contac
t
with in
fected
agents
Recovery
Increase before beingsymptomatic symptomatic
Decrease after being
Decrease by time
Infectivity
Immunity
Figure 18 The transitions of agent health status
119879Lat denotes the duration of latent period 120592 120572 and 120573 arethe location parameter scale parameter and shape parameterof Weibull distribution respectively 120574 is a uniform randomnumber in the range [0 1] According to statistics [47] 120592 120572and 120573 are set as 0 18 and 121 respectively Then the mean(standard deviation) of latent period is calculated as 159 days(058 day2) Furthermore infectious period is set as 7 days
A susceptible studentrsquos probability of getting infecteddepended on the infectivity of infectious agents his ownimmunity level the duration of the contact action and soon The infectivity of an infected agent evolved with timeduring the infectious period According to the statistics ofinfectivity in the chart with red bars in Figure 19 an infectedstudent had the highest level of infectivity in the second dayafter he had the first symptom The infectivity levels in otherdays in comparison with the second day are in Table 5 Theday labeled as ldquominus1rdquo means the day before the starting of thesymptomatic period
Based on the algorithm discussed before theH1N1modelis developed with the help of MDT The settings of influenzamodel will be discussed in Section 521
512 Modelling Intervention Measures by MDT In corre-spondence intervention measures model is used to ceasethe outbreak of H1N1 As discussed in [42 48 49] if theappropriate measures are taken when the infectious diseaseemerges the transmission of the disease could be sloweddown and the damage could be decreased Therefore it isnecessary and important to design emergency interventionmeasures in the artificial Beijing According to the MinistryofHealth inChina the intervention policies [50] are designedto control the spread of influenza such as H1N1 The inter-ventionmeasures of these policies including are interventionsactivated time vaccination rate antibiotic delay to hospitalisolation duration close workspace duration and limitationof activities listed in Table 6 Similarly the models of theseintervention measures are also developed with the help ofMDT Using the reasonable data ranges listed in Table 6intervention models can be initialized to do the computa-tional experiments These experiments are used to show how
Symptomatic1 2 3 4 5 6 7minus1
Recovered
Tim
e (da
y)
Infe
ctiv
ity
Figure 19 The distribution of infectivity rate of an infected agent[37]
Table 5 The proportion of infectivity
Days (119894) minus1 1 2 3 4 5 6 6Proportion of infectivity 03 05 1 08 04 02 01 005
to restrain the H1N1 transmissionThe settings of these inter-vention measures in IMCT are introduced in Section 522
52 Experiment Settings
521 H1N1 Settings by EECT EECT is designed to initializethe emergent events models such as H1N1The settings comefrom the specified emergent event models As shown inFigure 20 in theH1N1 used case the settings are composed ofthe distribution of infection source infection rate infectionperiod and so on The infection rate changes with envi-ronments and infection periods The settings are initializedfrom both the research of medicine on H1N1 and the data ofH1N1 outbreak in Beijing in 2009The process of influenza inartificial Beijing is simulated to repeat the possible scenariosof H1N1 break During the computational experiments ofartificial Beijing theH1N1 settings such as the period of beingsymptomatic are updated by the dynamic data registrationsfrom OsdRT The updating corrects the H1N1 model duringthe experiments With the help of OsdRT artificial Beijing isable to approach actual society It is also an implementationof parallel execution
522 Intervention Measures Settings by IMCT In the H1N1case the settings of intervention measures are shown inFigure 21 in IMCT These measures are designed under theinstruction of Emergency Decision Organizations like ChinaDisease Center (CDC) The measures such as vaccinationrate the rate of closedworkspace and the starting time can beinitialized in the tool Emergency response plans are imple-mented in these settings As illustrated in Figure 7 all the pos-sible compositions within value settings in reasonable inter-vals are tested to find the most optimal emergency responseplan For example different compositions of measures suchas vaccination rates and rates of workspace closed are usedto design the plans of experiments By the analysis of theresults the best composition of interventionmeasures can beobtainedThe analysis will be discussed in case study in detail
Mathematical Problems in Engineering 19
Table 6 The intervention measures of influenza
Intervention measures ValuesInterventions activated Time 119879Intervention isin [0 80th day]Vaccination rate 119877Vaccination isin [0 1]Antibiotic rate 119877Antibiotic isin [0 1]Delay to hospital 119867ospital(119860 119894) 119879infected lt 119905 lt 119879recovered + 119879HDelay 119879HDelay isin (0 3th day]Isolation duration 119868solation(119860 119894) 119879Intervention lt 119905 lt 119879Intervention + 119879IDuration 119879IDuration isin (0 14th day]Close workspace duration 119862lose(119864119894) 119879Intervention lt 119905 lt 119879Intervention + 119879CWDuration 119864119894 isinWorkspace 119879CWDuration isin (0 7th day]Limitation of agent activity duration 119871 imitation(119860 119894) 119879Intervention lt 119905 lt 119879Intervention + 119879LAADuration 119879LAADuration isin (0 7th day]119877Vaccination means the vaccination rate of artificial Beijing while 119877Antibiotic means the antibiotic rate used for infected agents in hospital119867ospital (119860119894) means that the infected agent 119860119894 is sent to hospital he or she will receive the retreatment immediately119868solation (119860119894) means that the infected agent 119860119894 is isolated at home he or she can only contact with the other agents until isolation duration is over119862lose (119864119894) means that the environment 119864119894 typed of workspace is closed until the close workspace duration is over119871 imitation (119860119894) means that the activities of agent 119860119894 are limited only the workspace and home are allowed until limitation duration is over
Incubation Weibulldistribution
Symbolic LogNormaldistribution
Recovering LogNormaldistribution
H1N1influenza
Settings of H1N1 influenza model
Disease model
Transmissionmodel
Figure 20 The settings of H1N1 in EECT
It is worth notifying that the settings of interventionmeasures are also influenced by OsdRT For example theH1N1 outbreak areas are notified by OsdRTThe interventionmeasures such as close workspace are used only in theoutbreak areas Therefore the unnecessary computation ofintervention measures is avoided So the data injection byOsdRT not only makes artificial society approach actualsociety but also increases the performance of computationalexperiments
53 Experiments by EMT When the initialization of arti-ficial Beijing is ready EMT is used to do the computa-tional experiments EMT is composed of two parts thecentral controller in the central node and the residen-tial service deployed in the computational nodes Centralcontroller loads the experiment plans and controls the whole
experiments process the GUI of controller is shown inFigure 22 EMT manages the computational experimentsin four steps cluster configuration load experiment plansmodels deployment and running experiments
531 Cluster Configuration Before the configuration clusterinformation is collected from the residential services insidethe cluster The information includes nodes amount nodenames node IPs and CPU occupations It is used to quantifythe computing power of the clusterThe cluster configurationis based on this information The nodes are selected forexperiments Then the LP (local process) [51] number foreach node is set according to the number of the CPU kernelsAfter the configuration the experiment plans are loaded tocustomize the models and data
20 Mathematical Problems in Engineering
Streets of close workspaces
Limitations ofactivities
IsolationsRespirator
Closeworkspace
Antibiotic
VaccinationHospital
Surveillance
Figure 21 The settings of intervention measures in IMCT
Information ofexperiment plan
List of experiment plans
Start | Pause | Reset | Stop | Speedup | Slowdown | Back
NodeName | CPU occupation | Memory occupation |Network | Models | Status
Figure 22 The deployment and running of experiments in EMT
532 Load Experiment Plan As discussed in Section 35 thedescription of experiment plans is composed of the infor-mation in three aspects models and the initial data for theexperiments the deployment mapping tables from the mod-els to the nodes and the plan of experiment executionWhenthe experiment plan is loaded by the EMT the file namesand model descriptions of models and data are listed Withthe support of cluster configuration models can be mappedto the nodes in the GUI according to the computationrequirements EMT also provides the setting of experimentsitself including the running times start time and end time
533 Models and Data Deployment Models are stored inthe agent population and services repository According tothe settings in experiment plan the customized models aredownloaded from the repository first then the models suchas agent model environment model emergent events modeland intervention measures model are integrated to be thecompositionmodels of artificial BeijingThemodels and datalocations in the nodes are configured in the ldquomodel pathrdquoand ldquodata pathrdquo in the GUI of central controller Finallythe models are uploaded to the nodes in cluster with theconsideration of partition The partition of artificial society
Mathematical Problems in Engineering 21
models is different from the multiagent system in [52]Models in artificial society are divided into three levels agentsand environments grids and city Agents and environmentsare encapsulated in grids which compose the whole cityThus the partition of artificial society becomes the partitionof grids The parallelism degrees are found to support thepartition of models agent independence and grid indepen-dence According to the parallelisms agents are partitioned toLPs according to the population distribution in grids Agentmovements among grids are simulated by the movementevents among grids in different LPs As a result a two-tierparallel architecture is proposed to support the simulation ofartificial society The architecture is detailed in [53]
534 Running of Experiments When the preparation isready the experiments can be done by the central controllerof EMT The running and control commands of the experi-ments are grouped into three aspects start pause or stop theexperiments get the running information of the nodes suchas the snapshot of the desktop of node With the help of thesecommands users can run the computational experimentsto do the research on the emergency management It isworth notifying that the simulation engine is optimized inorder to support the ten million agent simulation UsingGPU as coprocessor a two-tier parallel simulation engine isdesigned with support of MPI and OpenCL through phasedsynchronization [54] One-sided communication is used forreflection of remote simulation objects and message passingbetween processes A general kernel function prototype iselaborately designed and conditionally compiled for execu-tion on both CPU and GPU Moreover the optimizationoperations like load balance is developed under the instruc-tion of activity-based simulation [55] The densely populatedgrids are given more CPU and memory resource accordingto the activity predictions
54 Results of H1N1 Experiments with KD-ACP On thebasis of artificial society models and the intervention mea-sures mentioned in Section 512 a series of computationalexperiments are performed to study the H1N1 influenza inBeijing In our case of artificial Beijing 19610000 agentsare simulated in the cluster within 48 CPU cores and 128Gmemories the nodes are connected by kilomega networks Ittakes 18 hours to simulate a 250-day disease spreading
In order to illustrate how the optimal plans are obtainedbyKD-ACP the experiments are divided into four groups theexperiments ofmodel validation the experiments of sensitiv-ity analysis of vaccination rate the experiments of sensitivityanalysis of isolation and the experiments of the combinationof interventionmeasuresThese four groups show the generalprocess of the research on emergency management in H1N1influenza Firstly the models of H1N1 are validated withthe support of historical data Secondly the traditionalmedical interventions of pandemic such as vaccination areanalyzed to find the optimal plan Thirdly the traditionalnonmedical interventions of pandemic such as isolation areanalyzed to find the optimal plan Finally the combination ofinterventions is tested to show the combined effect
It is worth mentioning that all the experiments in ourwork are performed 100 times These experiments are ini-tialized with different settings of interventions in IMCT Theresults shown in the figures are the mean values
541 The Experiments of Model Validation In our artifi-cial Beijing the models of H1N1 influenza are built basedon the existing researches Wang and her colleagues hadstudied the H1N1 influenza by SEIR (SusceptibleExposedInfectiveRecovered) models [56] With the support of theirwork we build the models of artificial Beijing in the mannerof multiagents
In Figure 23 the control group shown in red line isdrawn based on the realistic statistical data collected in H1N1influenza in Beijing in 2009 The influenza lasts more thansix months and more than 170000 people are infectedComparing with the control group the simulated resultswithout interventions are drawn in blue It is obvious that thesimulated case fits the control group well before 70 days Butthe ascending of control group is slowed after 70 days andthe peak value (17883) is much less than the simulated results(25135) It is because the vaccination is activated at day 67 incontrol group according to the intervention measures usedin 2009 The restrain by vaccination becomes effective eightdays later The change of infected number after vaccinationis shown by red spotted line The trend of control group isdifferent from the simulation data after 75 days
However the simulated results (blue line) in compu-tational experiments fit the control group (as shown inFigure 23) in the first 70 days when no intervention mea-sures are executed The key features such as effective basicreproduction number (119877
0) are also almost the same in both
cases Therefore it can be concluded that the models ofartificial society built in microview are validated by therealistic statistical data in influenza in Beijing in 2009
In this group of experiments no interventions are exe-cuted in order to validate the models Therefore the simu-lated results of this extreme case which cannot happen todaycould be obtained As shown by blue line the number ofinfected agents without interventions reaches peak value atday 91 with 25135 consisting of 8052 in symptomatic phaseand 28950 in recovery phase It is obvious that the totalnumber of infected agents grows slowly at the beginningtime Then the number increases quickly when the influenzabreaks and it reaches its maximum value with 383870 in theend This phenomenon can be explained as in the followingAt the beginning of influenza due to the limited numberof infected agents the propagation of the disease remainsslow but many agents become susceptible When the numberof infected agents increases rapidly the spreading becomessignificant However the number of infected agents decreasesafter day 91 It is because the disease spreading relies on thesocial networks and spatial contact networks [49] It meansthat the infected agents can only infect agents either in thesocial relationship networks or in the spatially contacted
542 The Experiments of Sensitivity Analysis of VaccinationRate According to Section 512 vaccination is a typical
22 Mathematical Problems in Engineering
0
Infe
cted
agen
ts
0 50 100 150 200 250
Statistical dataSimulation data
25000
20000
15000
10000
5000
Day d
Figure 23 The results with no intervention executed and the com-parison with realistic statistical data in Beijing in 2009
medical intervention in influenza Traditionally the effective-ness of vaccination can only be evaluated after the influenzaand the key settings such as vaccination rate are determinedby the experience It is always difficult for emergency man-agement organizations to find the most optimal vaccinationrate in advanceTherefore the sensitivity of vaccination rate isanalyzed in this group of experiments The settings are listedin Table 7 average results are shown in Figure 24
Vaccination rates are set linearly (10 30 50 and70) to test the changes of influenza The peak value ofinfected agents decreases from 25135 22392 15787 6907to 638 respectively It shows that the more the agents arevaccinated the smaller the peak value of the infected agentswill be In the case of 70 vaccination the total infected agentis only 13799 It means that the influenza in this case does nothappen due to the high vaccination rate
The change of peak values also shows an interestingphenomenon Vaccination rates increase linearly the peakvalues decrease slower than the ascending of vaccination ratesat beginning but the descending becomes more and morefaster with the linear increase of vaccination rate It meansthat the small scale of vaccination (less than 30) does notmake senseWith the increase of vaccination rate the numberof susceptible agents decreases significantly In sequencethe possibility of infections is lowered greatly Additionallyit could be found that the decrease of peak values comeslater than the increase of the vaccination rates This is quitereasonable according to Mei et alrsquos work in [57]
As a result it is concluded that high vaccination rate iseffective in the control of H1N1 influenza though costly
543 The Experiments of Sensitivity Analysis of IsolationAccording to Section 512 isolation is a typical nonmedicalintervention It is the most used intervention measure in
Table 7 The initial settings of experiments of vaccination
Parameters ValueAgent count 19610000Initial infected count 40Interventions activated time(119879Intervention)
0
Vaccination rate (119877Vaccination) 0 10 30 50 70
0
Infe
cted
agen
ts
Loading intervention of vaccination
No vaccination10 vaccination30 vaccination
50 vaccination70 vaccination
0 50 100 150 200 250
25000
20000
15000
10000
5000
Day d
Figure 24 The results of loading intervention of vaccination
the emergency management in influenza As mentionedbefore a problem also exists that the measure cannot bevalidated before the intervention Therefore the sensitivityof isolation is analyzed in this group of experiments It isworth mentioning that the isolation is usually combinedwith the hospital policy It means the infected individual issent to hospital meanwhile the people who contact withthe infected are all isolated As a result parameters delay tohospital and isolation duration are designed together in thesettings of experiment Delay to hospital stands for the periodfrom infection time to hospital time while Isolation durationrepresents the period when an agent cannot contact others ifhe has contact with an infected agent The settings of theseparameters are listed in Table 8 average results are shown inFigure 25
Firstly the parameter delay to hospital is simulated aloneto test the sensitivity The simulated results are shown insolid line in Figure 25 The peak value decreases from 6708to 2218 when the delay to hospital decreases from 15 daysto 1 day The half day change brings the 67 reductionMoreover the peak value almost approximates to zero inthe case of 05 days The influenza does not happen in thiscase It can be concluded that the earlier the infected agents
Mathematical Problems in Engineering 23
Table 8 The initial settings of experiments of isolation
Parameters ValueAgent count 19610000Initial infected count 40Interventions activated time (119879Intervention) 80Delay to hospital (119879HDelay) isolation duration (119879IDuration) (15 0) (1 0) (05 0) (15 3) (1 3) (05 3)
0 50 100 150 200 2500
1000
2000
3000
4000
5000
6000
7000
Infe
cted
agen
ts
Loading intervention of isolation
Delay to hospital = 15
Delay to hospital = 10
Delay to hospital = 05
Delay to hospital = 15 isolation duration = 3
Delay to hospital = 10 isolation duration = 3
Delay to hospital = 05 isolation duration = 3
Day d
Figure 25 The results of loading intervention of isolation
are sent to hospital the less the agents will be infectedSo the construction of effective mechanism to find infectedindividuals is important in the emergency management ofinfluenza
Secondly the parameter isolation duration is added tosimulate the isolation of interventionObviously the isolationmeasure decreases the infected agents greatly Compared tothe parameter delay to hospital of 15-day case the peak valuedecreases from 6708 to 362when the isolation is added in theinterventionThe reduction of 6346 agents comes from the 3-day isolation of agents who had contacted with the infectedagents However the decrease of parameter delay to hospitalcannot bring the similar reduction in the 3-day isolationcases According to the figure the peak value decreases from362 to 51 from 15-day case to 05-day case only 311 agents aresaved from infection As a result the sensitivity of simulatedresults does not change obviously with delay to hospital inthe case of isolation It can be concluded that isolation is amore effective measure Though it is not easy for emergencymanagement organizations to decrease the delay to hospitaldue to the current monitoring mechanism the influenza canstill be controlled by isolation
544 The Experiments of Combination of Intervention Mea-sures Isolation is not the only nonmedical intervention ininfluenza close workspace is another common measureWhen close workspace is activated the workspaces are allclosed agents can only stay home at the closed time Luckilyisolation and close workspace are independent from oneanother It is possible for emergency management organi-zations to use these two measures together So this groupof experiments is designed to obtain the effectiveness of thecombination of intervention measures The parameters areinitialized with different settings listed in Table 9
Parameters delay to hospital and isolation duration is setin the middle values (1 day 3 days) according to the lastsection According to the experience close duration is set asseven days It is known that close workspace is a really costintervention measure the activated time is a key parameterWith the fixed parameters mentioned before the changeof interventions activated time is analyzed in this group ofcomputational experimentsThe average results are shown inFigure 26
Compared with the simulated results of experimentsdiscussed above Figure 26 strongly illustrates that influenzais ceased when the intervention measures are combinedtogether The peak value of infected agents is 141 in the ldquo80thdays of close-workspace and isolation with 1 day delay tohospitalrdquo case while 25135 in ldquono interventionrdquo case 2218 inldquo1 day delay to hospitalrdquo case 203 in ldquo1 day delay to hospitaland isolationrdquo case The interventions bring a great reductionof infection the peak values decrease from ten thousandslevel to hundred level
On the other hand the peak value even decreases to 108when the execution time of close workspace is 10 days earlierThe date of maximum value of infected agents is advancedfrom the 135th day to the 108th day The whole circle of H1N1influenza is reducedThe results of this group of experimentsshow that intervention measures should be activated as soonas possible and the compositions of interventions are moreeffective in emergency management of H1N1 influenza
In summary based on the general process lots of exper-iments could be done by KD-ACP to study emergency man-agement in H1N1 influenza Many conclusions are obtainedbased on the analysis of the experiments in this sectionAccording to these conclusions themost optimal response inour case study is the case of ldquo70th days of close-workspace andisolation with 1 day delay to hospitalrdquo It is because this casehas the fewest infected agents relatively But the modelling ofH1N1 influenza is not sufficient Economic and social factorsof interventions such as the money cost by vaccination andisolation the influence of the close-workspace in society
24 Mathematical Problems in Engineering
Table 9 The initial settings of experiments of isolation and closeworkspace
Parameters ValueAgent count 19610000Initial infected count 40Interventions activated time (119879Intervention) 70th day 80th dayDelay to hospital (119879HDelay) 1 dayIsolation duration (119879IDuration) 3 daysClose workspace duration (119879CWDuration) 7 days
50 100 150 200 2500
50
100
150
200
Infe
cted
agen
ts
Loading intervention of isolation and close workspace
Delay to hospital = 1 isolation duration = 3Start close time = 70 close duration time = 7Delay to hospital = 1 isolation duration = 3Start close time = 80 close duration time = 7
Day d
Figure 26The results of loading interventions of isolation and closeworkspace
and even opinion transmission of H1N1 in internet are notconsidered in our case study Without the analysis of thesefactors it does not make sense to talk about the most optimalintervention measures However KD-ACP had provided asoftware framework and general process of the research onemergency management Supported by KD-ACP the expertsof emergency management are able to study emergent eventssuch as H1N1 influenza as detailed as possible Based on theexperiments and analysis again and again the most optimalintervention measures will be found ultimately
55 Parallel Execution by ASST and OsdRT As discussed inSection 35 ASST andOsdRT are used to support the parallelexecution of artificial Beijing The statistical data is collectedand stored in artificial society runtime database at runtimeASST downloads the infection data from the database anddisplays it in the map As shown in Figure 27 the infectedagents are distributed in the districts in BeijingThe snapshotgives the geospatial distribution of the H1N1 epidemic Thelegends of the agents in different status are listed in the left
right corner in Figure 27 Five statuses are shown in the figurehealth without vaccination health with vaccination incuba-tion being symptomatic and convalescence ASST providesa very direct feeling for the emergency experts Partly withthe situation from ASST Emergency Decision Organizationmakes the emergency response plans to the response of theemergency scenario With the help of plans the epidemicin actual society can be restrained In the meantime theinfluenza information is reflected in the Internet Accordingto Figure 8 the data acquisition data extraction and datastandardization of H1N1 influenza are processed in OsdRTOsdRT is implemented by Dr Caorsquos team from Institute ofAutomation from Chinese Academy of Sciences [58] Thetool provides not only the reports of H1N1 SARS hand-foot-mouth disease and intestinal diseases are all monitored Theweb page is shown in Figure 28 H1N1 reports from Internetare collected first Based on the data extraction of the reportsH1N1 case of distribution is obtained and shown in theincidencemap As a result H1N1 incidence data is regarded asthe knowledge which is sent to EECT and IMCTThe settingsof emergent events and intervention measures are modifiedin order to make artificial society approach actual society innext turn of computational experiments With the help ofOsdRT the loop of parallel execution process integrates ACPapproach
6 Conclusion
This work provides an integrated software framework forsocial Computing in emergencymanagement From a systemperspective KD-ACP provides a reliable flexible low costand effective platform for the scientists to do the research onemergency response problems The analysis and predictionof these problems with inherent complexity can be solvedby the repetition of possible alternative experiments onthe software framework Therefore KD-ACP which is anattempt to implement ACP approach seemed as an effectivecomputational framework to support the decision making inemergency response
Currently KD-ACP is used to study the H1N1 epidemicin Beijing in 2009 The most optimal compositions of inter-vention measures are testified to support the response tonext influenza However it is still a long way ahead beforethe automation systematization and practicalization of KD-ACP A lot of work will be carried out along several directionsas in the following
(i) The parallel execution process in KD-ACP is actuallynot strongly connected with the emergency responseorganization Traditional experience and theory stillplay a dominant role in the decision of emergencymanagement It is necessary to propose a mechanismof hall for workshop to facilitate the parallel executionloop in actual society
(ii) Section 33 introduces the principle and process ofmodelling artificial society Metamodels and modelsof artificial Beijing are built by GME Model transfor-mation and code generation are used to implementcodes from FSM based models But many details
Mathematical Problems in Engineering 25
Fangshan district Daxing district Fengtai district Xicheng district
Mentougou district Haidian districtShijingshan district Dongcheng district
Incubation
Symptomatic
Convalescence
Health withoutvaccination
Health withvaccinationLegend of agent
health status
Figure 27 The situation of artificial Beijing by ASST
H1N1
Districts of Beijing
H1N1 case ofdistribution
Incidence map
Distribution ofH1N1 reportsfrom internet
Incidence of H1N1
Fangshan districtLiangxiang town
Population 19000Incidence 135789
(1100000)
Figure 28 The implementation of OsdRT by Institute of Automation from Chinese Academy of Sciences
are still not illustrated due to the topic of thispaper The key techniques such as template of codeframework and model transformations from specificdomains to FSM will be focused and introduced indetail in our next paper
(iii) The performance of computational experiments inKD-ACP is still slower than expectation The opti-mization of simulation engine and models structureis necessary to improve the performance So it ispossible to run artificial society with the real worldin parallel
In summary KD-ACP paves a new way for scientists inboth emergencymanagement and computation simulation to
collaborate with each other in solving the emergency man-agement problems by social computing
Research Highlights
(i) KD-ACP is implemented based on ACP approach
(ii) Artificial Beijing is built with the help of KD-ACP
(iii) H1N1 is modeled to simulate the influenza in Beijingin 2009
(iv) Computational experiments testify the effectivenessof intervention measures
26 Mathematical Problems in Engineering
Conflict of Interests
The authors declare that there is no conflict of interestsregarding the publication of this paper
Acknowledgment
This work was supported in part by the National ScienceFoundation of China (nos 91024030 71303252 61403402and 91324013)
References
[1] W Mao and F-Y Wang Advances in Intelligence and SecurityInformatics Academic Press Oxford UK 2012
[2] E Bonabeau ldquoAgent-based modeling methods and techniquesfor simulating human systemsrdquo Proceedings of the NationalAcademy of Sciences of the United States of America vol 99 no3 pp 7280ndash7287 2002
[3] K M Carley D B Fridsma E Casman et al ldquoBioWar scalableagent-basedmodel of bioattacksrdquo IEEETransactions on SystemsMan and Cybernetics A Systems and Humans vol 36 no 2 pp252ndash265 2006
[4] J M Epstein ldquoModelling to contain pandemicsrdquo Nature vol460 no 7256 p 687 2009
[5] Los Alamos National Laboratory Reports about EPISIMS LosAlamos National Laboratory 2003
[6] C H Builder and S C Bankes ldquoArtificial societies a conceptfor basic research on the societal impacts of informationtechnologyrdquo RAND Report 149 RAND Corp 1991
[7] F-Y Wang ldquoComputational theory and method on complexsystemrdquo China Basic Science vol 6 no 5 pp 3ndash10 2004
[8] F-Y Wang ldquoToward a paradigm shift in social computing theACP approachrdquo IEEE Intelligent Systems vol 22 no 5 pp 65ndash67 2007
[9] F-Y Wang and S Tang ldquoArtificial societies for integratedand sustainable development of metropolitan systemsrdquo IEEEIntelligent Systems vol 19 no 4 pp 82ndash87 2004
[10] D Wen Y Yuan and X-R Li ldquoArtificial societies computa-tional experiments and parallel systems an investigation ona computational theory for complex socioeconomic systemsrdquoIEEE Transactions on Services Computing vol 6 no 2 pp 177ndash185 2013
[11] F Wang F Zeng and Y Yuan ldquoAn ACP-based approach forcomplexity analysis of E-commerce systemrdquo Complex Systemsand Complexity Science vol 5 no 3 pp 1ndash8 2008 (Chinese)
[12] F-Y Wang and P K Wong ldquoResearch commentary intelligentsystems and technology for integrative and predictivemedicinean ACP approachrdquoACMTransactions on Intelligent Systems andTechnology vol 4 no 2 pp 1ndash6 2013
[13] J Sifeng X Gang F Dong and H Chunpeng ldquoStudy on theemergency rescue decision support system of petrochemicalplant based on ACP theoryrdquo in Proceedings of the 6th Manage-ment Annual Meeting (MAM rsquo11) 2011 (Chinese)
[14] X Dong D Fan G Xiong F Zhu Z Zhang and Y Yao ldquoPar-allel Bus Rapid Transit (BRT) operation management systembased on ACP approachrdquo in Proceedings of the 6th ManagementAnnual Meeting (MAM rsquo11) Beijing China 2011
[15] B Ning F-Y Wang H-R Dong R-M Li D Wen and L LildquoParallel systems for Urban rail transportation based on ACP
approachrdquo Journal of Transportation Systems Engineering andInformation Technology vol 10 no 6 pp 23ndash28 2010 (Chinese)
[16] F-Y Wang J Zhao and S-X Lun ldquoArtificial power systemsfor the operation and management of complex power gridsrdquoSouthern Power System Technology vol 2 no 3 pp 1ndash11 2008(Chinese)
[17] M J North T R Howe N T Collier and J R Vos ldquoRepastsimphony development environmentrdquo in Proceedings of theAgent 2005 Conference on Generative Social Processes Modelsand Mechanisms 2005
[18] M J North T RHowe N T Collier and J R Vos ldquoRepast Sim-phony runtime systemrdquo in Proceedings of the Agent Conferenceon Generative Social Processes Models and Mechanisms 2005
[19] SWARM Development Group Swarm 211 Reference Guide2000 httpwwwswarmorg
[20] S Tisue and U Wilensky ldquoNetLogo a simple environment formodeling complexityrdquo in International Conference on ComplexSystems pp 16ndash21 2004
[21] S Luke C Cioffi-Revilla L Panait K Sullivan and G BalanldquoMASON a multiagent simulation environmentrdquo Simulationvol 81 no 7 pp 517ndash527 2005
[22] M Ichikawa H Tanuma Y Koyama andH Deguchi ldquoSOARSintroduction as a social microscope for simulations of socialinteractions and gamingrdquo in Organizing and Learning throughGaming and Simulation Proceedings of the 38th Conference ofthe International Simulation and Gaming Association (ISAGArsquo07) pp 149ndash158 2007
[23] D S Burke J M Epstein D A T Cummings et al ldquoIndividual-based computational modeling of smallpox epidemic controlstrategiesrdquo Academic Emergency Medicine vol 13 no 11 pp1142ndash1149 2006
[24] F-Y Wang ldquoParallel control and management for intelligenttransportation systems concepts architectures and applica-tionsrdquo IEEE Transactions on Intelligent Transportation Systemsvol 11 no 3 pp 630ndash638 2010
[25] Z Molnar D Balasubramanian and A Ledeczi ldquoAn introduc-tion to the genericmodeling environmentrdquo in Proceedings of theTOOLS Europe 2007 Workshop on Model-Driven DevelopmentTool Implementers Forum Zurich Switzerland 2007
[26] S Kelly and J-P Tolvanen Domain Framework in Domain-Specific Modeling Enabling Full Code Generation JohnWiley ampSons Hoboken NJ USA 2007
[27] B P Zeigler H Praehofer and T G Kim Theory of Modelingand Simulation Integrating Discrete Event and ContinuousComplex Dynamic Systems Academic Press San Diego CalifUSA 2000
[28] G Booch J Rumbaugh and I JacobsonThe Unified ModelingLanguage User Guide Addison-Wesley Reading Mass USA1999
[29] F Wagner R Schmuki T Wagne and P Wolstenholme Mod-eling Software with Finite State Machines A Practical ApproachCRC amp Taylor amp Francis Boca Raton Fla USA 2006
[30] T Espiner ldquoChina builds worldrsquos fastest supercomputerrdquo ZDNetUK October 2010
[31] B Chen and G Guo ldquoA two-tier parallel architecture for arti-ficial society simulationrdquo in Proceedings of the ACMIEEESCS26th Workshop on Principles of Advanced and DistributedSimulation (PADS rsquo12) pp 184ndash186 July 2012
[32] G Guo B Chen X G Qiu and Z Li ldquoParallel simulation oflarge-scale artificial society on CPUGPU mixed architecturerdquo
Mathematical Problems in Engineering 27
in Proceedings of the ACMIEEESCS 26th Workshop on Prin-ciples of Advanced and Distributed Simulation (PADS rsquo12) pp174ndash177 July 2012
[33] S-X Lun ldquoResearch on the classification of parallel executionmodes of ACP theoryrdquo Acta Automatica Sinica vol 38 no 10pp 1602ndash1608 2012
[34] F-YWang ldquoA framework for social signal processing and anal-ysis from social sensing networks to computational dialecticalanalyticsrdquo Science China vol 43 no 12 pp 1598ndash1611 2013
[35] G GuoOne Model Userrsquos Manual version 12 National Univer-sity of Defense University 2011
[36] Y Ge R Meng Z Cao X Qiu and K Huang ldquoVirtual citymdashanindividual-based digital environment for human mobility andinteractive behaviorrdquo Simulation Transactions of the Society forthe Modeling and Simulation International 2014
[37] M Kretzschmar and R T Mikolajczyk ldquoContact profiles ineight European countries and implications for modelling thespread of airborne infectious diseasesrdquo PLoS ONE vol 4 no6 Article ID e5931 2009
[38] W Duan and X Qiu ldquoFostering artificial societies using sociallearning and social control in parallel emergency managementsystemsrdquo Frontiers in Computer Science vol 6 no 5 pp 604ndash610 2012
[39] W J Edmunds C J OrsquoCallaghan and D J Nokes ldquoWhomixes with whom Amethod to determine the contact patternsof adults that may lead to the spread of airborne infectionsrdquoProceedings of the Royal Society B Biological Sciences vol 264no 1384 pp 949ndash957 1997
[40] J Banks J S Carson B L Neison and D M Nicol Discrete-Event System Simulation Prentice Hall Englewood Cliffs NJUSA 4th edition 2007
[41] H Yongxia ldquoInvestigation and analysis of college studentsrsquo workand restrdquo Value Engineering vol 31 no 3 2012 (Chinese)
[42] W Duan Z Cao Y Wang et al ldquoAn ACP approach to publichealth emergency management using a campus outbreak ofh1n1 influenza as a case studyrdquo IEEE Transactions on SystemsMan and Cybernetics Part A Systems and Humans vol 43 no5 pp 1028ndash1041 2013
[43] HWHethcote ldquoThemathematics of infectious diseasesrdquo SIAMReview vol 42 no 4 pp 599ndash653 2000
[44] G Chowell E Shim F Brauer and et al ldquoModelling thetransmission dynamics of acute haemorrhagic conjunctivitisapplication to the 2003 outbreak in Mexicordquo Statistics inMedicine vol 25 no 11 pp 1840ndash1857 2006
[45] Y Ge L Liu B Chen X Qiu and K Huang ldquoAgent-basedmodeling for InfluenzaH1N1 in an artificial classroomrdquo SystemsEngineering Procedia vol 2 pp 94ndash104 2011
[46] A R Tuite A L Greer MWhelan et al ldquoEstimated epidemio-logic parameters andmorbidity associatedwith pandemicH1N1influenzardquo CMAJ vol 182 no 2 pp 131ndash136 2010
[47] L Brouwers B Cakici M Camitz A Tegnell and MBoman ldquoEconomic consequences to society of pandemic H1N1influenza 2009mdashpreliminary results for Swedenrdquo Euro Surveil-lance vol 14 no 37 pp 1ndash7 2009
[48] Y Ge L Liu X Qiu H Song Y Wang and K Huang ldquoAframework of multilayer social networks for communicationbehavior with agent-based modelingrdquo Simulation Transactionsof the Society for the Modeling and Simulation International vol89 no 7 pp 810ndash828 2013
[49] W Duan X Qiu Z Cao X Zheng and K Cui ldquoHeterogeneousand stochastic agent-based models for analyzing infectious
diseasesrsquo super spreadersrdquo IEEE Intelligent Systems vol 28 no4 pp 18ndash25 2013
[50] The Preparation and Emergency Response Plans to PandemicPublished by Ministry of Health Ministry of Health 2005(Chinese)
[51] R M Fujimoto Parallel and Distributed Simulation SystemsWiley Interscience 2000
[52] F Cicirelli A Giordano and L Nigro ldquoDistributed simulationof situated multi-agent systemsrdquo in Proceedings of the 15thIEEEACM International Symposium on Distributed Simulationand Real Time Applications (DS-RT rsquo11) pp 28ndash35 September2011
[53] B Chen and G Guo ldquoA two-tier parallel architecturefor artificial society simulationrdquo in Proceedings of the 26thACMIEEESCS Workshop on Principles of Advanced and Dis-tributed Simulation (PADS rsquo12) pp 184ndash186 July 2012
[54] G Guo B Chen and X Qiu ldquoParallel simulation of large-scale artificial society with GPU as coprocessorrdquo InternationalJournal of Modeling Simulation and Scientific Computing vol4 no 2 Article ID 1350005 2013
[55] B Chen L-B Zhang X-C Liu and H Vangheluwe ldquoActivity-based simulation using DEVS increasing performance by anactivity model in parallel DEVS simulationrdquo Journal of ZhejiangUniversity Science C vol 15 no 1 pp 13ndash30 2014
[56] X-L Wang P Yang Z-D Cao et al ldquoQuantitative evaluationon the effectiveness of prevention and control measures againstpandemic influenza A (H1N1) in Beijing 2009rdquoChinese Journalof Epidemiology vol 31 no 12 pp 1374ndash1378 2010 (Chinese)
[57] S Mei A D van de Vijver L Xuan Y Zhu and P M ASloot ldquoQuantitatively evaluating interventions in the influenzaA (H1N1) epidemic on China campus grounded on individual-based simulationsrdquo Procedia Computer Science vol 1 no 1 pp1675ndash1682 2010
[58] X Qiu Z Cao and Z Li Mid-Term Reports of China NationalScience Foundation 91024030 2014
Submit your manuscripts athttpwwwhindawicom
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
MathematicsJournal of
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Mathematical Problems in Engineering
Hindawi Publishing Corporationhttpwwwhindawicom
Differential EquationsInternational Journal of
Volume 2014
Applied MathematicsJournal of
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Probability and StatisticsHindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Journal of
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Mathematical PhysicsAdvances in
Complex AnalysisJournal of
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
OptimizationJournal of
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
CombinatoricsHindawi Publishing Corporationhttpwwwhindawicom Volume 2014
International Journal of
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Operations ResearchAdvances in
Journal of
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Function Spaces
Abstract and Applied AnalysisHindawi Publishing Corporationhttpwwwhindawicom Volume 2014
International Journal of Mathematics and Mathematical Sciences
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
The Scientific World JournalHindawi Publishing Corporation httpwwwhindawicom Volume 2014
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Algebra
Discrete Dynamics in Nature and Society
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Decision SciencesAdvances in
Discrete MathematicsJournal of
Hindawi Publishing Corporationhttpwwwhindawicom
Volume 2014 Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Stochastic AnalysisInternational Journal of
8 Mathematical Problems in Engineering
Artificial society description in emergency
Description ofemergency
ArtificialSociety Editor
(ASE)
Agent behavior set
Agent role setAgent status set
Environment setAgent relationship set
Statistical data of real society
Descriptionof artificial
societyframework
Terminal of agent modeling andartificial society describing
middot middot middotAgent
metamodeling
Customized agentmodeling
Relationship modelDailyEmergency
Mod
el D
evelo
pmen
t Too
l (M
DT)
Agent modeling
Agent model Encapsulate
Encapsulate
Behavior model
Intervention
Emergency
Response
Daily
Social relationshipmodeling
Socialbehaviormodeling
Socialbehaviormodel
Emergencybehavior
model
Emergencybehaviormodeling
Building
Geospatial statistical data
Population statistical data Datacollection
Actualsociety
Artificial societypopulation and
geospatial database
Organizationservice
Transportationservice+
MappingAgent model andservice repository+ emergency model
repository
Climateservice
Geographicalservice
+
Agent status in emergencyEmergency management
Emergent event set
Social relationship statistical data
Environment statistical data
Population behaviorstatistical data
Emergency statistical data
Emergency organization statistical data
Geospatial and SocialEnvironment Generation
Tool (GSeT)
Agent model repository andpopulation database
C++C++C++C++
Modeling
Statistical modelJob ageSex location
Public service modeling
Organizationmodel
Transportationmodel
Geographicalmodel
Climatemodel
Interventionservice
Publ
ic se
rvic
e sta
ndar
d
Figure 6 The editing and initialization of KD-ACP
MDT provides domain experts with a tool to obtain thecode implementations of models With the help of compileenvironment like visual studio MDT also supports thefurther programming development of the specific domaindetails which cannot be described in modelling step
Both the FSM models and services are developed by theMDT first and then stored in the agent model and servicerepository The repository manages the models according tothe requirements from emergency problems and provides themodels for EDT to make the experiment plans
34 The Editing and Initialization of Artificial Society byASE and PGET As shown in Figure 6 ASE is used to editthe scenarios of artificial society within emergent eventsThe editing is composed of two parts (1) the statisticalinformation of artificial society in daily life such as the rolesof agents the relationships of agents and the types of environ-ments and (2) the statistical information of artificial societyin emergency including the statistical data of emergent
events the emergency organization and emergency relatedbehaviors of agents
According to the requirements of the editing thesestatistical data are collected from actual society manually bythe domain experts Based on these statistical data PGETgenerates the artificial society population and geospatialenvironment database The database supports the instanti-ation of artificial society at individual level For examplethe attributes such as age and gender of each agent can befound in the database With the support of the databaseFSM models and service repository discussed before itis sufficient for domain experts to study the emergencyproblems by computational experiments
35The Computational Experiments and Parallel Execution inKD-ACP The tools of ldquoCrdquo part and ldquoPrdquo part in KD-ACP areused to support the process of computational experimentsand parallel execution The working process is shown inFigure 7 EECT and IMCT are both the starting and ending
Mathematical Problems in Engineering 9
Agent model andservice repository
repository
Emergency EventsConfigurationTool (EECT)
Open Source DataRegistration Tool (OsdRT)
Intervention MeasuresConfigurationTool (IMCT)
Load
Output Output
Internet
Emergencyresponse plans
Load
∙ Input models of emergencyevents∙ Output settings of agentbehavior in emergent events+ settings of emergent events
∙ Input models of interventionmeasures∙ Output settings of agentbehavior in interventionmeasures + settings ofintervention measures
+ emergency model
∙ Input data from internetnetworks∙ Output population data +emergency events data +intervention measures data
Artificial societypopulation and
geospatialenvironment database
Artificial societyruntime database
OutputThe most optimal
emergencyresponse plan
Actual society
OutputOutputOutput
Input
Load
Load
Load
Load
Output
Artificial SocietySituation Tool (ASST)
Emergency decisionorganization
ClusterRuntime infrastructure
Deploy
Deploy
Experimentsdesign tool (EDT)
Artificial societyinitial data file
Experimentplan
Artificial societymodels
Experiments Management
large scale artificial societyruntime infrastructure
Tool(EMT) +super computerTIANHE-1A
Load
∙ Input experiment plan +artificial society models +artificial society initial data file∙ Output artificial societyruntime state data
∙ Input description of artificialsociety + artificial society populationand geospatial environment society+ emergent events configuration +
intervention measures configuration∙ Output experiment plan + artificialsociety models + artificial societyinitial data file
∙ Input runtime statistical data∙ Output graphics charts ofartificial society statistical data +situation of artificial society
Figure 7 The computational experiments and parallel execution of KD-ACP
point The emergent events and intervention measures areconfigured by EECT and IMCT respectively The configu-rations of emergent events are used to simulate both thereal emergencies like SARS and H1N1 and the supposedemergencies for experiments Similarly the configurationsof intervention measures are also used to reproduce thereal one and simulate the supposed one The repeat of theemergency is used to verify the models while the supposedconfigurations are used to obtain the optimized decision planto the response of the possible emergencies
With the input of artificial society model and servicerepository artificial society population and geospatial envi-ronment database and the configurations discussed beforeEDT generates the experiment plans to meet the require-ments of research on emergency management The outputof EDT includes artificial society models artificial societyinitial data files and experiment plan The models aredownloaded from the repository while the data file is thecollection of data from the database to initialize the modelsWhen the models and data files are ready EMT loads theexperiment plan and deploys the models and data to thecluster or TIANHE-1A supercomputer [30] which was theworldrsquos fastest supercomputer built by National University of
Defense Technology (NUDT) in China in 2010 According tothe plan the experiments are done repeatedly on the largescale artificial society runtime infrastructure [31 32] by themultisample settingsThework process is the implementationof computational experiments in ACP approach
Traditionally emergency response plans are made byemergency management theories and experiences The onlyway to test the effective of plans is the feedback results of realworld ACP approach provides a novel method to supportemergency response plan making by parallel execution Asshown in Figure 7 the work process of KD-ACP is composedof two loops The inner loop composed by red arrowsdescribes the process of computational experiments while theouter loop of yellow arrows illustrates the process of parallelexecution During the runtime of computational experi-ments the statistical data of artificial society is collected andstored in the artificial society runtime database Based on thedatabase ASST outputs the customized situation of runningartificial society by graphics charts and situation mapsThe information is sent to the organizations of emergencydecision to support making the emergency response plansWith the help of computational experiments loop these plansare simulated repeatedly to find the most optimal one
10 Mathematical Problems in Engineering
Open Source Data Registration Tool (OsdRT)
Emergent EventsConfiguration Tool (EECT)
Intervention MeasuresConfiguration Tool (IMCT)
∙ Web information collecting
∙ Deep mining in social media
∙ Web information denoising
∙ Information filtering
Data acquisition
Actual society
Open source informationin social media
Internet
∙ Basic element extracting∙ Individuals and organizationsextraction∙ Sentiment analyzing andopinion mining∙ Social network analyzing
Data extraction
∙ Domain knowledge description∙ Information fusion and collisiondetection∙ Domain knowledge construction∙ Social media informationstandardization
Data standardization
Knowledge
Figure 8 The compositions of OsdRT [34]
Moreover the most optimal plan is used to the responseof emergency in actual society According to the idea ofparallel control in [24] the feedback of actual society ispartly collected from Internet networks by OsdRT As shownin Figure 8 OsdRT is composed of three components dataacquisition data extraction and data standardization Dataacquisition collects mines and filters information fromsocial sensing networks Data extraction includes basicelement extracting individual and organization extractingsentiment analyzing and social networks analyzing Datastandardization specifies the useful knowledge and sent it toconfiguration tools in KD-ACP
By processing in OsdRT the knowledge about emergentevents and intervention measures are analyzed first andregistered in the EECT and IMCT The registration updatesthe settings of emergent events and intervention measuresThis loop composed of yellow arrows implements the parallelexecution in ACP approach The implementation of parallelcontrol and management provides a data-driven approachthat considers both the engineering and social complexityfor modelling analysis and decision making in emergencymanagement
4 Modelling Beijing City with KD-ACP
41 How to Build the Artificial Beijing According to themodelling of artificial society in KD-ACP discussed beforethe Beijing city is modeled as follows
To meet the requirements of emergency managementsthe basic elements of artificial society are extended Asshown in Figure 9 six elements are required to simulate thecity agents environments transportation activity schedulecommunication and agent activity
411 Modelling Artificial Beijing Figure 10 shows the mainGUI of GME for the modelling of artificial modelling inpublic health events Metamodels are listed in the left area ofFigure 10 the list provides basic syntax elements for domainexperts to model artificial society Domain experts build
models based on the knowledge of their own MeanwhileGME supports hierarchy for building large scale systemsThesyntax symbol listed in GUI can be extended in new tab bydouble clicking Take agent for example the model of agentcan be detailed by edition in another tab page of agent
As shown in the center of Figure 10 the models ofartificial Beijing consist of five parts models of agent andenvironment domain models of public health events inter-ventionmodels controllermodels and services Agentmodeldescribes individuals in society it is composed of basicpopulation information action social relationships activityschedule and disease related information Activity schedulerepresents individualrsquos physical actionmodel focusing on thedaily action of agents Environment model includes physicalentities such as buildings playground transportations andagents contained in environment Domain models of publichealth events are composed of the propagation model of dis-ease disease state transition model and so on Interventionmodels include the settings of vaccination isolation and soon The models mentioned before are all FSM models Themechanisms of these models will be detailed in next sectionsController models and services are the public service mod-ules they are implemented in the development in MDT
412 Modelling Agents and Environments Under the speci-fication of FSM agent and environment models are imple-mented in two parts the state space and state transitionsThe state space is composed of the demographic attributesand behavior related attributes The transitions are triggeredwhen the conditions of states are satisfied As shown inFigure 11(a) the action of agent is changed when the ldquonexttimerdquo condition is satisfied in agent model while the agentslist is changed when the agent arrival condition is satisfied inthe environment model
413 Modelling Activities Agent activities come from theagent state transitions of actions such as movements andcommunications The actions of agents are instructed by theactivity schedule shown in Table 2 Activity schedule lists
Mathematical Problems in Engineering 11
Transportation EnvironmentsEnvironments
Agents
Artificial Beijing
Agent activityActivity schedule
Agent activity
Communication by socialrelationship networks
Communication by socialrelationship networks
Figure 9 The basic elements of artificial society
Syntax elements of DSM
Services
Models of agentenvironment andactivity schedule
Domain models of publichealth events
Intervention models
Modelingaspect ofdomainexperts
Controllermodels
Figure 10 The models of artificial Beijing in GME
all the actions with probability in one day for agents inboth normal and emergent situation [38] There are severaltypes of activity schedule in artificial Beijing student agentactivity schedule worker agent activity schedule emergentagent schedule and so on For example Table 2 gives an agentactivity schedule Upon the instruction of activity schedulestudent agent changes the actions by 119901
119894after state transitions
The119901119894in the tablemeans the action probability in the relevant
period In the duration from 0800 to 1200 a student agenteither goes to classroom to have class or goes to libraryto study The probability of class action is 119901
2while the
probability of study action is 1 minus 1199012
Agent behaviors are decided by the settings of activityschedule In addition to the daily activity schedulementionedbefore emergent activity schedules are also considered inour work Take public health events for instance an infectedagent changes schedule from a normal one to emergentone The workflow of a susceptible agent is illustratedin Figure 12 to show the change of behaviors After theinfection the agent is set in incubation phase Not all theincubation agents will become symptomatic Some of themturn back to being susceptible and some of them becomesymptomatic The symptomatic agents change their activityschedule from normal to emergent In the emergent case
12 Mathematical Problems in Engineering
Next actionNext environment (location)
Advance to next actiontime
Find next environment inactivity chain by action
Current actionCurrent environment (location)
Q
q0
120575 Find next action inactivity schedule by time
F
Σ
(a) The state transitions of agent
Q
q0
120575 F
Current agent list
Current pollution level
Next agents list
Next pollution level
Agents entered
Infected agents entered
Add agents to currentagents list
Compute pollution levelby infected agents
Σ
(b) The state transitions of environment
Figure 11 The state transitions of models
Table 2 The activity schedule of a student agent
Duration (Δ119905) Activity Location Probability 119879Location (minute)0000ndash0600 Sleep Dormitory 119875
0(100) 360
0600ndash0800 Breakfast Dormitoryrestaurant 1198751(068) 120
Sports-breakfast-travel Playground-dormitoryrestaurant 1 minus 1198751(032)
0800ndash1200 Class Classroom 1198752(077) 240
Study Library 1 minus 1198752(023)
1200ndash1400 Lunch Restaurant 1198753(090) 120
Lunch-shopping Restaurant-convenience store 1 minus 1198753(010)
1400ndash1800 Class Classroom 1198754(070) 240
Study Library 1 minus 1198754(030)
1800-1900Dinner Dormitoryrestaurant 119875
5(063)
60Sports-shopping Playgroundconvenience store 1198756(025)
Study Classroomlibrary 1 minus (1198755+ 1198756) (012)
1900ndash2200Rest Dormitory 119875
7(045)
180Dinner-rest Dormitoryrestaurant-dormitory 1198758(020)
Study Classroomlibrary 1 minus (1198757+ 1198758) (035)
2200ndash2400 Rest Dormitory 1198759(072) 120
Sleep Dormitory 1 minus 1198759(018)
infected agents go to hospital according to the treatmentschedule or stay in dormitory according to isolation scheduleAfter the treatment in hospital or self-immunoprocess agentsbecome healthy and immune of disease If the agents aretreated in the hospital they are not allowed to get out untilrecovered Moreover activity schedules are also influencedby the emergency response plans For example in the case ofisolation in emergency response plans the activity scheduleof agents who had contacted with infected is changed Onlythe locations including home and dormitory compose thisemergent schedule Likewise based on the statistical datafrom emergency response plans the additional possibilities
are added into activity schedulesThe behaviors of agents alsochange with activity schedules In the view of ACP approachthe injection of new emergency response plans implementsthe parallel execution of microagent behaviors
In another aspect contact frequency is also anothercrucial element to determine the infected rate in public healthevent It is important to model the contact frequency andcontact time of individuals Based on the studies on thecontact behavior of human being by questionnaire surveyEdmunds found that the contact frequency of individualcould be fitted approximately into a normal distributionand the mean and the standard deviation distributions are
Mathematical Problems in Engineering 13
Susceptible agent ImmuneYes
Healthy and immune
No
Choose schedule
Self-isolation Go to hospital
Staysymptomatic
No
Receive treatment
Contact withinfected agents
Symptomatic
No
Self-immunoprocess
Yes
Recovered
Yes
Incubation
Infection
No
Yes No
NoYes
Figure 12 The workflow of a susceptible agent
168 and 85 [39] So it is possible to apply normal randomvariable to model the contact frequency In our work Box-Muller method [40] is used to generate the random variableof contact frequency shown in the following
119865119862= 120583119865+ 120590119865(minus2 ln (120574
1))12 cos (2120587120574
2) (1)
in which 119865119862is the random variable of contact frequency 120583
119865
is the mean value of normal distribution 120590119865is the standard
deviation of normal distribution and 1205741and 120574
2are the
uniform random variables distributed in the interval [0 1]Based on (1) and survey data [41] the contact frequenciesof agent 119865contact(119860 119894) are discretized as in Table 3 within theconsideration of activity differences
Similarly the duration per contact between individualsis another key factor of AHC transmission The duration ofcontact could also bemodeled by the normal randomvariablein the following [42]
119879119862= 120583119879+ 120590119879(minus2 ln (120574
1))12 cos (2120587120574
2) (2)
in which 119879119862is the random variable of duration per contact
120583119879
is the mean value of normal distribution 120590119879
is thestandard deviation of normal distribution and 120574
1and 120574
2
are the uniform random variables distributed in the interval
[0 1] In addition because the time spent in specific location119879Location(119860 119894) also follows the random distribution like (3)in our work In order to make the duration of contact119879contact(119860 119894) shorter than the 119879Location(119860 119894) the mean andstandard deviation of (2) is set up in (3) and (4) [42]
120583119879=119879Location(120583119865+ 120590119865) (3)
120590119879=119879Location(10120583119865+ 10120590
119865) (4)
in which 119879Location is the time of an activity in a specificlocation set in Table 2 120583
119879and 120590
119879are the mean and standard
deviation of contact frequency As a result the duration ofagent 119879contact(119860 119894) in specific locations (listed in Table 2) isalso discretized in Table 3 within the consideration of activitydifferences
414 Modelling Transportation In artificial Beijing threekinds of travel models are considered walking inside thedistrict travelling by road networks and travelling by subwaynetworks According to the traveling models the agentmovements are implemented by the compositions of travelingmodels before So it is important to build the basic road
14 Mathematical Problems in Engineering
Table 3 The contact frequency and duration in different locations and activities by social relationships in case of student and teacher
Social relationship Activity Location 119879contact(119860 119894) (minute) 119865contact(119860 119894)
Student-student
Sleep Dormitory 0 0
Breakfast Restaurant 119873(52) 119873(32)Breakfast Dormitory 119873(52) 119873(21)Sports Playground 119873(21) 119873(53)Class Classroom 119873(21) 119873(21)Study Library 119873(21) 119873(21)Lunch Restaurant 119873(158) 119873(32)
Shopping Convenience Store 119873(203) 119873(21)Dinner Restaurant 119873(108) 119873(21)Dinner Dormitory 119873(158) 119873(32)Rest Dormitory 119873(63) 119873(52)
Teacher-student Class Classroom 119873(21) 119873(21)
Subway network of Beijing city (red lines) Road network of Beijing city (green lines)
Figure 13 The implementation of transportation for artificial Beijing
and subway networks With the help of Google map thekey points of road are sampled first and the road networksare generated by the links among point Similarly subwaynetworks are generated by the links between subway stationsmapped from the real stations in Google map Figure 13shows the road and subway networks in artificial Beijingwith the road networks in green and the subway networks inred Most of the districts in Beijing city are covered by thesetwo networks The algorithm of hierarchical route planningis proposed as follows
(i) Find the next position where agent will be located innext action
(ii) Obtain the travelingmodels from the statistical trans-portation data
(iii) Search the path from the starting position to thenearest road entry or subway stations
(iv) Search the optimal path from the starting station orroad entry to the nearest station or road exit of targetposition
(v) Search the path from the target subway station or roadexit to the target position
(vi) Attach the subway train number or the bus numberincluding the transfer information to the path accord-ing to the timetable of subway or bus
(vii) Connect all the paths obtained before Generate thepath from the starting position to the target position
Based on the algorithm for each agent in artificial Beijingthe transportation is simulated during the computationalexperiments It is worth noting that the activities insidethe buses or train cars are considered because the agentcommunication during travelling is also necessary to besimulated
Mathematical Problems in Engineering 15
(a) The implementation of agent in MDT (b) The code framework of agent
Figure 14 The agent model implementation in MDT [35]
415 The Implementation of Models by MDT All the modelsare developed in the MDT Figure 14 shows the GUI of thetool MDT is used to implement the state space and statetransitions of FSM models The code framework is shownin Figure 14(b) Using the C++ inheritance features and thetechniques of dynamic link library (dll) the models areencapsulated in the dll componentsThe flexible compositionmechanism is designed to support the models evolution incomputational experiments
416 The Description of Artificial Beijing by ASE As men-tioned in Section 34 ASE describes the artificial Beijingin macroview The statistical information of Beijing city iscollected by ASE Table 4 lists part of the description ofartificial Beijing generated by ASE
42 How to Obtain the Data of Beijing City
421 Population and Geospatial Environment Database ofBeijing City Based on the models developed for ArtificialBeijing it is necessary to generate the initial data to do thecomputational experiments Because the individual level dataare not available it is necessary to construct an individual-based population database for both accurate computationalexperiment and determining optimal decisions
According to the state spaces of models discussed inSection 412 the geospatial and population database of Arti-ficial Beijing is designed as shown in Figure 15 The kernelpart of database consists of the tables such as agent list tablegeospatial environment list table household list table andagent distribution table Agent list table is an individual leveltable used to store the data to initialize the state space of agentmodel such as id and gender age Geospatial environmentlist table is also an individual level table which stores thedata of environment attributes such as id street id andtype Household list table and agent distribution table storethe statistical data for the data generation for agent andenvironment models With the help of the geospatial andpopulation database the data generating of artificial Beijingis proposed in the next section
422 Generating Geospatial and Social Environment Datafor Artificial Beijing by PGET The algorithm of generatinggeospatial and social environment data for artificial Beijingis used to quantify the spatial distribution of population andformalize geospatial behavior of each agent The algorithmshown in Figure 16 is capable of generating a syntheticartificial society which allowsmultiresolution statistical datasocial interactive behavior and multilayer social networksto be integrated together The synthetic population canrepresent individual agents in the form of households andhousehold members and the synthetic population is statis-tically equivalent to a real population For each householdcharacteristics such as address household size family typesand relationships are generated Each person is described bycharacteristics such as age gender social role and correlatedlocations The algorithm provides an effective methodologyto reconstruct the computing environment in high resolutionby using statistical data in low resolution leading to betterprediction and management of emergencies The details ofalgorithm are illustrated in [36]
As shown in Figure 17 the implementation of the algo-rithm is embedded inside the GEPTThe collective input datasuch as the data files of population distribution is collected inthe tool first then the algorithm is activated to generate thedatabase The generation lasts almost twelve hours one timewith the specified parameters settings When the generationis finished the data of artificial Beijing is obtained in thedatabase with 19610000 agents and 16000 environments
5 H1N1 Experiments in ArtificialBeijing with KD-ACP
With the support of the artificial Beijing modeled before thecomputational experiments and parallel execution are alsoimplemented by KD-ACP The experiments are designed tofind the most optimal emergency decision response plansEpidemic in city is a typical scenario in emergency manage-ment H1N1 epidemic in Beijing in 2009 is used to be the usedcase to test KD-ACP According to Section 34 it is necessarytomodel emergencymanagement first in order to support thecomputational experiments
16 Mathematical Problems in Engineering
Table 4 The description of artificial Beijing generated by ASE
Statistical features ofartificial Beijing Descriptions of statistical features Data sourcedefault values
Agent roles Baby child primary student middle school student college student worker retired
Demographic data file ofBeijing
Environment types Super market store restaurant park stadium hospital school cliniccommercial building dormitory building residential building
Demographic data file ofBeijing
Demographic data Age distribution of population sex distribution of population householddistribution of population retired age distribution children distribution
Demographic data file ofBeijing
Environment data Distribution of all types of environments Environment distributiondata file of Beijing
Relationships Classmate relative college friendship family Demographic data file ofBeijing
Varying parameters
Age difference of couples119873(120583120590) 119873(03) isin [0 12]Minimummarriage age (male female) (22 20)Age difference of children119873(120583120590) 119873(13) isin [1 10]Enrollment rate of kindergarten 120572 08Enrollment rate of primary school 120573 1Enrollment rate of middle school 120574 1Enrollment rate of college 120575 08Entrance age of kindergarten 1205721015840 3Entrance age of primary school 1205731015840 7Entrance age of middle school 1205741015840 13Entrance age of middle school 1205751015840 19
Age distributioninformation table
Agent settable
Geospatialenvironment set table
Household settable
Figure 15 The geospatial and population database of Beijing city
51 Modelling Emergency Management The modelling ofemergency management consists of two parts the modellingof emergent events and the modelling of intervention mea-sures H1N1 and the intervention measures for H1N1 aremodeled in our case
511 Modelling H1N1 by MDT H1N1 Model in agent simu-lates the states transitions of health status and the relevantactions Referring to SIR (SusceptibleInfectiveRecovered)[43 44] models agent has three health statuses susceptibleinfected and healthy with immunity Only susceptible agentcan be infected by the contact with infected agents After
the three infected phases (incubation being symptomaticand recovery) of H1N1 model agent is healthy again It isworthy mentioning that the recovered agents can be infectediteratively when the immunity disappears gradually by timeThe state transitions are illustrated in Figure 18
According to [45] the latent period of H1N1 was in aWeibull distribution within people from one to seven days[46 47] The distribution was usually centered in the rangebetween one and three days AWeibull random variable [40]is used to model the latent period as described in
119879Lat = 120572 times [minus ln (1 minus 120574)]1120573+ 120592 (119879Lat ge 120592) (5)
Mathematical Problems in Engineering 17
Censusdata
Generate GUID
Assign individualhousehold
Generate familystructure
Assign family roleto every member
Endow individualwith age gender
social role
Distribute environmentlocations on the map
School Workplace Entertainmentlocations
Consumptionlocations
Assign school orworkplace to
individual
Correlatedindividual withother locations
Generate friendshipin school
Generate coworkerrelations inworkplace
Assign individualadministrative
region
Generate housebuildings
Locate householdto house building
Generate familyrelation
Generateneighborhood
relation
PopulationEnvironment
Environment listSocial network
Figure 16 The algorithm of generating geospatial and social environment data [36]
Edit
Edit
Edit
Edit
Edit
Edit
BrowseStatistical characteristic of population
Geospatial population data
Population data of street
Population types
Geospatial and social environment generation tool
Workspaces
Parameters distribution settings
Household types
Roles in household
Household structure
Social roles
BrowseBrowseBrowseBrowseBrowseBrowse
Population distribution data
File path
Retired distributionPop distributionHousehold numHousehold distributionSex distributionAge distributionData type
Figure 17 The GUI of Population and Geospatial Environment generation Tool (PGET)
18 Mathematical Problems in Engineering
Infected
SusceptibleHealth
withimmunity
Infected
by contac
t
with in
fected
agents
Recovery
Increase before beingsymptomatic symptomatic
Decrease after being
Decrease by time
Infectivity
Immunity
Figure 18 The transitions of agent health status
119879Lat denotes the duration of latent period 120592 120572 and 120573 arethe location parameter scale parameter and shape parameterof Weibull distribution respectively 120574 is a uniform randomnumber in the range [0 1] According to statistics [47] 120592 120572and 120573 are set as 0 18 and 121 respectively Then the mean(standard deviation) of latent period is calculated as 159 days(058 day2) Furthermore infectious period is set as 7 days
A susceptible studentrsquos probability of getting infecteddepended on the infectivity of infectious agents his ownimmunity level the duration of the contact action and soon The infectivity of an infected agent evolved with timeduring the infectious period According to the statistics ofinfectivity in the chart with red bars in Figure 19 an infectedstudent had the highest level of infectivity in the second dayafter he had the first symptom The infectivity levels in otherdays in comparison with the second day are in Table 5 Theday labeled as ldquominus1rdquo means the day before the starting of thesymptomatic period
Based on the algorithm discussed before theH1N1modelis developed with the help of MDT The settings of influenzamodel will be discussed in Section 521
512 Modelling Intervention Measures by MDT In corre-spondence intervention measures model is used to ceasethe outbreak of H1N1 As discussed in [42 48 49] if theappropriate measures are taken when the infectious diseaseemerges the transmission of the disease could be sloweddown and the damage could be decreased Therefore it isnecessary and important to design emergency interventionmeasures in the artificial Beijing According to the MinistryofHealth inChina the intervention policies [50] are designedto control the spread of influenza such as H1N1 The inter-ventionmeasures of these policies including are interventionsactivated time vaccination rate antibiotic delay to hospitalisolation duration close workspace duration and limitationof activities listed in Table 6 Similarly the models of theseintervention measures are also developed with the help ofMDT Using the reasonable data ranges listed in Table 6intervention models can be initialized to do the computa-tional experiments These experiments are used to show how
Symptomatic1 2 3 4 5 6 7minus1
Recovered
Tim
e (da
y)
Infe
ctiv
ity
Figure 19 The distribution of infectivity rate of an infected agent[37]
Table 5 The proportion of infectivity
Days (119894) minus1 1 2 3 4 5 6 6Proportion of infectivity 03 05 1 08 04 02 01 005
to restrain the H1N1 transmissionThe settings of these inter-vention measures in IMCT are introduced in Section 522
52 Experiment Settings
521 H1N1 Settings by EECT EECT is designed to initializethe emergent events models such as H1N1The settings comefrom the specified emergent event models As shown inFigure 20 in theH1N1 used case the settings are composed ofthe distribution of infection source infection rate infectionperiod and so on The infection rate changes with envi-ronments and infection periods The settings are initializedfrom both the research of medicine on H1N1 and the data ofH1N1 outbreak in Beijing in 2009The process of influenza inartificial Beijing is simulated to repeat the possible scenariosof H1N1 break During the computational experiments ofartificial Beijing theH1N1 settings such as the period of beingsymptomatic are updated by the dynamic data registrationsfrom OsdRT The updating corrects the H1N1 model duringthe experiments With the help of OsdRT artificial Beijing isable to approach actual society It is also an implementationof parallel execution
522 Intervention Measures Settings by IMCT In the H1N1case the settings of intervention measures are shown inFigure 21 in IMCT These measures are designed under theinstruction of Emergency Decision Organizations like ChinaDisease Center (CDC) The measures such as vaccinationrate the rate of closedworkspace and the starting time can beinitialized in the tool Emergency response plans are imple-mented in these settings As illustrated in Figure 7 all the pos-sible compositions within value settings in reasonable inter-vals are tested to find the most optimal emergency responseplan For example different compositions of measures suchas vaccination rates and rates of workspace closed are usedto design the plans of experiments By the analysis of theresults the best composition of interventionmeasures can beobtainedThe analysis will be discussed in case study in detail
Mathematical Problems in Engineering 19
Table 6 The intervention measures of influenza
Intervention measures ValuesInterventions activated Time 119879Intervention isin [0 80th day]Vaccination rate 119877Vaccination isin [0 1]Antibiotic rate 119877Antibiotic isin [0 1]Delay to hospital 119867ospital(119860 119894) 119879infected lt 119905 lt 119879recovered + 119879HDelay 119879HDelay isin (0 3th day]Isolation duration 119868solation(119860 119894) 119879Intervention lt 119905 lt 119879Intervention + 119879IDuration 119879IDuration isin (0 14th day]Close workspace duration 119862lose(119864119894) 119879Intervention lt 119905 lt 119879Intervention + 119879CWDuration 119864119894 isinWorkspace 119879CWDuration isin (0 7th day]Limitation of agent activity duration 119871 imitation(119860 119894) 119879Intervention lt 119905 lt 119879Intervention + 119879LAADuration 119879LAADuration isin (0 7th day]119877Vaccination means the vaccination rate of artificial Beijing while 119877Antibiotic means the antibiotic rate used for infected agents in hospital119867ospital (119860119894) means that the infected agent 119860119894 is sent to hospital he or she will receive the retreatment immediately119868solation (119860119894) means that the infected agent 119860119894 is isolated at home he or she can only contact with the other agents until isolation duration is over119862lose (119864119894) means that the environment 119864119894 typed of workspace is closed until the close workspace duration is over119871 imitation (119860119894) means that the activities of agent 119860119894 are limited only the workspace and home are allowed until limitation duration is over
Incubation Weibulldistribution
Symbolic LogNormaldistribution
Recovering LogNormaldistribution
H1N1influenza
Settings of H1N1 influenza model
Disease model
Transmissionmodel
Figure 20 The settings of H1N1 in EECT
It is worth notifying that the settings of interventionmeasures are also influenced by OsdRT For example theH1N1 outbreak areas are notified by OsdRTThe interventionmeasures such as close workspace are used only in theoutbreak areas Therefore the unnecessary computation ofintervention measures is avoided So the data injection byOsdRT not only makes artificial society approach actualsociety but also increases the performance of computationalexperiments
53 Experiments by EMT When the initialization of arti-ficial Beijing is ready EMT is used to do the computa-tional experiments EMT is composed of two parts thecentral controller in the central node and the residen-tial service deployed in the computational nodes Centralcontroller loads the experiment plans and controls the whole
experiments process the GUI of controller is shown inFigure 22 EMT manages the computational experimentsin four steps cluster configuration load experiment plansmodels deployment and running experiments
531 Cluster Configuration Before the configuration clusterinformation is collected from the residential services insidethe cluster The information includes nodes amount nodenames node IPs and CPU occupations It is used to quantifythe computing power of the clusterThe cluster configurationis based on this information The nodes are selected forexperiments Then the LP (local process) [51] number foreach node is set according to the number of the CPU kernelsAfter the configuration the experiment plans are loaded tocustomize the models and data
20 Mathematical Problems in Engineering
Streets of close workspaces
Limitations ofactivities
IsolationsRespirator
Closeworkspace
Antibiotic
VaccinationHospital
Surveillance
Figure 21 The settings of intervention measures in IMCT
Information ofexperiment plan
List of experiment plans
Start | Pause | Reset | Stop | Speedup | Slowdown | Back
NodeName | CPU occupation | Memory occupation |Network | Models | Status
Figure 22 The deployment and running of experiments in EMT
532 Load Experiment Plan As discussed in Section 35 thedescription of experiment plans is composed of the infor-mation in three aspects models and the initial data for theexperiments the deployment mapping tables from the mod-els to the nodes and the plan of experiment executionWhenthe experiment plan is loaded by the EMT the file namesand model descriptions of models and data are listed Withthe support of cluster configuration models can be mappedto the nodes in the GUI according to the computationrequirements EMT also provides the setting of experimentsitself including the running times start time and end time
533 Models and Data Deployment Models are stored inthe agent population and services repository According tothe settings in experiment plan the customized models aredownloaded from the repository first then the models suchas agent model environment model emergent events modeland intervention measures model are integrated to be thecompositionmodels of artificial BeijingThemodels and datalocations in the nodes are configured in the ldquomodel pathrdquoand ldquodata pathrdquo in the GUI of central controller Finallythe models are uploaded to the nodes in cluster with theconsideration of partition The partition of artificial society
Mathematical Problems in Engineering 21
models is different from the multiagent system in [52]Models in artificial society are divided into three levels agentsand environments grids and city Agents and environmentsare encapsulated in grids which compose the whole cityThus the partition of artificial society becomes the partitionof grids The parallelism degrees are found to support thepartition of models agent independence and grid indepen-dence According to the parallelisms agents are partitioned toLPs according to the population distribution in grids Agentmovements among grids are simulated by the movementevents among grids in different LPs As a result a two-tierparallel architecture is proposed to support the simulation ofartificial society The architecture is detailed in [53]
534 Running of Experiments When the preparation isready the experiments can be done by the central controllerof EMT The running and control commands of the experi-ments are grouped into three aspects start pause or stop theexperiments get the running information of the nodes suchas the snapshot of the desktop of node With the help of thesecommands users can run the computational experimentsto do the research on the emergency management It isworth notifying that the simulation engine is optimized inorder to support the ten million agent simulation UsingGPU as coprocessor a two-tier parallel simulation engine isdesigned with support of MPI and OpenCL through phasedsynchronization [54] One-sided communication is used forreflection of remote simulation objects and message passingbetween processes A general kernel function prototype iselaborately designed and conditionally compiled for execu-tion on both CPU and GPU Moreover the optimizationoperations like load balance is developed under the instruc-tion of activity-based simulation [55] The densely populatedgrids are given more CPU and memory resource accordingto the activity predictions
54 Results of H1N1 Experiments with KD-ACP On thebasis of artificial society models and the intervention mea-sures mentioned in Section 512 a series of computationalexperiments are performed to study the H1N1 influenza inBeijing In our case of artificial Beijing 19610000 agentsare simulated in the cluster within 48 CPU cores and 128Gmemories the nodes are connected by kilomega networks Ittakes 18 hours to simulate a 250-day disease spreading
In order to illustrate how the optimal plans are obtainedbyKD-ACP the experiments are divided into four groups theexperiments ofmodel validation the experiments of sensitiv-ity analysis of vaccination rate the experiments of sensitivityanalysis of isolation and the experiments of the combinationof interventionmeasuresThese four groups show the generalprocess of the research on emergency management in H1N1influenza Firstly the models of H1N1 are validated withthe support of historical data Secondly the traditionalmedical interventions of pandemic such as vaccination areanalyzed to find the optimal plan Thirdly the traditionalnonmedical interventions of pandemic such as isolation areanalyzed to find the optimal plan Finally the combination ofinterventions is tested to show the combined effect
It is worth mentioning that all the experiments in ourwork are performed 100 times These experiments are ini-tialized with different settings of interventions in IMCT Theresults shown in the figures are the mean values
541 The Experiments of Model Validation In our artifi-cial Beijing the models of H1N1 influenza are built basedon the existing researches Wang and her colleagues hadstudied the H1N1 influenza by SEIR (SusceptibleExposedInfectiveRecovered) models [56] With the support of theirwork we build the models of artificial Beijing in the mannerof multiagents
In Figure 23 the control group shown in red line isdrawn based on the realistic statistical data collected in H1N1influenza in Beijing in 2009 The influenza lasts more thansix months and more than 170000 people are infectedComparing with the control group the simulated resultswithout interventions are drawn in blue It is obvious that thesimulated case fits the control group well before 70 days Butthe ascending of control group is slowed after 70 days andthe peak value (17883) is much less than the simulated results(25135) It is because the vaccination is activated at day 67 incontrol group according to the intervention measures usedin 2009 The restrain by vaccination becomes effective eightdays later The change of infected number after vaccinationis shown by red spotted line The trend of control group isdifferent from the simulation data after 75 days
However the simulated results (blue line) in compu-tational experiments fit the control group (as shown inFigure 23) in the first 70 days when no intervention mea-sures are executed The key features such as effective basicreproduction number (119877
0) are also almost the same in both
cases Therefore it can be concluded that the models ofartificial society built in microview are validated by therealistic statistical data in influenza in Beijing in 2009
In this group of experiments no interventions are exe-cuted in order to validate the models Therefore the simu-lated results of this extreme case which cannot happen todaycould be obtained As shown by blue line the number ofinfected agents without interventions reaches peak value atday 91 with 25135 consisting of 8052 in symptomatic phaseand 28950 in recovery phase It is obvious that the totalnumber of infected agents grows slowly at the beginningtime Then the number increases quickly when the influenzabreaks and it reaches its maximum value with 383870 in theend This phenomenon can be explained as in the followingAt the beginning of influenza due to the limited numberof infected agents the propagation of the disease remainsslow but many agents become susceptible When the numberof infected agents increases rapidly the spreading becomessignificant However the number of infected agents decreasesafter day 91 It is because the disease spreading relies on thesocial networks and spatial contact networks [49] It meansthat the infected agents can only infect agents either in thesocial relationship networks or in the spatially contacted
542 The Experiments of Sensitivity Analysis of VaccinationRate According to Section 512 vaccination is a typical
22 Mathematical Problems in Engineering
0
Infe
cted
agen
ts
0 50 100 150 200 250
Statistical dataSimulation data
25000
20000
15000
10000
5000
Day d
Figure 23 The results with no intervention executed and the com-parison with realistic statistical data in Beijing in 2009
medical intervention in influenza Traditionally the effective-ness of vaccination can only be evaluated after the influenzaand the key settings such as vaccination rate are determinedby the experience It is always difficult for emergency man-agement organizations to find the most optimal vaccinationrate in advanceTherefore the sensitivity of vaccination rate isanalyzed in this group of experiments The settings are listedin Table 7 average results are shown in Figure 24
Vaccination rates are set linearly (10 30 50 and70) to test the changes of influenza The peak value ofinfected agents decreases from 25135 22392 15787 6907to 638 respectively It shows that the more the agents arevaccinated the smaller the peak value of the infected agentswill be In the case of 70 vaccination the total infected agentis only 13799 It means that the influenza in this case does nothappen due to the high vaccination rate
The change of peak values also shows an interestingphenomenon Vaccination rates increase linearly the peakvalues decrease slower than the ascending of vaccination ratesat beginning but the descending becomes more and morefaster with the linear increase of vaccination rate It meansthat the small scale of vaccination (less than 30) does notmake senseWith the increase of vaccination rate the numberof susceptible agents decreases significantly In sequencethe possibility of infections is lowered greatly Additionallyit could be found that the decrease of peak values comeslater than the increase of the vaccination rates This is quitereasonable according to Mei et alrsquos work in [57]
As a result it is concluded that high vaccination rate iseffective in the control of H1N1 influenza though costly
543 The Experiments of Sensitivity Analysis of IsolationAccording to Section 512 isolation is a typical nonmedicalintervention It is the most used intervention measure in
Table 7 The initial settings of experiments of vaccination
Parameters ValueAgent count 19610000Initial infected count 40Interventions activated time(119879Intervention)
0
Vaccination rate (119877Vaccination) 0 10 30 50 70
0
Infe
cted
agen
ts
Loading intervention of vaccination
No vaccination10 vaccination30 vaccination
50 vaccination70 vaccination
0 50 100 150 200 250
25000
20000
15000
10000
5000
Day d
Figure 24 The results of loading intervention of vaccination
the emergency management in influenza As mentionedbefore a problem also exists that the measure cannot bevalidated before the intervention Therefore the sensitivityof isolation is analyzed in this group of experiments It isworth mentioning that the isolation is usually combinedwith the hospital policy It means the infected individual issent to hospital meanwhile the people who contact withthe infected are all isolated As a result parameters delay tohospital and isolation duration are designed together in thesettings of experiment Delay to hospital stands for the periodfrom infection time to hospital time while Isolation durationrepresents the period when an agent cannot contact others ifhe has contact with an infected agent The settings of theseparameters are listed in Table 8 average results are shown inFigure 25
Firstly the parameter delay to hospital is simulated aloneto test the sensitivity The simulated results are shown insolid line in Figure 25 The peak value decreases from 6708to 2218 when the delay to hospital decreases from 15 daysto 1 day The half day change brings the 67 reductionMoreover the peak value almost approximates to zero inthe case of 05 days The influenza does not happen in thiscase It can be concluded that the earlier the infected agents
Mathematical Problems in Engineering 23
Table 8 The initial settings of experiments of isolation
Parameters ValueAgent count 19610000Initial infected count 40Interventions activated time (119879Intervention) 80Delay to hospital (119879HDelay) isolation duration (119879IDuration) (15 0) (1 0) (05 0) (15 3) (1 3) (05 3)
0 50 100 150 200 2500
1000
2000
3000
4000
5000
6000
7000
Infe
cted
agen
ts
Loading intervention of isolation
Delay to hospital = 15
Delay to hospital = 10
Delay to hospital = 05
Delay to hospital = 15 isolation duration = 3
Delay to hospital = 10 isolation duration = 3
Delay to hospital = 05 isolation duration = 3
Day d
Figure 25 The results of loading intervention of isolation
are sent to hospital the less the agents will be infectedSo the construction of effective mechanism to find infectedindividuals is important in the emergency management ofinfluenza
Secondly the parameter isolation duration is added tosimulate the isolation of interventionObviously the isolationmeasure decreases the infected agents greatly Compared tothe parameter delay to hospital of 15-day case the peak valuedecreases from 6708 to 362when the isolation is added in theinterventionThe reduction of 6346 agents comes from the 3-day isolation of agents who had contacted with the infectedagents However the decrease of parameter delay to hospitalcannot bring the similar reduction in the 3-day isolationcases According to the figure the peak value decreases from362 to 51 from 15-day case to 05-day case only 311 agents aresaved from infection As a result the sensitivity of simulatedresults does not change obviously with delay to hospital inthe case of isolation It can be concluded that isolation is amore effective measure Though it is not easy for emergencymanagement organizations to decrease the delay to hospitaldue to the current monitoring mechanism the influenza canstill be controlled by isolation
544 The Experiments of Combination of Intervention Mea-sures Isolation is not the only nonmedical intervention ininfluenza close workspace is another common measureWhen close workspace is activated the workspaces are allclosed agents can only stay home at the closed time Luckilyisolation and close workspace are independent from oneanother It is possible for emergency management organi-zations to use these two measures together So this groupof experiments is designed to obtain the effectiveness of thecombination of intervention measures The parameters areinitialized with different settings listed in Table 9
Parameters delay to hospital and isolation duration is setin the middle values (1 day 3 days) according to the lastsection According to the experience close duration is set asseven days It is known that close workspace is a really costintervention measure the activated time is a key parameterWith the fixed parameters mentioned before the changeof interventions activated time is analyzed in this group ofcomputational experimentsThe average results are shown inFigure 26
Compared with the simulated results of experimentsdiscussed above Figure 26 strongly illustrates that influenzais ceased when the intervention measures are combinedtogether The peak value of infected agents is 141 in the ldquo80thdays of close-workspace and isolation with 1 day delay tohospitalrdquo case while 25135 in ldquono interventionrdquo case 2218 inldquo1 day delay to hospitalrdquo case 203 in ldquo1 day delay to hospitaland isolationrdquo case The interventions bring a great reductionof infection the peak values decrease from ten thousandslevel to hundred level
On the other hand the peak value even decreases to 108when the execution time of close workspace is 10 days earlierThe date of maximum value of infected agents is advancedfrom the 135th day to the 108th day The whole circle of H1N1influenza is reducedThe results of this group of experimentsshow that intervention measures should be activated as soonas possible and the compositions of interventions are moreeffective in emergency management of H1N1 influenza
In summary based on the general process lots of exper-iments could be done by KD-ACP to study emergency man-agement in H1N1 influenza Many conclusions are obtainedbased on the analysis of the experiments in this sectionAccording to these conclusions themost optimal response inour case study is the case of ldquo70th days of close-workspace andisolation with 1 day delay to hospitalrdquo It is because this casehas the fewest infected agents relatively But the modelling ofH1N1 influenza is not sufficient Economic and social factorsof interventions such as the money cost by vaccination andisolation the influence of the close-workspace in society
24 Mathematical Problems in Engineering
Table 9 The initial settings of experiments of isolation and closeworkspace
Parameters ValueAgent count 19610000Initial infected count 40Interventions activated time (119879Intervention) 70th day 80th dayDelay to hospital (119879HDelay) 1 dayIsolation duration (119879IDuration) 3 daysClose workspace duration (119879CWDuration) 7 days
50 100 150 200 2500
50
100
150
200
Infe
cted
agen
ts
Loading intervention of isolation and close workspace
Delay to hospital = 1 isolation duration = 3Start close time = 70 close duration time = 7Delay to hospital = 1 isolation duration = 3Start close time = 80 close duration time = 7
Day d
Figure 26The results of loading interventions of isolation and closeworkspace
and even opinion transmission of H1N1 in internet are notconsidered in our case study Without the analysis of thesefactors it does not make sense to talk about the most optimalintervention measures However KD-ACP had provided asoftware framework and general process of the research onemergency management Supported by KD-ACP the expertsof emergency management are able to study emergent eventssuch as H1N1 influenza as detailed as possible Based on theexperiments and analysis again and again the most optimalintervention measures will be found ultimately
55 Parallel Execution by ASST and OsdRT As discussed inSection 35 ASST andOsdRT are used to support the parallelexecution of artificial Beijing The statistical data is collectedand stored in artificial society runtime database at runtimeASST downloads the infection data from the database anddisplays it in the map As shown in Figure 27 the infectedagents are distributed in the districts in BeijingThe snapshotgives the geospatial distribution of the H1N1 epidemic Thelegends of the agents in different status are listed in the left
right corner in Figure 27 Five statuses are shown in the figurehealth without vaccination health with vaccination incuba-tion being symptomatic and convalescence ASST providesa very direct feeling for the emergency experts Partly withthe situation from ASST Emergency Decision Organizationmakes the emergency response plans to the response of theemergency scenario With the help of plans the epidemicin actual society can be restrained In the meantime theinfluenza information is reflected in the Internet Accordingto Figure 8 the data acquisition data extraction and datastandardization of H1N1 influenza are processed in OsdRTOsdRT is implemented by Dr Caorsquos team from Institute ofAutomation from Chinese Academy of Sciences [58] Thetool provides not only the reports of H1N1 SARS hand-foot-mouth disease and intestinal diseases are all monitored Theweb page is shown in Figure 28 H1N1 reports from Internetare collected first Based on the data extraction of the reportsH1N1 case of distribution is obtained and shown in theincidencemap As a result H1N1 incidence data is regarded asthe knowledge which is sent to EECT and IMCTThe settingsof emergent events and intervention measures are modifiedin order to make artificial society approach actual society innext turn of computational experiments With the help ofOsdRT the loop of parallel execution process integrates ACPapproach
6 Conclusion
This work provides an integrated software framework forsocial Computing in emergencymanagement From a systemperspective KD-ACP provides a reliable flexible low costand effective platform for the scientists to do the research onemergency response problems The analysis and predictionof these problems with inherent complexity can be solvedby the repetition of possible alternative experiments onthe software framework Therefore KD-ACP which is anattempt to implement ACP approach seemed as an effectivecomputational framework to support the decision making inemergency response
Currently KD-ACP is used to study the H1N1 epidemicin Beijing in 2009 The most optimal compositions of inter-vention measures are testified to support the response tonext influenza However it is still a long way ahead beforethe automation systematization and practicalization of KD-ACP A lot of work will be carried out along several directionsas in the following
(i) The parallel execution process in KD-ACP is actuallynot strongly connected with the emergency responseorganization Traditional experience and theory stillplay a dominant role in the decision of emergencymanagement It is necessary to propose a mechanismof hall for workshop to facilitate the parallel executionloop in actual society
(ii) Section 33 introduces the principle and process ofmodelling artificial society Metamodels and modelsof artificial Beijing are built by GME Model transfor-mation and code generation are used to implementcodes from FSM based models But many details
Mathematical Problems in Engineering 25
Fangshan district Daxing district Fengtai district Xicheng district
Mentougou district Haidian districtShijingshan district Dongcheng district
Incubation
Symptomatic
Convalescence
Health withoutvaccination
Health withvaccinationLegend of agent
health status
Figure 27 The situation of artificial Beijing by ASST
H1N1
Districts of Beijing
H1N1 case ofdistribution
Incidence map
Distribution ofH1N1 reportsfrom internet
Incidence of H1N1
Fangshan districtLiangxiang town
Population 19000Incidence 135789
(1100000)
Figure 28 The implementation of OsdRT by Institute of Automation from Chinese Academy of Sciences
are still not illustrated due to the topic of thispaper The key techniques such as template of codeframework and model transformations from specificdomains to FSM will be focused and introduced indetail in our next paper
(iii) The performance of computational experiments inKD-ACP is still slower than expectation The opti-mization of simulation engine and models structureis necessary to improve the performance So it ispossible to run artificial society with the real worldin parallel
In summary KD-ACP paves a new way for scientists inboth emergencymanagement and computation simulation to
collaborate with each other in solving the emergency man-agement problems by social computing
Research Highlights
(i) KD-ACP is implemented based on ACP approach
(ii) Artificial Beijing is built with the help of KD-ACP
(iii) H1N1 is modeled to simulate the influenza in Beijingin 2009
(iv) Computational experiments testify the effectivenessof intervention measures
26 Mathematical Problems in Engineering
Conflict of Interests
The authors declare that there is no conflict of interestsregarding the publication of this paper
Acknowledgment
This work was supported in part by the National ScienceFoundation of China (nos 91024030 71303252 61403402and 91324013)
References
[1] W Mao and F-Y Wang Advances in Intelligence and SecurityInformatics Academic Press Oxford UK 2012
[2] E Bonabeau ldquoAgent-based modeling methods and techniquesfor simulating human systemsrdquo Proceedings of the NationalAcademy of Sciences of the United States of America vol 99 no3 pp 7280ndash7287 2002
[3] K M Carley D B Fridsma E Casman et al ldquoBioWar scalableagent-basedmodel of bioattacksrdquo IEEETransactions on SystemsMan and Cybernetics A Systems and Humans vol 36 no 2 pp252ndash265 2006
[4] J M Epstein ldquoModelling to contain pandemicsrdquo Nature vol460 no 7256 p 687 2009
[5] Los Alamos National Laboratory Reports about EPISIMS LosAlamos National Laboratory 2003
[6] C H Builder and S C Bankes ldquoArtificial societies a conceptfor basic research on the societal impacts of informationtechnologyrdquo RAND Report 149 RAND Corp 1991
[7] F-Y Wang ldquoComputational theory and method on complexsystemrdquo China Basic Science vol 6 no 5 pp 3ndash10 2004
[8] F-Y Wang ldquoToward a paradigm shift in social computing theACP approachrdquo IEEE Intelligent Systems vol 22 no 5 pp 65ndash67 2007
[9] F-Y Wang and S Tang ldquoArtificial societies for integratedand sustainable development of metropolitan systemsrdquo IEEEIntelligent Systems vol 19 no 4 pp 82ndash87 2004
[10] D Wen Y Yuan and X-R Li ldquoArtificial societies computa-tional experiments and parallel systems an investigation ona computational theory for complex socioeconomic systemsrdquoIEEE Transactions on Services Computing vol 6 no 2 pp 177ndash185 2013
[11] F Wang F Zeng and Y Yuan ldquoAn ACP-based approach forcomplexity analysis of E-commerce systemrdquo Complex Systemsand Complexity Science vol 5 no 3 pp 1ndash8 2008 (Chinese)
[12] F-Y Wang and P K Wong ldquoResearch commentary intelligentsystems and technology for integrative and predictivemedicinean ACP approachrdquoACMTransactions on Intelligent Systems andTechnology vol 4 no 2 pp 1ndash6 2013
[13] J Sifeng X Gang F Dong and H Chunpeng ldquoStudy on theemergency rescue decision support system of petrochemicalplant based on ACP theoryrdquo in Proceedings of the 6th Manage-ment Annual Meeting (MAM rsquo11) 2011 (Chinese)
[14] X Dong D Fan G Xiong F Zhu Z Zhang and Y Yao ldquoPar-allel Bus Rapid Transit (BRT) operation management systembased on ACP approachrdquo in Proceedings of the 6th ManagementAnnual Meeting (MAM rsquo11) Beijing China 2011
[15] B Ning F-Y Wang H-R Dong R-M Li D Wen and L LildquoParallel systems for Urban rail transportation based on ACP
approachrdquo Journal of Transportation Systems Engineering andInformation Technology vol 10 no 6 pp 23ndash28 2010 (Chinese)
[16] F-Y Wang J Zhao and S-X Lun ldquoArtificial power systemsfor the operation and management of complex power gridsrdquoSouthern Power System Technology vol 2 no 3 pp 1ndash11 2008(Chinese)
[17] M J North T R Howe N T Collier and J R Vos ldquoRepastsimphony development environmentrdquo in Proceedings of theAgent 2005 Conference on Generative Social Processes Modelsand Mechanisms 2005
[18] M J North T RHowe N T Collier and J R Vos ldquoRepast Sim-phony runtime systemrdquo in Proceedings of the Agent Conferenceon Generative Social Processes Models and Mechanisms 2005
[19] SWARM Development Group Swarm 211 Reference Guide2000 httpwwwswarmorg
[20] S Tisue and U Wilensky ldquoNetLogo a simple environment formodeling complexityrdquo in International Conference on ComplexSystems pp 16ndash21 2004
[21] S Luke C Cioffi-Revilla L Panait K Sullivan and G BalanldquoMASON a multiagent simulation environmentrdquo Simulationvol 81 no 7 pp 517ndash527 2005
[22] M Ichikawa H Tanuma Y Koyama andH Deguchi ldquoSOARSintroduction as a social microscope for simulations of socialinteractions and gamingrdquo in Organizing and Learning throughGaming and Simulation Proceedings of the 38th Conference ofthe International Simulation and Gaming Association (ISAGArsquo07) pp 149ndash158 2007
[23] D S Burke J M Epstein D A T Cummings et al ldquoIndividual-based computational modeling of smallpox epidemic controlstrategiesrdquo Academic Emergency Medicine vol 13 no 11 pp1142ndash1149 2006
[24] F-Y Wang ldquoParallel control and management for intelligenttransportation systems concepts architectures and applica-tionsrdquo IEEE Transactions on Intelligent Transportation Systemsvol 11 no 3 pp 630ndash638 2010
[25] Z Molnar D Balasubramanian and A Ledeczi ldquoAn introduc-tion to the genericmodeling environmentrdquo in Proceedings of theTOOLS Europe 2007 Workshop on Model-Driven DevelopmentTool Implementers Forum Zurich Switzerland 2007
[26] S Kelly and J-P Tolvanen Domain Framework in Domain-Specific Modeling Enabling Full Code Generation JohnWiley ampSons Hoboken NJ USA 2007
[27] B P Zeigler H Praehofer and T G Kim Theory of Modelingand Simulation Integrating Discrete Event and ContinuousComplex Dynamic Systems Academic Press San Diego CalifUSA 2000
[28] G Booch J Rumbaugh and I JacobsonThe Unified ModelingLanguage User Guide Addison-Wesley Reading Mass USA1999
[29] F Wagner R Schmuki T Wagne and P Wolstenholme Mod-eling Software with Finite State Machines A Practical ApproachCRC amp Taylor amp Francis Boca Raton Fla USA 2006
[30] T Espiner ldquoChina builds worldrsquos fastest supercomputerrdquo ZDNetUK October 2010
[31] B Chen and G Guo ldquoA two-tier parallel architecture for arti-ficial society simulationrdquo in Proceedings of the ACMIEEESCS26th Workshop on Principles of Advanced and DistributedSimulation (PADS rsquo12) pp 184ndash186 July 2012
[32] G Guo B Chen X G Qiu and Z Li ldquoParallel simulation oflarge-scale artificial society on CPUGPU mixed architecturerdquo
Mathematical Problems in Engineering 27
in Proceedings of the ACMIEEESCS 26th Workshop on Prin-ciples of Advanced and Distributed Simulation (PADS rsquo12) pp174ndash177 July 2012
[33] S-X Lun ldquoResearch on the classification of parallel executionmodes of ACP theoryrdquo Acta Automatica Sinica vol 38 no 10pp 1602ndash1608 2012
[34] F-YWang ldquoA framework for social signal processing and anal-ysis from social sensing networks to computational dialecticalanalyticsrdquo Science China vol 43 no 12 pp 1598ndash1611 2013
[35] G GuoOne Model Userrsquos Manual version 12 National Univer-sity of Defense University 2011
[36] Y Ge R Meng Z Cao X Qiu and K Huang ldquoVirtual citymdashanindividual-based digital environment for human mobility andinteractive behaviorrdquo Simulation Transactions of the Society forthe Modeling and Simulation International 2014
[37] M Kretzschmar and R T Mikolajczyk ldquoContact profiles ineight European countries and implications for modelling thespread of airborne infectious diseasesrdquo PLoS ONE vol 4 no6 Article ID e5931 2009
[38] W Duan and X Qiu ldquoFostering artificial societies using sociallearning and social control in parallel emergency managementsystemsrdquo Frontiers in Computer Science vol 6 no 5 pp 604ndash610 2012
[39] W J Edmunds C J OrsquoCallaghan and D J Nokes ldquoWhomixes with whom Amethod to determine the contact patternsof adults that may lead to the spread of airborne infectionsrdquoProceedings of the Royal Society B Biological Sciences vol 264no 1384 pp 949ndash957 1997
[40] J Banks J S Carson B L Neison and D M Nicol Discrete-Event System Simulation Prentice Hall Englewood Cliffs NJUSA 4th edition 2007
[41] H Yongxia ldquoInvestigation and analysis of college studentsrsquo workand restrdquo Value Engineering vol 31 no 3 2012 (Chinese)
[42] W Duan Z Cao Y Wang et al ldquoAn ACP approach to publichealth emergency management using a campus outbreak ofh1n1 influenza as a case studyrdquo IEEE Transactions on SystemsMan and Cybernetics Part A Systems and Humans vol 43 no5 pp 1028ndash1041 2013
[43] HWHethcote ldquoThemathematics of infectious diseasesrdquo SIAMReview vol 42 no 4 pp 599ndash653 2000
[44] G Chowell E Shim F Brauer and et al ldquoModelling thetransmission dynamics of acute haemorrhagic conjunctivitisapplication to the 2003 outbreak in Mexicordquo Statistics inMedicine vol 25 no 11 pp 1840ndash1857 2006
[45] Y Ge L Liu B Chen X Qiu and K Huang ldquoAgent-basedmodeling for InfluenzaH1N1 in an artificial classroomrdquo SystemsEngineering Procedia vol 2 pp 94ndash104 2011
[46] A R Tuite A L Greer MWhelan et al ldquoEstimated epidemio-logic parameters andmorbidity associatedwith pandemicH1N1influenzardquo CMAJ vol 182 no 2 pp 131ndash136 2010
[47] L Brouwers B Cakici M Camitz A Tegnell and MBoman ldquoEconomic consequences to society of pandemic H1N1influenza 2009mdashpreliminary results for Swedenrdquo Euro Surveil-lance vol 14 no 37 pp 1ndash7 2009
[48] Y Ge L Liu X Qiu H Song Y Wang and K Huang ldquoAframework of multilayer social networks for communicationbehavior with agent-based modelingrdquo Simulation Transactionsof the Society for the Modeling and Simulation International vol89 no 7 pp 810ndash828 2013
[49] W Duan X Qiu Z Cao X Zheng and K Cui ldquoHeterogeneousand stochastic agent-based models for analyzing infectious
diseasesrsquo super spreadersrdquo IEEE Intelligent Systems vol 28 no4 pp 18ndash25 2013
[50] The Preparation and Emergency Response Plans to PandemicPublished by Ministry of Health Ministry of Health 2005(Chinese)
[51] R M Fujimoto Parallel and Distributed Simulation SystemsWiley Interscience 2000
[52] F Cicirelli A Giordano and L Nigro ldquoDistributed simulationof situated multi-agent systemsrdquo in Proceedings of the 15thIEEEACM International Symposium on Distributed Simulationand Real Time Applications (DS-RT rsquo11) pp 28ndash35 September2011
[53] B Chen and G Guo ldquoA two-tier parallel architecturefor artificial society simulationrdquo in Proceedings of the 26thACMIEEESCS Workshop on Principles of Advanced and Dis-tributed Simulation (PADS rsquo12) pp 184ndash186 July 2012
[54] G Guo B Chen and X Qiu ldquoParallel simulation of large-scale artificial society with GPU as coprocessorrdquo InternationalJournal of Modeling Simulation and Scientific Computing vol4 no 2 Article ID 1350005 2013
[55] B Chen L-B Zhang X-C Liu and H Vangheluwe ldquoActivity-based simulation using DEVS increasing performance by anactivity model in parallel DEVS simulationrdquo Journal of ZhejiangUniversity Science C vol 15 no 1 pp 13ndash30 2014
[56] X-L Wang P Yang Z-D Cao et al ldquoQuantitative evaluationon the effectiveness of prevention and control measures againstpandemic influenza A (H1N1) in Beijing 2009rdquoChinese Journalof Epidemiology vol 31 no 12 pp 1374ndash1378 2010 (Chinese)
[57] S Mei A D van de Vijver L Xuan Y Zhu and P M ASloot ldquoQuantitatively evaluating interventions in the influenzaA (H1N1) epidemic on China campus grounded on individual-based simulationsrdquo Procedia Computer Science vol 1 no 1 pp1675ndash1682 2010
[58] X Qiu Z Cao and Z Li Mid-Term Reports of China NationalScience Foundation 91024030 2014
Submit your manuscripts athttpwwwhindawicom
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
MathematicsJournal of
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Mathematical Problems in Engineering
Hindawi Publishing Corporationhttpwwwhindawicom
Differential EquationsInternational Journal of
Volume 2014
Applied MathematicsJournal of
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Probability and StatisticsHindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Journal of
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Mathematical PhysicsAdvances in
Complex AnalysisJournal of
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
OptimizationJournal of
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
CombinatoricsHindawi Publishing Corporationhttpwwwhindawicom Volume 2014
International Journal of
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Operations ResearchAdvances in
Journal of
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Function Spaces
Abstract and Applied AnalysisHindawi Publishing Corporationhttpwwwhindawicom Volume 2014
International Journal of Mathematics and Mathematical Sciences
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
The Scientific World JournalHindawi Publishing Corporation httpwwwhindawicom Volume 2014
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Algebra
Discrete Dynamics in Nature and Society
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Decision SciencesAdvances in
Discrete MathematicsJournal of
Hindawi Publishing Corporationhttpwwwhindawicom
Volume 2014 Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Stochastic AnalysisInternational Journal of
Mathematical Problems in Engineering 9
Agent model andservice repository
repository
Emergency EventsConfigurationTool (EECT)
Open Source DataRegistration Tool (OsdRT)
Intervention MeasuresConfigurationTool (IMCT)
Load
Output Output
Internet
Emergencyresponse plans
Load
∙ Input models of emergencyevents∙ Output settings of agentbehavior in emergent events+ settings of emergent events
∙ Input models of interventionmeasures∙ Output settings of agentbehavior in interventionmeasures + settings ofintervention measures
+ emergency model
∙ Input data from internetnetworks∙ Output population data +emergency events data +intervention measures data
Artificial societypopulation and
geospatialenvironment database
Artificial societyruntime database
OutputThe most optimal
emergencyresponse plan
Actual society
OutputOutputOutput
Input
Load
Load
Load
Load
Output
Artificial SocietySituation Tool (ASST)
Emergency decisionorganization
ClusterRuntime infrastructure
Deploy
Deploy
Experimentsdesign tool (EDT)
Artificial societyinitial data file
Experimentplan
Artificial societymodels
Experiments Management
large scale artificial societyruntime infrastructure
Tool(EMT) +super computerTIANHE-1A
Load
∙ Input experiment plan +artificial society models +artificial society initial data file∙ Output artificial societyruntime state data
∙ Input description of artificialsociety + artificial society populationand geospatial environment society+ emergent events configuration +
intervention measures configuration∙ Output experiment plan + artificialsociety models + artificial societyinitial data file
∙ Input runtime statistical data∙ Output graphics charts ofartificial society statistical data +situation of artificial society
Figure 7 The computational experiments and parallel execution of KD-ACP
point The emergent events and intervention measures areconfigured by EECT and IMCT respectively The configu-rations of emergent events are used to simulate both thereal emergencies like SARS and H1N1 and the supposedemergencies for experiments Similarly the configurationsof intervention measures are also used to reproduce thereal one and simulate the supposed one The repeat of theemergency is used to verify the models while the supposedconfigurations are used to obtain the optimized decision planto the response of the possible emergencies
With the input of artificial society model and servicerepository artificial society population and geospatial envi-ronment database and the configurations discussed beforeEDT generates the experiment plans to meet the require-ments of research on emergency management The outputof EDT includes artificial society models artificial societyinitial data files and experiment plan The models aredownloaded from the repository while the data file is thecollection of data from the database to initialize the modelsWhen the models and data files are ready EMT loads theexperiment plan and deploys the models and data to thecluster or TIANHE-1A supercomputer [30] which was theworldrsquos fastest supercomputer built by National University of
Defense Technology (NUDT) in China in 2010 According tothe plan the experiments are done repeatedly on the largescale artificial society runtime infrastructure [31 32] by themultisample settingsThework process is the implementationof computational experiments in ACP approach
Traditionally emergency response plans are made byemergency management theories and experiences The onlyway to test the effective of plans is the feedback results of realworld ACP approach provides a novel method to supportemergency response plan making by parallel execution Asshown in Figure 7 the work process of KD-ACP is composedof two loops The inner loop composed by red arrowsdescribes the process of computational experiments while theouter loop of yellow arrows illustrates the process of parallelexecution During the runtime of computational experi-ments the statistical data of artificial society is collected andstored in the artificial society runtime database Based on thedatabase ASST outputs the customized situation of runningartificial society by graphics charts and situation mapsThe information is sent to the organizations of emergencydecision to support making the emergency response plansWith the help of computational experiments loop these plansare simulated repeatedly to find the most optimal one
10 Mathematical Problems in Engineering
Open Source Data Registration Tool (OsdRT)
Emergent EventsConfiguration Tool (EECT)
Intervention MeasuresConfiguration Tool (IMCT)
∙ Web information collecting
∙ Deep mining in social media
∙ Web information denoising
∙ Information filtering
Data acquisition
Actual society
Open source informationin social media
Internet
∙ Basic element extracting∙ Individuals and organizationsextraction∙ Sentiment analyzing andopinion mining∙ Social network analyzing
Data extraction
∙ Domain knowledge description∙ Information fusion and collisiondetection∙ Domain knowledge construction∙ Social media informationstandardization
Data standardization
Knowledge
Figure 8 The compositions of OsdRT [34]
Moreover the most optimal plan is used to the responseof emergency in actual society According to the idea ofparallel control in [24] the feedback of actual society ispartly collected from Internet networks by OsdRT As shownin Figure 8 OsdRT is composed of three components dataacquisition data extraction and data standardization Dataacquisition collects mines and filters information fromsocial sensing networks Data extraction includes basicelement extracting individual and organization extractingsentiment analyzing and social networks analyzing Datastandardization specifies the useful knowledge and sent it toconfiguration tools in KD-ACP
By processing in OsdRT the knowledge about emergentevents and intervention measures are analyzed first andregistered in the EECT and IMCT The registration updatesthe settings of emergent events and intervention measuresThis loop composed of yellow arrows implements the parallelexecution in ACP approach The implementation of parallelcontrol and management provides a data-driven approachthat considers both the engineering and social complexityfor modelling analysis and decision making in emergencymanagement
4 Modelling Beijing City with KD-ACP
41 How to Build the Artificial Beijing According to themodelling of artificial society in KD-ACP discussed beforethe Beijing city is modeled as follows
To meet the requirements of emergency managementsthe basic elements of artificial society are extended Asshown in Figure 9 six elements are required to simulate thecity agents environments transportation activity schedulecommunication and agent activity
411 Modelling Artificial Beijing Figure 10 shows the mainGUI of GME for the modelling of artificial modelling inpublic health events Metamodels are listed in the left area ofFigure 10 the list provides basic syntax elements for domainexperts to model artificial society Domain experts build
models based on the knowledge of their own MeanwhileGME supports hierarchy for building large scale systemsThesyntax symbol listed in GUI can be extended in new tab bydouble clicking Take agent for example the model of agentcan be detailed by edition in another tab page of agent
As shown in the center of Figure 10 the models ofartificial Beijing consist of five parts models of agent andenvironment domain models of public health events inter-ventionmodels controllermodels and services Agentmodeldescribes individuals in society it is composed of basicpopulation information action social relationships activityschedule and disease related information Activity schedulerepresents individualrsquos physical actionmodel focusing on thedaily action of agents Environment model includes physicalentities such as buildings playground transportations andagents contained in environment Domain models of publichealth events are composed of the propagation model of dis-ease disease state transition model and so on Interventionmodels include the settings of vaccination isolation and soon The models mentioned before are all FSM models Themechanisms of these models will be detailed in next sectionsController models and services are the public service mod-ules they are implemented in the development in MDT
412 Modelling Agents and Environments Under the speci-fication of FSM agent and environment models are imple-mented in two parts the state space and state transitionsThe state space is composed of the demographic attributesand behavior related attributes The transitions are triggeredwhen the conditions of states are satisfied As shown inFigure 11(a) the action of agent is changed when the ldquonexttimerdquo condition is satisfied in agent model while the agentslist is changed when the agent arrival condition is satisfied inthe environment model
413 Modelling Activities Agent activities come from theagent state transitions of actions such as movements andcommunications The actions of agents are instructed by theactivity schedule shown in Table 2 Activity schedule lists
Mathematical Problems in Engineering 11
Transportation EnvironmentsEnvironments
Agents
Artificial Beijing
Agent activityActivity schedule
Agent activity
Communication by socialrelationship networks
Communication by socialrelationship networks
Figure 9 The basic elements of artificial society
Syntax elements of DSM
Services
Models of agentenvironment andactivity schedule
Domain models of publichealth events
Intervention models
Modelingaspect ofdomainexperts
Controllermodels
Figure 10 The models of artificial Beijing in GME
all the actions with probability in one day for agents inboth normal and emergent situation [38] There are severaltypes of activity schedule in artificial Beijing student agentactivity schedule worker agent activity schedule emergentagent schedule and so on For example Table 2 gives an agentactivity schedule Upon the instruction of activity schedulestudent agent changes the actions by 119901
119894after state transitions
The119901119894in the tablemeans the action probability in the relevant
period In the duration from 0800 to 1200 a student agenteither goes to classroom to have class or goes to libraryto study The probability of class action is 119901
2while the
probability of study action is 1 minus 1199012
Agent behaviors are decided by the settings of activityschedule In addition to the daily activity schedulementionedbefore emergent activity schedules are also considered inour work Take public health events for instance an infectedagent changes schedule from a normal one to emergentone The workflow of a susceptible agent is illustratedin Figure 12 to show the change of behaviors After theinfection the agent is set in incubation phase Not all theincubation agents will become symptomatic Some of themturn back to being susceptible and some of them becomesymptomatic The symptomatic agents change their activityschedule from normal to emergent In the emergent case
12 Mathematical Problems in Engineering
Next actionNext environment (location)
Advance to next actiontime
Find next environment inactivity chain by action
Current actionCurrent environment (location)
Q
q0
120575 Find next action inactivity schedule by time
F
Σ
(a) The state transitions of agent
Q
q0
120575 F
Current agent list
Current pollution level
Next agents list
Next pollution level
Agents entered
Infected agents entered
Add agents to currentagents list
Compute pollution levelby infected agents
Σ
(b) The state transitions of environment
Figure 11 The state transitions of models
Table 2 The activity schedule of a student agent
Duration (Δ119905) Activity Location Probability 119879Location (minute)0000ndash0600 Sleep Dormitory 119875
0(100) 360
0600ndash0800 Breakfast Dormitoryrestaurant 1198751(068) 120
Sports-breakfast-travel Playground-dormitoryrestaurant 1 minus 1198751(032)
0800ndash1200 Class Classroom 1198752(077) 240
Study Library 1 minus 1198752(023)
1200ndash1400 Lunch Restaurant 1198753(090) 120
Lunch-shopping Restaurant-convenience store 1 minus 1198753(010)
1400ndash1800 Class Classroom 1198754(070) 240
Study Library 1 minus 1198754(030)
1800-1900Dinner Dormitoryrestaurant 119875
5(063)
60Sports-shopping Playgroundconvenience store 1198756(025)
Study Classroomlibrary 1 minus (1198755+ 1198756) (012)
1900ndash2200Rest Dormitory 119875
7(045)
180Dinner-rest Dormitoryrestaurant-dormitory 1198758(020)
Study Classroomlibrary 1 minus (1198757+ 1198758) (035)
2200ndash2400 Rest Dormitory 1198759(072) 120
Sleep Dormitory 1 minus 1198759(018)
infected agents go to hospital according to the treatmentschedule or stay in dormitory according to isolation scheduleAfter the treatment in hospital or self-immunoprocess agentsbecome healthy and immune of disease If the agents aretreated in the hospital they are not allowed to get out untilrecovered Moreover activity schedules are also influencedby the emergency response plans For example in the case ofisolation in emergency response plans the activity scheduleof agents who had contacted with infected is changed Onlythe locations including home and dormitory compose thisemergent schedule Likewise based on the statistical datafrom emergency response plans the additional possibilities
are added into activity schedulesThe behaviors of agents alsochange with activity schedules In the view of ACP approachthe injection of new emergency response plans implementsthe parallel execution of microagent behaviors
In another aspect contact frequency is also anothercrucial element to determine the infected rate in public healthevent It is important to model the contact frequency andcontact time of individuals Based on the studies on thecontact behavior of human being by questionnaire surveyEdmunds found that the contact frequency of individualcould be fitted approximately into a normal distributionand the mean and the standard deviation distributions are
Mathematical Problems in Engineering 13
Susceptible agent ImmuneYes
Healthy and immune
No
Choose schedule
Self-isolation Go to hospital
Staysymptomatic
No
Receive treatment
Contact withinfected agents
Symptomatic
No
Self-immunoprocess
Yes
Recovered
Yes
Incubation
Infection
No
Yes No
NoYes
Figure 12 The workflow of a susceptible agent
168 and 85 [39] So it is possible to apply normal randomvariable to model the contact frequency In our work Box-Muller method [40] is used to generate the random variableof contact frequency shown in the following
119865119862= 120583119865+ 120590119865(minus2 ln (120574
1))12 cos (2120587120574
2) (1)
in which 119865119862is the random variable of contact frequency 120583
119865
is the mean value of normal distribution 120590119865is the standard
deviation of normal distribution and 1205741and 120574
2are the
uniform random variables distributed in the interval [0 1]Based on (1) and survey data [41] the contact frequenciesof agent 119865contact(119860 119894) are discretized as in Table 3 within theconsideration of activity differences
Similarly the duration per contact between individualsis another key factor of AHC transmission The duration ofcontact could also bemodeled by the normal randomvariablein the following [42]
119879119862= 120583119879+ 120590119879(minus2 ln (120574
1))12 cos (2120587120574
2) (2)
in which 119879119862is the random variable of duration per contact
120583119879
is the mean value of normal distribution 120590119879
is thestandard deviation of normal distribution and 120574
1and 120574
2
are the uniform random variables distributed in the interval
[0 1] In addition because the time spent in specific location119879Location(119860 119894) also follows the random distribution like (3)in our work In order to make the duration of contact119879contact(119860 119894) shorter than the 119879Location(119860 119894) the mean andstandard deviation of (2) is set up in (3) and (4) [42]
120583119879=119879Location(120583119865+ 120590119865) (3)
120590119879=119879Location(10120583119865+ 10120590
119865) (4)
in which 119879Location is the time of an activity in a specificlocation set in Table 2 120583
119879and 120590
119879are the mean and standard
deviation of contact frequency As a result the duration ofagent 119879contact(119860 119894) in specific locations (listed in Table 2) isalso discretized in Table 3 within the consideration of activitydifferences
414 Modelling Transportation In artificial Beijing threekinds of travel models are considered walking inside thedistrict travelling by road networks and travelling by subwaynetworks According to the traveling models the agentmovements are implemented by the compositions of travelingmodels before So it is important to build the basic road
14 Mathematical Problems in Engineering
Table 3 The contact frequency and duration in different locations and activities by social relationships in case of student and teacher
Social relationship Activity Location 119879contact(119860 119894) (minute) 119865contact(119860 119894)
Student-student
Sleep Dormitory 0 0
Breakfast Restaurant 119873(52) 119873(32)Breakfast Dormitory 119873(52) 119873(21)Sports Playground 119873(21) 119873(53)Class Classroom 119873(21) 119873(21)Study Library 119873(21) 119873(21)Lunch Restaurant 119873(158) 119873(32)
Shopping Convenience Store 119873(203) 119873(21)Dinner Restaurant 119873(108) 119873(21)Dinner Dormitory 119873(158) 119873(32)Rest Dormitory 119873(63) 119873(52)
Teacher-student Class Classroom 119873(21) 119873(21)
Subway network of Beijing city (red lines) Road network of Beijing city (green lines)
Figure 13 The implementation of transportation for artificial Beijing
and subway networks With the help of Google map thekey points of road are sampled first and the road networksare generated by the links among point Similarly subwaynetworks are generated by the links between subway stationsmapped from the real stations in Google map Figure 13shows the road and subway networks in artificial Beijingwith the road networks in green and the subway networks inred Most of the districts in Beijing city are covered by thesetwo networks The algorithm of hierarchical route planningis proposed as follows
(i) Find the next position where agent will be located innext action
(ii) Obtain the travelingmodels from the statistical trans-portation data
(iii) Search the path from the starting position to thenearest road entry or subway stations
(iv) Search the optimal path from the starting station orroad entry to the nearest station or road exit of targetposition
(v) Search the path from the target subway station or roadexit to the target position
(vi) Attach the subway train number or the bus numberincluding the transfer information to the path accord-ing to the timetable of subway or bus
(vii) Connect all the paths obtained before Generate thepath from the starting position to the target position
Based on the algorithm for each agent in artificial Beijingthe transportation is simulated during the computationalexperiments It is worth noting that the activities insidethe buses or train cars are considered because the agentcommunication during travelling is also necessary to besimulated
Mathematical Problems in Engineering 15
(a) The implementation of agent in MDT (b) The code framework of agent
Figure 14 The agent model implementation in MDT [35]
415 The Implementation of Models by MDT All the modelsare developed in the MDT Figure 14 shows the GUI of thetool MDT is used to implement the state space and statetransitions of FSM models The code framework is shownin Figure 14(b) Using the C++ inheritance features and thetechniques of dynamic link library (dll) the models areencapsulated in the dll componentsThe flexible compositionmechanism is designed to support the models evolution incomputational experiments
416 The Description of Artificial Beijing by ASE As men-tioned in Section 34 ASE describes the artificial Beijingin macroview The statistical information of Beijing city iscollected by ASE Table 4 lists part of the description ofartificial Beijing generated by ASE
42 How to Obtain the Data of Beijing City
421 Population and Geospatial Environment Database ofBeijing City Based on the models developed for ArtificialBeijing it is necessary to generate the initial data to do thecomputational experiments Because the individual level dataare not available it is necessary to construct an individual-based population database for both accurate computationalexperiment and determining optimal decisions
According to the state spaces of models discussed inSection 412 the geospatial and population database of Arti-ficial Beijing is designed as shown in Figure 15 The kernelpart of database consists of the tables such as agent list tablegeospatial environment list table household list table andagent distribution table Agent list table is an individual leveltable used to store the data to initialize the state space of agentmodel such as id and gender age Geospatial environmentlist table is also an individual level table which stores thedata of environment attributes such as id street id andtype Household list table and agent distribution table storethe statistical data for the data generation for agent andenvironment models With the help of the geospatial andpopulation database the data generating of artificial Beijingis proposed in the next section
422 Generating Geospatial and Social Environment Datafor Artificial Beijing by PGET The algorithm of generatinggeospatial and social environment data for artificial Beijingis used to quantify the spatial distribution of population andformalize geospatial behavior of each agent The algorithmshown in Figure 16 is capable of generating a syntheticartificial society which allowsmultiresolution statistical datasocial interactive behavior and multilayer social networksto be integrated together The synthetic population canrepresent individual agents in the form of households andhousehold members and the synthetic population is statis-tically equivalent to a real population For each householdcharacteristics such as address household size family typesand relationships are generated Each person is described bycharacteristics such as age gender social role and correlatedlocations The algorithm provides an effective methodologyto reconstruct the computing environment in high resolutionby using statistical data in low resolution leading to betterprediction and management of emergencies The details ofalgorithm are illustrated in [36]
As shown in Figure 17 the implementation of the algo-rithm is embedded inside the GEPTThe collective input datasuch as the data files of population distribution is collected inthe tool first then the algorithm is activated to generate thedatabase The generation lasts almost twelve hours one timewith the specified parameters settings When the generationis finished the data of artificial Beijing is obtained in thedatabase with 19610000 agents and 16000 environments
5 H1N1 Experiments in ArtificialBeijing with KD-ACP
With the support of the artificial Beijing modeled before thecomputational experiments and parallel execution are alsoimplemented by KD-ACP The experiments are designed tofind the most optimal emergency decision response plansEpidemic in city is a typical scenario in emergency manage-ment H1N1 epidemic in Beijing in 2009 is used to be the usedcase to test KD-ACP According to Section 34 it is necessarytomodel emergencymanagement first in order to support thecomputational experiments
16 Mathematical Problems in Engineering
Table 4 The description of artificial Beijing generated by ASE
Statistical features ofartificial Beijing Descriptions of statistical features Data sourcedefault values
Agent roles Baby child primary student middle school student college student worker retired
Demographic data file ofBeijing
Environment types Super market store restaurant park stadium hospital school cliniccommercial building dormitory building residential building
Demographic data file ofBeijing
Demographic data Age distribution of population sex distribution of population householddistribution of population retired age distribution children distribution
Demographic data file ofBeijing
Environment data Distribution of all types of environments Environment distributiondata file of Beijing
Relationships Classmate relative college friendship family Demographic data file ofBeijing
Varying parameters
Age difference of couples119873(120583120590) 119873(03) isin [0 12]Minimummarriage age (male female) (22 20)Age difference of children119873(120583120590) 119873(13) isin [1 10]Enrollment rate of kindergarten 120572 08Enrollment rate of primary school 120573 1Enrollment rate of middle school 120574 1Enrollment rate of college 120575 08Entrance age of kindergarten 1205721015840 3Entrance age of primary school 1205731015840 7Entrance age of middle school 1205741015840 13Entrance age of middle school 1205751015840 19
Age distributioninformation table
Agent settable
Geospatialenvironment set table
Household settable
Figure 15 The geospatial and population database of Beijing city
51 Modelling Emergency Management The modelling ofemergency management consists of two parts the modellingof emergent events and the modelling of intervention mea-sures H1N1 and the intervention measures for H1N1 aremodeled in our case
511 Modelling H1N1 by MDT H1N1 Model in agent simu-lates the states transitions of health status and the relevantactions Referring to SIR (SusceptibleInfectiveRecovered)[43 44] models agent has three health statuses susceptibleinfected and healthy with immunity Only susceptible agentcan be infected by the contact with infected agents After
the three infected phases (incubation being symptomaticand recovery) of H1N1 model agent is healthy again It isworthy mentioning that the recovered agents can be infectediteratively when the immunity disappears gradually by timeThe state transitions are illustrated in Figure 18
According to [45] the latent period of H1N1 was in aWeibull distribution within people from one to seven days[46 47] The distribution was usually centered in the rangebetween one and three days AWeibull random variable [40]is used to model the latent period as described in
119879Lat = 120572 times [minus ln (1 minus 120574)]1120573+ 120592 (119879Lat ge 120592) (5)
Mathematical Problems in Engineering 17
Censusdata
Generate GUID
Assign individualhousehold
Generate familystructure
Assign family roleto every member
Endow individualwith age gender
social role
Distribute environmentlocations on the map
School Workplace Entertainmentlocations
Consumptionlocations
Assign school orworkplace to
individual
Correlatedindividual withother locations
Generate friendshipin school
Generate coworkerrelations inworkplace
Assign individualadministrative
region
Generate housebuildings
Locate householdto house building
Generate familyrelation
Generateneighborhood
relation
PopulationEnvironment
Environment listSocial network
Figure 16 The algorithm of generating geospatial and social environment data [36]
Edit
Edit
Edit
Edit
Edit
Edit
BrowseStatistical characteristic of population
Geospatial population data
Population data of street
Population types
Geospatial and social environment generation tool
Workspaces
Parameters distribution settings
Household types
Roles in household
Household structure
Social roles
BrowseBrowseBrowseBrowseBrowseBrowse
Population distribution data
File path
Retired distributionPop distributionHousehold numHousehold distributionSex distributionAge distributionData type
Figure 17 The GUI of Population and Geospatial Environment generation Tool (PGET)
18 Mathematical Problems in Engineering
Infected
SusceptibleHealth
withimmunity
Infected
by contac
t
with in
fected
agents
Recovery
Increase before beingsymptomatic symptomatic
Decrease after being
Decrease by time
Infectivity
Immunity
Figure 18 The transitions of agent health status
119879Lat denotes the duration of latent period 120592 120572 and 120573 arethe location parameter scale parameter and shape parameterof Weibull distribution respectively 120574 is a uniform randomnumber in the range [0 1] According to statistics [47] 120592 120572and 120573 are set as 0 18 and 121 respectively Then the mean(standard deviation) of latent period is calculated as 159 days(058 day2) Furthermore infectious period is set as 7 days
A susceptible studentrsquos probability of getting infecteddepended on the infectivity of infectious agents his ownimmunity level the duration of the contact action and soon The infectivity of an infected agent evolved with timeduring the infectious period According to the statistics ofinfectivity in the chart with red bars in Figure 19 an infectedstudent had the highest level of infectivity in the second dayafter he had the first symptom The infectivity levels in otherdays in comparison with the second day are in Table 5 Theday labeled as ldquominus1rdquo means the day before the starting of thesymptomatic period
Based on the algorithm discussed before theH1N1modelis developed with the help of MDT The settings of influenzamodel will be discussed in Section 521
512 Modelling Intervention Measures by MDT In corre-spondence intervention measures model is used to ceasethe outbreak of H1N1 As discussed in [42 48 49] if theappropriate measures are taken when the infectious diseaseemerges the transmission of the disease could be sloweddown and the damage could be decreased Therefore it isnecessary and important to design emergency interventionmeasures in the artificial Beijing According to the MinistryofHealth inChina the intervention policies [50] are designedto control the spread of influenza such as H1N1 The inter-ventionmeasures of these policies including are interventionsactivated time vaccination rate antibiotic delay to hospitalisolation duration close workspace duration and limitationof activities listed in Table 6 Similarly the models of theseintervention measures are also developed with the help ofMDT Using the reasonable data ranges listed in Table 6intervention models can be initialized to do the computa-tional experiments These experiments are used to show how
Symptomatic1 2 3 4 5 6 7minus1
Recovered
Tim
e (da
y)
Infe
ctiv
ity
Figure 19 The distribution of infectivity rate of an infected agent[37]
Table 5 The proportion of infectivity
Days (119894) minus1 1 2 3 4 5 6 6Proportion of infectivity 03 05 1 08 04 02 01 005
to restrain the H1N1 transmissionThe settings of these inter-vention measures in IMCT are introduced in Section 522
52 Experiment Settings
521 H1N1 Settings by EECT EECT is designed to initializethe emergent events models such as H1N1The settings comefrom the specified emergent event models As shown inFigure 20 in theH1N1 used case the settings are composed ofthe distribution of infection source infection rate infectionperiod and so on The infection rate changes with envi-ronments and infection periods The settings are initializedfrom both the research of medicine on H1N1 and the data ofH1N1 outbreak in Beijing in 2009The process of influenza inartificial Beijing is simulated to repeat the possible scenariosof H1N1 break During the computational experiments ofartificial Beijing theH1N1 settings such as the period of beingsymptomatic are updated by the dynamic data registrationsfrom OsdRT The updating corrects the H1N1 model duringthe experiments With the help of OsdRT artificial Beijing isable to approach actual society It is also an implementationof parallel execution
522 Intervention Measures Settings by IMCT In the H1N1case the settings of intervention measures are shown inFigure 21 in IMCT These measures are designed under theinstruction of Emergency Decision Organizations like ChinaDisease Center (CDC) The measures such as vaccinationrate the rate of closedworkspace and the starting time can beinitialized in the tool Emergency response plans are imple-mented in these settings As illustrated in Figure 7 all the pos-sible compositions within value settings in reasonable inter-vals are tested to find the most optimal emergency responseplan For example different compositions of measures suchas vaccination rates and rates of workspace closed are usedto design the plans of experiments By the analysis of theresults the best composition of interventionmeasures can beobtainedThe analysis will be discussed in case study in detail
Mathematical Problems in Engineering 19
Table 6 The intervention measures of influenza
Intervention measures ValuesInterventions activated Time 119879Intervention isin [0 80th day]Vaccination rate 119877Vaccination isin [0 1]Antibiotic rate 119877Antibiotic isin [0 1]Delay to hospital 119867ospital(119860 119894) 119879infected lt 119905 lt 119879recovered + 119879HDelay 119879HDelay isin (0 3th day]Isolation duration 119868solation(119860 119894) 119879Intervention lt 119905 lt 119879Intervention + 119879IDuration 119879IDuration isin (0 14th day]Close workspace duration 119862lose(119864119894) 119879Intervention lt 119905 lt 119879Intervention + 119879CWDuration 119864119894 isinWorkspace 119879CWDuration isin (0 7th day]Limitation of agent activity duration 119871 imitation(119860 119894) 119879Intervention lt 119905 lt 119879Intervention + 119879LAADuration 119879LAADuration isin (0 7th day]119877Vaccination means the vaccination rate of artificial Beijing while 119877Antibiotic means the antibiotic rate used for infected agents in hospital119867ospital (119860119894) means that the infected agent 119860119894 is sent to hospital he or she will receive the retreatment immediately119868solation (119860119894) means that the infected agent 119860119894 is isolated at home he or she can only contact with the other agents until isolation duration is over119862lose (119864119894) means that the environment 119864119894 typed of workspace is closed until the close workspace duration is over119871 imitation (119860119894) means that the activities of agent 119860119894 are limited only the workspace and home are allowed until limitation duration is over
Incubation Weibulldistribution
Symbolic LogNormaldistribution
Recovering LogNormaldistribution
H1N1influenza
Settings of H1N1 influenza model
Disease model
Transmissionmodel
Figure 20 The settings of H1N1 in EECT
It is worth notifying that the settings of interventionmeasures are also influenced by OsdRT For example theH1N1 outbreak areas are notified by OsdRTThe interventionmeasures such as close workspace are used only in theoutbreak areas Therefore the unnecessary computation ofintervention measures is avoided So the data injection byOsdRT not only makes artificial society approach actualsociety but also increases the performance of computationalexperiments
53 Experiments by EMT When the initialization of arti-ficial Beijing is ready EMT is used to do the computa-tional experiments EMT is composed of two parts thecentral controller in the central node and the residen-tial service deployed in the computational nodes Centralcontroller loads the experiment plans and controls the whole
experiments process the GUI of controller is shown inFigure 22 EMT manages the computational experimentsin four steps cluster configuration load experiment plansmodels deployment and running experiments
531 Cluster Configuration Before the configuration clusterinformation is collected from the residential services insidethe cluster The information includes nodes amount nodenames node IPs and CPU occupations It is used to quantifythe computing power of the clusterThe cluster configurationis based on this information The nodes are selected forexperiments Then the LP (local process) [51] number foreach node is set according to the number of the CPU kernelsAfter the configuration the experiment plans are loaded tocustomize the models and data
20 Mathematical Problems in Engineering
Streets of close workspaces
Limitations ofactivities
IsolationsRespirator
Closeworkspace
Antibiotic
VaccinationHospital
Surveillance
Figure 21 The settings of intervention measures in IMCT
Information ofexperiment plan
List of experiment plans
Start | Pause | Reset | Stop | Speedup | Slowdown | Back
NodeName | CPU occupation | Memory occupation |Network | Models | Status
Figure 22 The deployment and running of experiments in EMT
532 Load Experiment Plan As discussed in Section 35 thedescription of experiment plans is composed of the infor-mation in three aspects models and the initial data for theexperiments the deployment mapping tables from the mod-els to the nodes and the plan of experiment executionWhenthe experiment plan is loaded by the EMT the file namesand model descriptions of models and data are listed Withthe support of cluster configuration models can be mappedto the nodes in the GUI according to the computationrequirements EMT also provides the setting of experimentsitself including the running times start time and end time
533 Models and Data Deployment Models are stored inthe agent population and services repository According tothe settings in experiment plan the customized models aredownloaded from the repository first then the models suchas agent model environment model emergent events modeland intervention measures model are integrated to be thecompositionmodels of artificial BeijingThemodels and datalocations in the nodes are configured in the ldquomodel pathrdquoand ldquodata pathrdquo in the GUI of central controller Finallythe models are uploaded to the nodes in cluster with theconsideration of partition The partition of artificial society
Mathematical Problems in Engineering 21
models is different from the multiagent system in [52]Models in artificial society are divided into three levels agentsand environments grids and city Agents and environmentsare encapsulated in grids which compose the whole cityThus the partition of artificial society becomes the partitionof grids The parallelism degrees are found to support thepartition of models agent independence and grid indepen-dence According to the parallelisms agents are partitioned toLPs according to the population distribution in grids Agentmovements among grids are simulated by the movementevents among grids in different LPs As a result a two-tierparallel architecture is proposed to support the simulation ofartificial society The architecture is detailed in [53]
534 Running of Experiments When the preparation isready the experiments can be done by the central controllerof EMT The running and control commands of the experi-ments are grouped into three aspects start pause or stop theexperiments get the running information of the nodes suchas the snapshot of the desktop of node With the help of thesecommands users can run the computational experimentsto do the research on the emergency management It isworth notifying that the simulation engine is optimized inorder to support the ten million agent simulation UsingGPU as coprocessor a two-tier parallel simulation engine isdesigned with support of MPI and OpenCL through phasedsynchronization [54] One-sided communication is used forreflection of remote simulation objects and message passingbetween processes A general kernel function prototype iselaborately designed and conditionally compiled for execu-tion on both CPU and GPU Moreover the optimizationoperations like load balance is developed under the instruc-tion of activity-based simulation [55] The densely populatedgrids are given more CPU and memory resource accordingto the activity predictions
54 Results of H1N1 Experiments with KD-ACP On thebasis of artificial society models and the intervention mea-sures mentioned in Section 512 a series of computationalexperiments are performed to study the H1N1 influenza inBeijing In our case of artificial Beijing 19610000 agentsare simulated in the cluster within 48 CPU cores and 128Gmemories the nodes are connected by kilomega networks Ittakes 18 hours to simulate a 250-day disease spreading
In order to illustrate how the optimal plans are obtainedbyKD-ACP the experiments are divided into four groups theexperiments ofmodel validation the experiments of sensitiv-ity analysis of vaccination rate the experiments of sensitivityanalysis of isolation and the experiments of the combinationof interventionmeasuresThese four groups show the generalprocess of the research on emergency management in H1N1influenza Firstly the models of H1N1 are validated withthe support of historical data Secondly the traditionalmedical interventions of pandemic such as vaccination areanalyzed to find the optimal plan Thirdly the traditionalnonmedical interventions of pandemic such as isolation areanalyzed to find the optimal plan Finally the combination ofinterventions is tested to show the combined effect
It is worth mentioning that all the experiments in ourwork are performed 100 times These experiments are ini-tialized with different settings of interventions in IMCT Theresults shown in the figures are the mean values
541 The Experiments of Model Validation In our artifi-cial Beijing the models of H1N1 influenza are built basedon the existing researches Wang and her colleagues hadstudied the H1N1 influenza by SEIR (SusceptibleExposedInfectiveRecovered) models [56] With the support of theirwork we build the models of artificial Beijing in the mannerof multiagents
In Figure 23 the control group shown in red line isdrawn based on the realistic statistical data collected in H1N1influenza in Beijing in 2009 The influenza lasts more thansix months and more than 170000 people are infectedComparing with the control group the simulated resultswithout interventions are drawn in blue It is obvious that thesimulated case fits the control group well before 70 days Butthe ascending of control group is slowed after 70 days andthe peak value (17883) is much less than the simulated results(25135) It is because the vaccination is activated at day 67 incontrol group according to the intervention measures usedin 2009 The restrain by vaccination becomes effective eightdays later The change of infected number after vaccinationis shown by red spotted line The trend of control group isdifferent from the simulation data after 75 days
However the simulated results (blue line) in compu-tational experiments fit the control group (as shown inFigure 23) in the first 70 days when no intervention mea-sures are executed The key features such as effective basicreproduction number (119877
0) are also almost the same in both
cases Therefore it can be concluded that the models ofartificial society built in microview are validated by therealistic statistical data in influenza in Beijing in 2009
In this group of experiments no interventions are exe-cuted in order to validate the models Therefore the simu-lated results of this extreme case which cannot happen todaycould be obtained As shown by blue line the number ofinfected agents without interventions reaches peak value atday 91 with 25135 consisting of 8052 in symptomatic phaseand 28950 in recovery phase It is obvious that the totalnumber of infected agents grows slowly at the beginningtime Then the number increases quickly when the influenzabreaks and it reaches its maximum value with 383870 in theend This phenomenon can be explained as in the followingAt the beginning of influenza due to the limited numberof infected agents the propagation of the disease remainsslow but many agents become susceptible When the numberof infected agents increases rapidly the spreading becomessignificant However the number of infected agents decreasesafter day 91 It is because the disease spreading relies on thesocial networks and spatial contact networks [49] It meansthat the infected agents can only infect agents either in thesocial relationship networks or in the spatially contacted
542 The Experiments of Sensitivity Analysis of VaccinationRate According to Section 512 vaccination is a typical
22 Mathematical Problems in Engineering
0
Infe
cted
agen
ts
0 50 100 150 200 250
Statistical dataSimulation data
25000
20000
15000
10000
5000
Day d
Figure 23 The results with no intervention executed and the com-parison with realistic statistical data in Beijing in 2009
medical intervention in influenza Traditionally the effective-ness of vaccination can only be evaluated after the influenzaand the key settings such as vaccination rate are determinedby the experience It is always difficult for emergency man-agement organizations to find the most optimal vaccinationrate in advanceTherefore the sensitivity of vaccination rate isanalyzed in this group of experiments The settings are listedin Table 7 average results are shown in Figure 24
Vaccination rates are set linearly (10 30 50 and70) to test the changes of influenza The peak value ofinfected agents decreases from 25135 22392 15787 6907to 638 respectively It shows that the more the agents arevaccinated the smaller the peak value of the infected agentswill be In the case of 70 vaccination the total infected agentis only 13799 It means that the influenza in this case does nothappen due to the high vaccination rate
The change of peak values also shows an interestingphenomenon Vaccination rates increase linearly the peakvalues decrease slower than the ascending of vaccination ratesat beginning but the descending becomes more and morefaster with the linear increase of vaccination rate It meansthat the small scale of vaccination (less than 30) does notmake senseWith the increase of vaccination rate the numberof susceptible agents decreases significantly In sequencethe possibility of infections is lowered greatly Additionallyit could be found that the decrease of peak values comeslater than the increase of the vaccination rates This is quitereasonable according to Mei et alrsquos work in [57]
As a result it is concluded that high vaccination rate iseffective in the control of H1N1 influenza though costly
543 The Experiments of Sensitivity Analysis of IsolationAccording to Section 512 isolation is a typical nonmedicalintervention It is the most used intervention measure in
Table 7 The initial settings of experiments of vaccination
Parameters ValueAgent count 19610000Initial infected count 40Interventions activated time(119879Intervention)
0
Vaccination rate (119877Vaccination) 0 10 30 50 70
0
Infe
cted
agen
ts
Loading intervention of vaccination
No vaccination10 vaccination30 vaccination
50 vaccination70 vaccination
0 50 100 150 200 250
25000
20000
15000
10000
5000
Day d
Figure 24 The results of loading intervention of vaccination
the emergency management in influenza As mentionedbefore a problem also exists that the measure cannot bevalidated before the intervention Therefore the sensitivityof isolation is analyzed in this group of experiments It isworth mentioning that the isolation is usually combinedwith the hospital policy It means the infected individual issent to hospital meanwhile the people who contact withthe infected are all isolated As a result parameters delay tohospital and isolation duration are designed together in thesettings of experiment Delay to hospital stands for the periodfrom infection time to hospital time while Isolation durationrepresents the period when an agent cannot contact others ifhe has contact with an infected agent The settings of theseparameters are listed in Table 8 average results are shown inFigure 25
Firstly the parameter delay to hospital is simulated aloneto test the sensitivity The simulated results are shown insolid line in Figure 25 The peak value decreases from 6708to 2218 when the delay to hospital decreases from 15 daysto 1 day The half day change brings the 67 reductionMoreover the peak value almost approximates to zero inthe case of 05 days The influenza does not happen in thiscase It can be concluded that the earlier the infected agents
Mathematical Problems in Engineering 23
Table 8 The initial settings of experiments of isolation
Parameters ValueAgent count 19610000Initial infected count 40Interventions activated time (119879Intervention) 80Delay to hospital (119879HDelay) isolation duration (119879IDuration) (15 0) (1 0) (05 0) (15 3) (1 3) (05 3)
0 50 100 150 200 2500
1000
2000
3000
4000
5000
6000
7000
Infe
cted
agen
ts
Loading intervention of isolation
Delay to hospital = 15
Delay to hospital = 10
Delay to hospital = 05
Delay to hospital = 15 isolation duration = 3
Delay to hospital = 10 isolation duration = 3
Delay to hospital = 05 isolation duration = 3
Day d
Figure 25 The results of loading intervention of isolation
are sent to hospital the less the agents will be infectedSo the construction of effective mechanism to find infectedindividuals is important in the emergency management ofinfluenza
Secondly the parameter isolation duration is added tosimulate the isolation of interventionObviously the isolationmeasure decreases the infected agents greatly Compared tothe parameter delay to hospital of 15-day case the peak valuedecreases from 6708 to 362when the isolation is added in theinterventionThe reduction of 6346 agents comes from the 3-day isolation of agents who had contacted with the infectedagents However the decrease of parameter delay to hospitalcannot bring the similar reduction in the 3-day isolationcases According to the figure the peak value decreases from362 to 51 from 15-day case to 05-day case only 311 agents aresaved from infection As a result the sensitivity of simulatedresults does not change obviously with delay to hospital inthe case of isolation It can be concluded that isolation is amore effective measure Though it is not easy for emergencymanagement organizations to decrease the delay to hospitaldue to the current monitoring mechanism the influenza canstill be controlled by isolation
544 The Experiments of Combination of Intervention Mea-sures Isolation is not the only nonmedical intervention ininfluenza close workspace is another common measureWhen close workspace is activated the workspaces are allclosed agents can only stay home at the closed time Luckilyisolation and close workspace are independent from oneanother It is possible for emergency management organi-zations to use these two measures together So this groupof experiments is designed to obtain the effectiveness of thecombination of intervention measures The parameters areinitialized with different settings listed in Table 9
Parameters delay to hospital and isolation duration is setin the middle values (1 day 3 days) according to the lastsection According to the experience close duration is set asseven days It is known that close workspace is a really costintervention measure the activated time is a key parameterWith the fixed parameters mentioned before the changeof interventions activated time is analyzed in this group ofcomputational experimentsThe average results are shown inFigure 26
Compared with the simulated results of experimentsdiscussed above Figure 26 strongly illustrates that influenzais ceased when the intervention measures are combinedtogether The peak value of infected agents is 141 in the ldquo80thdays of close-workspace and isolation with 1 day delay tohospitalrdquo case while 25135 in ldquono interventionrdquo case 2218 inldquo1 day delay to hospitalrdquo case 203 in ldquo1 day delay to hospitaland isolationrdquo case The interventions bring a great reductionof infection the peak values decrease from ten thousandslevel to hundred level
On the other hand the peak value even decreases to 108when the execution time of close workspace is 10 days earlierThe date of maximum value of infected agents is advancedfrom the 135th day to the 108th day The whole circle of H1N1influenza is reducedThe results of this group of experimentsshow that intervention measures should be activated as soonas possible and the compositions of interventions are moreeffective in emergency management of H1N1 influenza
In summary based on the general process lots of exper-iments could be done by KD-ACP to study emergency man-agement in H1N1 influenza Many conclusions are obtainedbased on the analysis of the experiments in this sectionAccording to these conclusions themost optimal response inour case study is the case of ldquo70th days of close-workspace andisolation with 1 day delay to hospitalrdquo It is because this casehas the fewest infected agents relatively But the modelling ofH1N1 influenza is not sufficient Economic and social factorsof interventions such as the money cost by vaccination andisolation the influence of the close-workspace in society
24 Mathematical Problems in Engineering
Table 9 The initial settings of experiments of isolation and closeworkspace
Parameters ValueAgent count 19610000Initial infected count 40Interventions activated time (119879Intervention) 70th day 80th dayDelay to hospital (119879HDelay) 1 dayIsolation duration (119879IDuration) 3 daysClose workspace duration (119879CWDuration) 7 days
50 100 150 200 2500
50
100
150
200
Infe
cted
agen
ts
Loading intervention of isolation and close workspace
Delay to hospital = 1 isolation duration = 3Start close time = 70 close duration time = 7Delay to hospital = 1 isolation duration = 3Start close time = 80 close duration time = 7
Day d
Figure 26The results of loading interventions of isolation and closeworkspace
and even opinion transmission of H1N1 in internet are notconsidered in our case study Without the analysis of thesefactors it does not make sense to talk about the most optimalintervention measures However KD-ACP had provided asoftware framework and general process of the research onemergency management Supported by KD-ACP the expertsof emergency management are able to study emergent eventssuch as H1N1 influenza as detailed as possible Based on theexperiments and analysis again and again the most optimalintervention measures will be found ultimately
55 Parallel Execution by ASST and OsdRT As discussed inSection 35 ASST andOsdRT are used to support the parallelexecution of artificial Beijing The statistical data is collectedand stored in artificial society runtime database at runtimeASST downloads the infection data from the database anddisplays it in the map As shown in Figure 27 the infectedagents are distributed in the districts in BeijingThe snapshotgives the geospatial distribution of the H1N1 epidemic Thelegends of the agents in different status are listed in the left
right corner in Figure 27 Five statuses are shown in the figurehealth without vaccination health with vaccination incuba-tion being symptomatic and convalescence ASST providesa very direct feeling for the emergency experts Partly withthe situation from ASST Emergency Decision Organizationmakes the emergency response plans to the response of theemergency scenario With the help of plans the epidemicin actual society can be restrained In the meantime theinfluenza information is reflected in the Internet Accordingto Figure 8 the data acquisition data extraction and datastandardization of H1N1 influenza are processed in OsdRTOsdRT is implemented by Dr Caorsquos team from Institute ofAutomation from Chinese Academy of Sciences [58] Thetool provides not only the reports of H1N1 SARS hand-foot-mouth disease and intestinal diseases are all monitored Theweb page is shown in Figure 28 H1N1 reports from Internetare collected first Based on the data extraction of the reportsH1N1 case of distribution is obtained and shown in theincidencemap As a result H1N1 incidence data is regarded asthe knowledge which is sent to EECT and IMCTThe settingsof emergent events and intervention measures are modifiedin order to make artificial society approach actual society innext turn of computational experiments With the help ofOsdRT the loop of parallel execution process integrates ACPapproach
6 Conclusion
This work provides an integrated software framework forsocial Computing in emergencymanagement From a systemperspective KD-ACP provides a reliable flexible low costand effective platform for the scientists to do the research onemergency response problems The analysis and predictionof these problems with inherent complexity can be solvedby the repetition of possible alternative experiments onthe software framework Therefore KD-ACP which is anattempt to implement ACP approach seemed as an effectivecomputational framework to support the decision making inemergency response
Currently KD-ACP is used to study the H1N1 epidemicin Beijing in 2009 The most optimal compositions of inter-vention measures are testified to support the response tonext influenza However it is still a long way ahead beforethe automation systematization and practicalization of KD-ACP A lot of work will be carried out along several directionsas in the following
(i) The parallel execution process in KD-ACP is actuallynot strongly connected with the emergency responseorganization Traditional experience and theory stillplay a dominant role in the decision of emergencymanagement It is necessary to propose a mechanismof hall for workshop to facilitate the parallel executionloop in actual society
(ii) Section 33 introduces the principle and process ofmodelling artificial society Metamodels and modelsof artificial Beijing are built by GME Model transfor-mation and code generation are used to implementcodes from FSM based models But many details
Mathematical Problems in Engineering 25
Fangshan district Daxing district Fengtai district Xicheng district
Mentougou district Haidian districtShijingshan district Dongcheng district
Incubation
Symptomatic
Convalescence
Health withoutvaccination
Health withvaccinationLegend of agent
health status
Figure 27 The situation of artificial Beijing by ASST
H1N1
Districts of Beijing
H1N1 case ofdistribution
Incidence map
Distribution ofH1N1 reportsfrom internet
Incidence of H1N1
Fangshan districtLiangxiang town
Population 19000Incidence 135789
(1100000)
Figure 28 The implementation of OsdRT by Institute of Automation from Chinese Academy of Sciences
are still not illustrated due to the topic of thispaper The key techniques such as template of codeframework and model transformations from specificdomains to FSM will be focused and introduced indetail in our next paper
(iii) The performance of computational experiments inKD-ACP is still slower than expectation The opti-mization of simulation engine and models structureis necessary to improve the performance So it ispossible to run artificial society with the real worldin parallel
In summary KD-ACP paves a new way for scientists inboth emergencymanagement and computation simulation to
collaborate with each other in solving the emergency man-agement problems by social computing
Research Highlights
(i) KD-ACP is implemented based on ACP approach
(ii) Artificial Beijing is built with the help of KD-ACP
(iii) H1N1 is modeled to simulate the influenza in Beijingin 2009
(iv) Computational experiments testify the effectivenessof intervention measures
26 Mathematical Problems in Engineering
Conflict of Interests
The authors declare that there is no conflict of interestsregarding the publication of this paper
Acknowledgment
This work was supported in part by the National ScienceFoundation of China (nos 91024030 71303252 61403402and 91324013)
References
[1] W Mao and F-Y Wang Advances in Intelligence and SecurityInformatics Academic Press Oxford UK 2012
[2] E Bonabeau ldquoAgent-based modeling methods and techniquesfor simulating human systemsrdquo Proceedings of the NationalAcademy of Sciences of the United States of America vol 99 no3 pp 7280ndash7287 2002
[3] K M Carley D B Fridsma E Casman et al ldquoBioWar scalableagent-basedmodel of bioattacksrdquo IEEETransactions on SystemsMan and Cybernetics A Systems and Humans vol 36 no 2 pp252ndash265 2006
[4] J M Epstein ldquoModelling to contain pandemicsrdquo Nature vol460 no 7256 p 687 2009
[5] Los Alamos National Laboratory Reports about EPISIMS LosAlamos National Laboratory 2003
[6] C H Builder and S C Bankes ldquoArtificial societies a conceptfor basic research on the societal impacts of informationtechnologyrdquo RAND Report 149 RAND Corp 1991
[7] F-Y Wang ldquoComputational theory and method on complexsystemrdquo China Basic Science vol 6 no 5 pp 3ndash10 2004
[8] F-Y Wang ldquoToward a paradigm shift in social computing theACP approachrdquo IEEE Intelligent Systems vol 22 no 5 pp 65ndash67 2007
[9] F-Y Wang and S Tang ldquoArtificial societies for integratedand sustainable development of metropolitan systemsrdquo IEEEIntelligent Systems vol 19 no 4 pp 82ndash87 2004
[10] D Wen Y Yuan and X-R Li ldquoArtificial societies computa-tional experiments and parallel systems an investigation ona computational theory for complex socioeconomic systemsrdquoIEEE Transactions on Services Computing vol 6 no 2 pp 177ndash185 2013
[11] F Wang F Zeng and Y Yuan ldquoAn ACP-based approach forcomplexity analysis of E-commerce systemrdquo Complex Systemsand Complexity Science vol 5 no 3 pp 1ndash8 2008 (Chinese)
[12] F-Y Wang and P K Wong ldquoResearch commentary intelligentsystems and technology for integrative and predictivemedicinean ACP approachrdquoACMTransactions on Intelligent Systems andTechnology vol 4 no 2 pp 1ndash6 2013
[13] J Sifeng X Gang F Dong and H Chunpeng ldquoStudy on theemergency rescue decision support system of petrochemicalplant based on ACP theoryrdquo in Proceedings of the 6th Manage-ment Annual Meeting (MAM rsquo11) 2011 (Chinese)
[14] X Dong D Fan G Xiong F Zhu Z Zhang and Y Yao ldquoPar-allel Bus Rapid Transit (BRT) operation management systembased on ACP approachrdquo in Proceedings of the 6th ManagementAnnual Meeting (MAM rsquo11) Beijing China 2011
[15] B Ning F-Y Wang H-R Dong R-M Li D Wen and L LildquoParallel systems for Urban rail transportation based on ACP
approachrdquo Journal of Transportation Systems Engineering andInformation Technology vol 10 no 6 pp 23ndash28 2010 (Chinese)
[16] F-Y Wang J Zhao and S-X Lun ldquoArtificial power systemsfor the operation and management of complex power gridsrdquoSouthern Power System Technology vol 2 no 3 pp 1ndash11 2008(Chinese)
[17] M J North T R Howe N T Collier and J R Vos ldquoRepastsimphony development environmentrdquo in Proceedings of theAgent 2005 Conference on Generative Social Processes Modelsand Mechanisms 2005
[18] M J North T RHowe N T Collier and J R Vos ldquoRepast Sim-phony runtime systemrdquo in Proceedings of the Agent Conferenceon Generative Social Processes Models and Mechanisms 2005
[19] SWARM Development Group Swarm 211 Reference Guide2000 httpwwwswarmorg
[20] S Tisue and U Wilensky ldquoNetLogo a simple environment formodeling complexityrdquo in International Conference on ComplexSystems pp 16ndash21 2004
[21] S Luke C Cioffi-Revilla L Panait K Sullivan and G BalanldquoMASON a multiagent simulation environmentrdquo Simulationvol 81 no 7 pp 517ndash527 2005
[22] M Ichikawa H Tanuma Y Koyama andH Deguchi ldquoSOARSintroduction as a social microscope for simulations of socialinteractions and gamingrdquo in Organizing and Learning throughGaming and Simulation Proceedings of the 38th Conference ofthe International Simulation and Gaming Association (ISAGArsquo07) pp 149ndash158 2007
[23] D S Burke J M Epstein D A T Cummings et al ldquoIndividual-based computational modeling of smallpox epidemic controlstrategiesrdquo Academic Emergency Medicine vol 13 no 11 pp1142ndash1149 2006
[24] F-Y Wang ldquoParallel control and management for intelligenttransportation systems concepts architectures and applica-tionsrdquo IEEE Transactions on Intelligent Transportation Systemsvol 11 no 3 pp 630ndash638 2010
[25] Z Molnar D Balasubramanian and A Ledeczi ldquoAn introduc-tion to the genericmodeling environmentrdquo in Proceedings of theTOOLS Europe 2007 Workshop on Model-Driven DevelopmentTool Implementers Forum Zurich Switzerland 2007
[26] S Kelly and J-P Tolvanen Domain Framework in Domain-Specific Modeling Enabling Full Code Generation JohnWiley ampSons Hoboken NJ USA 2007
[27] B P Zeigler H Praehofer and T G Kim Theory of Modelingand Simulation Integrating Discrete Event and ContinuousComplex Dynamic Systems Academic Press San Diego CalifUSA 2000
[28] G Booch J Rumbaugh and I JacobsonThe Unified ModelingLanguage User Guide Addison-Wesley Reading Mass USA1999
[29] F Wagner R Schmuki T Wagne and P Wolstenholme Mod-eling Software with Finite State Machines A Practical ApproachCRC amp Taylor amp Francis Boca Raton Fla USA 2006
[30] T Espiner ldquoChina builds worldrsquos fastest supercomputerrdquo ZDNetUK October 2010
[31] B Chen and G Guo ldquoA two-tier parallel architecture for arti-ficial society simulationrdquo in Proceedings of the ACMIEEESCS26th Workshop on Principles of Advanced and DistributedSimulation (PADS rsquo12) pp 184ndash186 July 2012
[32] G Guo B Chen X G Qiu and Z Li ldquoParallel simulation oflarge-scale artificial society on CPUGPU mixed architecturerdquo
Mathematical Problems in Engineering 27
in Proceedings of the ACMIEEESCS 26th Workshop on Prin-ciples of Advanced and Distributed Simulation (PADS rsquo12) pp174ndash177 July 2012
[33] S-X Lun ldquoResearch on the classification of parallel executionmodes of ACP theoryrdquo Acta Automatica Sinica vol 38 no 10pp 1602ndash1608 2012
[34] F-YWang ldquoA framework for social signal processing and anal-ysis from social sensing networks to computational dialecticalanalyticsrdquo Science China vol 43 no 12 pp 1598ndash1611 2013
[35] G GuoOne Model Userrsquos Manual version 12 National Univer-sity of Defense University 2011
[36] Y Ge R Meng Z Cao X Qiu and K Huang ldquoVirtual citymdashanindividual-based digital environment for human mobility andinteractive behaviorrdquo Simulation Transactions of the Society forthe Modeling and Simulation International 2014
[37] M Kretzschmar and R T Mikolajczyk ldquoContact profiles ineight European countries and implications for modelling thespread of airborne infectious diseasesrdquo PLoS ONE vol 4 no6 Article ID e5931 2009
[38] W Duan and X Qiu ldquoFostering artificial societies using sociallearning and social control in parallel emergency managementsystemsrdquo Frontiers in Computer Science vol 6 no 5 pp 604ndash610 2012
[39] W J Edmunds C J OrsquoCallaghan and D J Nokes ldquoWhomixes with whom Amethod to determine the contact patternsof adults that may lead to the spread of airborne infectionsrdquoProceedings of the Royal Society B Biological Sciences vol 264no 1384 pp 949ndash957 1997
[40] J Banks J S Carson B L Neison and D M Nicol Discrete-Event System Simulation Prentice Hall Englewood Cliffs NJUSA 4th edition 2007
[41] H Yongxia ldquoInvestigation and analysis of college studentsrsquo workand restrdquo Value Engineering vol 31 no 3 2012 (Chinese)
[42] W Duan Z Cao Y Wang et al ldquoAn ACP approach to publichealth emergency management using a campus outbreak ofh1n1 influenza as a case studyrdquo IEEE Transactions on SystemsMan and Cybernetics Part A Systems and Humans vol 43 no5 pp 1028ndash1041 2013
[43] HWHethcote ldquoThemathematics of infectious diseasesrdquo SIAMReview vol 42 no 4 pp 599ndash653 2000
[44] G Chowell E Shim F Brauer and et al ldquoModelling thetransmission dynamics of acute haemorrhagic conjunctivitisapplication to the 2003 outbreak in Mexicordquo Statistics inMedicine vol 25 no 11 pp 1840ndash1857 2006
[45] Y Ge L Liu B Chen X Qiu and K Huang ldquoAgent-basedmodeling for InfluenzaH1N1 in an artificial classroomrdquo SystemsEngineering Procedia vol 2 pp 94ndash104 2011
[46] A R Tuite A L Greer MWhelan et al ldquoEstimated epidemio-logic parameters andmorbidity associatedwith pandemicH1N1influenzardquo CMAJ vol 182 no 2 pp 131ndash136 2010
[47] L Brouwers B Cakici M Camitz A Tegnell and MBoman ldquoEconomic consequences to society of pandemic H1N1influenza 2009mdashpreliminary results for Swedenrdquo Euro Surveil-lance vol 14 no 37 pp 1ndash7 2009
[48] Y Ge L Liu X Qiu H Song Y Wang and K Huang ldquoAframework of multilayer social networks for communicationbehavior with agent-based modelingrdquo Simulation Transactionsof the Society for the Modeling and Simulation International vol89 no 7 pp 810ndash828 2013
[49] W Duan X Qiu Z Cao X Zheng and K Cui ldquoHeterogeneousand stochastic agent-based models for analyzing infectious
diseasesrsquo super spreadersrdquo IEEE Intelligent Systems vol 28 no4 pp 18ndash25 2013
[50] The Preparation and Emergency Response Plans to PandemicPublished by Ministry of Health Ministry of Health 2005(Chinese)
[51] R M Fujimoto Parallel and Distributed Simulation SystemsWiley Interscience 2000
[52] F Cicirelli A Giordano and L Nigro ldquoDistributed simulationof situated multi-agent systemsrdquo in Proceedings of the 15thIEEEACM International Symposium on Distributed Simulationand Real Time Applications (DS-RT rsquo11) pp 28ndash35 September2011
[53] B Chen and G Guo ldquoA two-tier parallel architecturefor artificial society simulationrdquo in Proceedings of the 26thACMIEEESCS Workshop on Principles of Advanced and Dis-tributed Simulation (PADS rsquo12) pp 184ndash186 July 2012
[54] G Guo B Chen and X Qiu ldquoParallel simulation of large-scale artificial society with GPU as coprocessorrdquo InternationalJournal of Modeling Simulation and Scientific Computing vol4 no 2 Article ID 1350005 2013
[55] B Chen L-B Zhang X-C Liu and H Vangheluwe ldquoActivity-based simulation using DEVS increasing performance by anactivity model in parallel DEVS simulationrdquo Journal of ZhejiangUniversity Science C vol 15 no 1 pp 13ndash30 2014
[56] X-L Wang P Yang Z-D Cao et al ldquoQuantitative evaluationon the effectiveness of prevention and control measures againstpandemic influenza A (H1N1) in Beijing 2009rdquoChinese Journalof Epidemiology vol 31 no 12 pp 1374ndash1378 2010 (Chinese)
[57] S Mei A D van de Vijver L Xuan Y Zhu and P M ASloot ldquoQuantitatively evaluating interventions in the influenzaA (H1N1) epidemic on China campus grounded on individual-based simulationsrdquo Procedia Computer Science vol 1 no 1 pp1675ndash1682 2010
[58] X Qiu Z Cao and Z Li Mid-Term Reports of China NationalScience Foundation 91024030 2014
Submit your manuscripts athttpwwwhindawicom
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
MathematicsJournal of
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Mathematical Problems in Engineering
Hindawi Publishing Corporationhttpwwwhindawicom
Differential EquationsInternational Journal of
Volume 2014
Applied MathematicsJournal of
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Probability and StatisticsHindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Journal of
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Mathematical PhysicsAdvances in
Complex AnalysisJournal of
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
OptimizationJournal of
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
CombinatoricsHindawi Publishing Corporationhttpwwwhindawicom Volume 2014
International Journal of
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Operations ResearchAdvances in
Journal of
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Function Spaces
Abstract and Applied AnalysisHindawi Publishing Corporationhttpwwwhindawicom Volume 2014
International Journal of Mathematics and Mathematical Sciences
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
The Scientific World JournalHindawi Publishing Corporation httpwwwhindawicom Volume 2014
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Algebra
Discrete Dynamics in Nature and Society
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Decision SciencesAdvances in
Discrete MathematicsJournal of
Hindawi Publishing Corporationhttpwwwhindawicom
Volume 2014 Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Stochastic AnalysisInternational Journal of
10 Mathematical Problems in Engineering
Open Source Data Registration Tool (OsdRT)
Emergent EventsConfiguration Tool (EECT)
Intervention MeasuresConfiguration Tool (IMCT)
∙ Web information collecting
∙ Deep mining in social media
∙ Web information denoising
∙ Information filtering
Data acquisition
Actual society
Open source informationin social media
Internet
∙ Basic element extracting∙ Individuals and organizationsextraction∙ Sentiment analyzing andopinion mining∙ Social network analyzing
Data extraction
∙ Domain knowledge description∙ Information fusion and collisiondetection∙ Domain knowledge construction∙ Social media informationstandardization
Data standardization
Knowledge
Figure 8 The compositions of OsdRT [34]
Moreover the most optimal plan is used to the responseof emergency in actual society According to the idea ofparallel control in [24] the feedback of actual society ispartly collected from Internet networks by OsdRT As shownin Figure 8 OsdRT is composed of three components dataacquisition data extraction and data standardization Dataacquisition collects mines and filters information fromsocial sensing networks Data extraction includes basicelement extracting individual and organization extractingsentiment analyzing and social networks analyzing Datastandardization specifies the useful knowledge and sent it toconfiguration tools in KD-ACP
By processing in OsdRT the knowledge about emergentevents and intervention measures are analyzed first andregistered in the EECT and IMCT The registration updatesthe settings of emergent events and intervention measuresThis loop composed of yellow arrows implements the parallelexecution in ACP approach The implementation of parallelcontrol and management provides a data-driven approachthat considers both the engineering and social complexityfor modelling analysis and decision making in emergencymanagement
4 Modelling Beijing City with KD-ACP
41 How to Build the Artificial Beijing According to themodelling of artificial society in KD-ACP discussed beforethe Beijing city is modeled as follows
To meet the requirements of emergency managementsthe basic elements of artificial society are extended Asshown in Figure 9 six elements are required to simulate thecity agents environments transportation activity schedulecommunication and agent activity
411 Modelling Artificial Beijing Figure 10 shows the mainGUI of GME for the modelling of artificial modelling inpublic health events Metamodels are listed in the left area ofFigure 10 the list provides basic syntax elements for domainexperts to model artificial society Domain experts build
models based on the knowledge of their own MeanwhileGME supports hierarchy for building large scale systemsThesyntax symbol listed in GUI can be extended in new tab bydouble clicking Take agent for example the model of agentcan be detailed by edition in another tab page of agent
As shown in the center of Figure 10 the models ofartificial Beijing consist of five parts models of agent andenvironment domain models of public health events inter-ventionmodels controllermodels and services Agentmodeldescribes individuals in society it is composed of basicpopulation information action social relationships activityschedule and disease related information Activity schedulerepresents individualrsquos physical actionmodel focusing on thedaily action of agents Environment model includes physicalentities such as buildings playground transportations andagents contained in environment Domain models of publichealth events are composed of the propagation model of dis-ease disease state transition model and so on Interventionmodels include the settings of vaccination isolation and soon The models mentioned before are all FSM models Themechanisms of these models will be detailed in next sectionsController models and services are the public service mod-ules they are implemented in the development in MDT
412 Modelling Agents and Environments Under the speci-fication of FSM agent and environment models are imple-mented in two parts the state space and state transitionsThe state space is composed of the demographic attributesand behavior related attributes The transitions are triggeredwhen the conditions of states are satisfied As shown inFigure 11(a) the action of agent is changed when the ldquonexttimerdquo condition is satisfied in agent model while the agentslist is changed when the agent arrival condition is satisfied inthe environment model
413 Modelling Activities Agent activities come from theagent state transitions of actions such as movements andcommunications The actions of agents are instructed by theactivity schedule shown in Table 2 Activity schedule lists
Mathematical Problems in Engineering 11
Transportation EnvironmentsEnvironments
Agents
Artificial Beijing
Agent activityActivity schedule
Agent activity
Communication by socialrelationship networks
Communication by socialrelationship networks
Figure 9 The basic elements of artificial society
Syntax elements of DSM
Services
Models of agentenvironment andactivity schedule
Domain models of publichealth events
Intervention models
Modelingaspect ofdomainexperts
Controllermodels
Figure 10 The models of artificial Beijing in GME
all the actions with probability in one day for agents inboth normal and emergent situation [38] There are severaltypes of activity schedule in artificial Beijing student agentactivity schedule worker agent activity schedule emergentagent schedule and so on For example Table 2 gives an agentactivity schedule Upon the instruction of activity schedulestudent agent changes the actions by 119901
119894after state transitions
The119901119894in the tablemeans the action probability in the relevant
period In the duration from 0800 to 1200 a student agenteither goes to classroom to have class or goes to libraryto study The probability of class action is 119901
2while the
probability of study action is 1 minus 1199012
Agent behaviors are decided by the settings of activityschedule In addition to the daily activity schedulementionedbefore emergent activity schedules are also considered inour work Take public health events for instance an infectedagent changes schedule from a normal one to emergentone The workflow of a susceptible agent is illustratedin Figure 12 to show the change of behaviors After theinfection the agent is set in incubation phase Not all theincubation agents will become symptomatic Some of themturn back to being susceptible and some of them becomesymptomatic The symptomatic agents change their activityschedule from normal to emergent In the emergent case
12 Mathematical Problems in Engineering
Next actionNext environment (location)
Advance to next actiontime
Find next environment inactivity chain by action
Current actionCurrent environment (location)
Q
q0
120575 Find next action inactivity schedule by time
F
Σ
(a) The state transitions of agent
Q
q0
120575 F
Current agent list
Current pollution level
Next agents list
Next pollution level
Agents entered
Infected agents entered
Add agents to currentagents list
Compute pollution levelby infected agents
Σ
(b) The state transitions of environment
Figure 11 The state transitions of models
Table 2 The activity schedule of a student agent
Duration (Δ119905) Activity Location Probability 119879Location (minute)0000ndash0600 Sleep Dormitory 119875
0(100) 360
0600ndash0800 Breakfast Dormitoryrestaurant 1198751(068) 120
Sports-breakfast-travel Playground-dormitoryrestaurant 1 minus 1198751(032)
0800ndash1200 Class Classroom 1198752(077) 240
Study Library 1 minus 1198752(023)
1200ndash1400 Lunch Restaurant 1198753(090) 120
Lunch-shopping Restaurant-convenience store 1 minus 1198753(010)
1400ndash1800 Class Classroom 1198754(070) 240
Study Library 1 minus 1198754(030)
1800-1900Dinner Dormitoryrestaurant 119875
5(063)
60Sports-shopping Playgroundconvenience store 1198756(025)
Study Classroomlibrary 1 minus (1198755+ 1198756) (012)
1900ndash2200Rest Dormitory 119875
7(045)
180Dinner-rest Dormitoryrestaurant-dormitory 1198758(020)
Study Classroomlibrary 1 minus (1198757+ 1198758) (035)
2200ndash2400 Rest Dormitory 1198759(072) 120
Sleep Dormitory 1 minus 1198759(018)
infected agents go to hospital according to the treatmentschedule or stay in dormitory according to isolation scheduleAfter the treatment in hospital or self-immunoprocess agentsbecome healthy and immune of disease If the agents aretreated in the hospital they are not allowed to get out untilrecovered Moreover activity schedules are also influencedby the emergency response plans For example in the case ofisolation in emergency response plans the activity scheduleof agents who had contacted with infected is changed Onlythe locations including home and dormitory compose thisemergent schedule Likewise based on the statistical datafrom emergency response plans the additional possibilities
are added into activity schedulesThe behaviors of agents alsochange with activity schedules In the view of ACP approachthe injection of new emergency response plans implementsthe parallel execution of microagent behaviors
In another aspect contact frequency is also anothercrucial element to determine the infected rate in public healthevent It is important to model the contact frequency andcontact time of individuals Based on the studies on thecontact behavior of human being by questionnaire surveyEdmunds found that the contact frequency of individualcould be fitted approximately into a normal distributionand the mean and the standard deviation distributions are
Mathematical Problems in Engineering 13
Susceptible agent ImmuneYes
Healthy and immune
No
Choose schedule
Self-isolation Go to hospital
Staysymptomatic
No
Receive treatment
Contact withinfected agents
Symptomatic
No
Self-immunoprocess
Yes
Recovered
Yes
Incubation
Infection
No
Yes No
NoYes
Figure 12 The workflow of a susceptible agent
168 and 85 [39] So it is possible to apply normal randomvariable to model the contact frequency In our work Box-Muller method [40] is used to generate the random variableof contact frequency shown in the following
119865119862= 120583119865+ 120590119865(minus2 ln (120574
1))12 cos (2120587120574
2) (1)
in which 119865119862is the random variable of contact frequency 120583
119865
is the mean value of normal distribution 120590119865is the standard
deviation of normal distribution and 1205741and 120574
2are the
uniform random variables distributed in the interval [0 1]Based on (1) and survey data [41] the contact frequenciesof agent 119865contact(119860 119894) are discretized as in Table 3 within theconsideration of activity differences
Similarly the duration per contact between individualsis another key factor of AHC transmission The duration ofcontact could also bemodeled by the normal randomvariablein the following [42]
119879119862= 120583119879+ 120590119879(minus2 ln (120574
1))12 cos (2120587120574
2) (2)
in which 119879119862is the random variable of duration per contact
120583119879
is the mean value of normal distribution 120590119879
is thestandard deviation of normal distribution and 120574
1and 120574
2
are the uniform random variables distributed in the interval
[0 1] In addition because the time spent in specific location119879Location(119860 119894) also follows the random distribution like (3)in our work In order to make the duration of contact119879contact(119860 119894) shorter than the 119879Location(119860 119894) the mean andstandard deviation of (2) is set up in (3) and (4) [42]
120583119879=119879Location(120583119865+ 120590119865) (3)
120590119879=119879Location(10120583119865+ 10120590
119865) (4)
in which 119879Location is the time of an activity in a specificlocation set in Table 2 120583
119879and 120590
119879are the mean and standard
deviation of contact frequency As a result the duration ofagent 119879contact(119860 119894) in specific locations (listed in Table 2) isalso discretized in Table 3 within the consideration of activitydifferences
414 Modelling Transportation In artificial Beijing threekinds of travel models are considered walking inside thedistrict travelling by road networks and travelling by subwaynetworks According to the traveling models the agentmovements are implemented by the compositions of travelingmodels before So it is important to build the basic road
14 Mathematical Problems in Engineering
Table 3 The contact frequency and duration in different locations and activities by social relationships in case of student and teacher
Social relationship Activity Location 119879contact(119860 119894) (minute) 119865contact(119860 119894)
Student-student
Sleep Dormitory 0 0
Breakfast Restaurant 119873(52) 119873(32)Breakfast Dormitory 119873(52) 119873(21)Sports Playground 119873(21) 119873(53)Class Classroom 119873(21) 119873(21)Study Library 119873(21) 119873(21)Lunch Restaurant 119873(158) 119873(32)
Shopping Convenience Store 119873(203) 119873(21)Dinner Restaurant 119873(108) 119873(21)Dinner Dormitory 119873(158) 119873(32)Rest Dormitory 119873(63) 119873(52)
Teacher-student Class Classroom 119873(21) 119873(21)
Subway network of Beijing city (red lines) Road network of Beijing city (green lines)
Figure 13 The implementation of transportation for artificial Beijing
and subway networks With the help of Google map thekey points of road are sampled first and the road networksare generated by the links among point Similarly subwaynetworks are generated by the links between subway stationsmapped from the real stations in Google map Figure 13shows the road and subway networks in artificial Beijingwith the road networks in green and the subway networks inred Most of the districts in Beijing city are covered by thesetwo networks The algorithm of hierarchical route planningis proposed as follows
(i) Find the next position where agent will be located innext action
(ii) Obtain the travelingmodels from the statistical trans-portation data
(iii) Search the path from the starting position to thenearest road entry or subway stations
(iv) Search the optimal path from the starting station orroad entry to the nearest station or road exit of targetposition
(v) Search the path from the target subway station or roadexit to the target position
(vi) Attach the subway train number or the bus numberincluding the transfer information to the path accord-ing to the timetable of subway or bus
(vii) Connect all the paths obtained before Generate thepath from the starting position to the target position
Based on the algorithm for each agent in artificial Beijingthe transportation is simulated during the computationalexperiments It is worth noting that the activities insidethe buses or train cars are considered because the agentcommunication during travelling is also necessary to besimulated
Mathematical Problems in Engineering 15
(a) The implementation of agent in MDT (b) The code framework of agent
Figure 14 The agent model implementation in MDT [35]
415 The Implementation of Models by MDT All the modelsare developed in the MDT Figure 14 shows the GUI of thetool MDT is used to implement the state space and statetransitions of FSM models The code framework is shownin Figure 14(b) Using the C++ inheritance features and thetechniques of dynamic link library (dll) the models areencapsulated in the dll componentsThe flexible compositionmechanism is designed to support the models evolution incomputational experiments
416 The Description of Artificial Beijing by ASE As men-tioned in Section 34 ASE describes the artificial Beijingin macroview The statistical information of Beijing city iscollected by ASE Table 4 lists part of the description ofartificial Beijing generated by ASE
42 How to Obtain the Data of Beijing City
421 Population and Geospatial Environment Database ofBeijing City Based on the models developed for ArtificialBeijing it is necessary to generate the initial data to do thecomputational experiments Because the individual level dataare not available it is necessary to construct an individual-based population database for both accurate computationalexperiment and determining optimal decisions
According to the state spaces of models discussed inSection 412 the geospatial and population database of Arti-ficial Beijing is designed as shown in Figure 15 The kernelpart of database consists of the tables such as agent list tablegeospatial environment list table household list table andagent distribution table Agent list table is an individual leveltable used to store the data to initialize the state space of agentmodel such as id and gender age Geospatial environmentlist table is also an individual level table which stores thedata of environment attributes such as id street id andtype Household list table and agent distribution table storethe statistical data for the data generation for agent andenvironment models With the help of the geospatial andpopulation database the data generating of artificial Beijingis proposed in the next section
422 Generating Geospatial and Social Environment Datafor Artificial Beijing by PGET The algorithm of generatinggeospatial and social environment data for artificial Beijingis used to quantify the spatial distribution of population andformalize geospatial behavior of each agent The algorithmshown in Figure 16 is capable of generating a syntheticartificial society which allowsmultiresolution statistical datasocial interactive behavior and multilayer social networksto be integrated together The synthetic population canrepresent individual agents in the form of households andhousehold members and the synthetic population is statis-tically equivalent to a real population For each householdcharacteristics such as address household size family typesand relationships are generated Each person is described bycharacteristics such as age gender social role and correlatedlocations The algorithm provides an effective methodologyto reconstruct the computing environment in high resolutionby using statistical data in low resolution leading to betterprediction and management of emergencies The details ofalgorithm are illustrated in [36]
As shown in Figure 17 the implementation of the algo-rithm is embedded inside the GEPTThe collective input datasuch as the data files of population distribution is collected inthe tool first then the algorithm is activated to generate thedatabase The generation lasts almost twelve hours one timewith the specified parameters settings When the generationis finished the data of artificial Beijing is obtained in thedatabase with 19610000 agents and 16000 environments
5 H1N1 Experiments in ArtificialBeijing with KD-ACP
With the support of the artificial Beijing modeled before thecomputational experiments and parallel execution are alsoimplemented by KD-ACP The experiments are designed tofind the most optimal emergency decision response plansEpidemic in city is a typical scenario in emergency manage-ment H1N1 epidemic in Beijing in 2009 is used to be the usedcase to test KD-ACP According to Section 34 it is necessarytomodel emergencymanagement first in order to support thecomputational experiments
16 Mathematical Problems in Engineering
Table 4 The description of artificial Beijing generated by ASE
Statistical features ofartificial Beijing Descriptions of statistical features Data sourcedefault values
Agent roles Baby child primary student middle school student college student worker retired
Demographic data file ofBeijing
Environment types Super market store restaurant park stadium hospital school cliniccommercial building dormitory building residential building
Demographic data file ofBeijing
Demographic data Age distribution of population sex distribution of population householddistribution of population retired age distribution children distribution
Demographic data file ofBeijing
Environment data Distribution of all types of environments Environment distributiondata file of Beijing
Relationships Classmate relative college friendship family Demographic data file ofBeijing
Varying parameters
Age difference of couples119873(120583120590) 119873(03) isin [0 12]Minimummarriage age (male female) (22 20)Age difference of children119873(120583120590) 119873(13) isin [1 10]Enrollment rate of kindergarten 120572 08Enrollment rate of primary school 120573 1Enrollment rate of middle school 120574 1Enrollment rate of college 120575 08Entrance age of kindergarten 1205721015840 3Entrance age of primary school 1205731015840 7Entrance age of middle school 1205741015840 13Entrance age of middle school 1205751015840 19
Age distributioninformation table
Agent settable
Geospatialenvironment set table
Household settable
Figure 15 The geospatial and population database of Beijing city
51 Modelling Emergency Management The modelling ofemergency management consists of two parts the modellingof emergent events and the modelling of intervention mea-sures H1N1 and the intervention measures for H1N1 aremodeled in our case
511 Modelling H1N1 by MDT H1N1 Model in agent simu-lates the states transitions of health status and the relevantactions Referring to SIR (SusceptibleInfectiveRecovered)[43 44] models agent has three health statuses susceptibleinfected and healthy with immunity Only susceptible agentcan be infected by the contact with infected agents After
the three infected phases (incubation being symptomaticand recovery) of H1N1 model agent is healthy again It isworthy mentioning that the recovered agents can be infectediteratively when the immunity disappears gradually by timeThe state transitions are illustrated in Figure 18
According to [45] the latent period of H1N1 was in aWeibull distribution within people from one to seven days[46 47] The distribution was usually centered in the rangebetween one and three days AWeibull random variable [40]is used to model the latent period as described in
119879Lat = 120572 times [minus ln (1 minus 120574)]1120573+ 120592 (119879Lat ge 120592) (5)
Mathematical Problems in Engineering 17
Censusdata
Generate GUID
Assign individualhousehold
Generate familystructure
Assign family roleto every member
Endow individualwith age gender
social role
Distribute environmentlocations on the map
School Workplace Entertainmentlocations
Consumptionlocations
Assign school orworkplace to
individual
Correlatedindividual withother locations
Generate friendshipin school
Generate coworkerrelations inworkplace
Assign individualadministrative
region
Generate housebuildings
Locate householdto house building
Generate familyrelation
Generateneighborhood
relation
PopulationEnvironment
Environment listSocial network
Figure 16 The algorithm of generating geospatial and social environment data [36]
Edit
Edit
Edit
Edit
Edit
Edit
BrowseStatistical characteristic of population
Geospatial population data
Population data of street
Population types
Geospatial and social environment generation tool
Workspaces
Parameters distribution settings
Household types
Roles in household
Household structure
Social roles
BrowseBrowseBrowseBrowseBrowseBrowse
Population distribution data
File path
Retired distributionPop distributionHousehold numHousehold distributionSex distributionAge distributionData type
Figure 17 The GUI of Population and Geospatial Environment generation Tool (PGET)
18 Mathematical Problems in Engineering
Infected
SusceptibleHealth
withimmunity
Infected
by contac
t
with in
fected
agents
Recovery
Increase before beingsymptomatic symptomatic
Decrease after being
Decrease by time
Infectivity
Immunity
Figure 18 The transitions of agent health status
119879Lat denotes the duration of latent period 120592 120572 and 120573 arethe location parameter scale parameter and shape parameterof Weibull distribution respectively 120574 is a uniform randomnumber in the range [0 1] According to statistics [47] 120592 120572and 120573 are set as 0 18 and 121 respectively Then the mean(standard deviation) of latent period is calculated as 159 days(058 day2) Furthermore infectious period is set as 7 days
A susceptible studentrsquos probability of getting infecteddepended on the infectivity of infectious agents his ownimmunity level the duration of the contact action and soon The infectivity of an infected agent evolved with timeduring the infectious period According to the statistics ofinfectivity in the chart with red bars in Figure 19 an infectedstudent had the highest level of infectivity in the second dayafter he had the first symptom The infectivity levels in otherdays in comparison with the second day are in Table 5 Theday labeled as ldquominus1rdquo means the day before the starting of thesymptomatic period
Based on the algorithm discussed before theH1N1modelis developed with the help of MDT The settings of influenzamodel will be discussed in Section 521
512 Modelling Intervention Measures by MDT In corre-spondence intervention measures model is used to ceasethe outbreak of H1N1 As discussed in [42 48 49] if theappropriate measures are taken when the infectious diseaseemerges the transmission of the disease could be sloweddown and the damage could be decreased Therefore it isnecessary and important to design emergency interventionmeasures in the artificial Beijing According to the MinistryofHealth inChina the intervention policies [50] are designedto control the spread of influenza such as H1N1 The inter-ventionmeasures of these policies including are interventionsactivated time vaccination rate antibiotic delay to hospitalisolation duration close workspace duration and limitationof activities listed in Table 6 Similarly the models of theseintervention measures are also developed with the help ofMDT Using the reasonable data ranges listed in Table 6intervention models can be initialized to do the computa-tional experiments These experiments are used to show how
Symptomatic1 2 3 4 5 6 7minus1
Recovered
Tim
e (da
y)
Infe
ctiv
ity
Figure 19 The distribution of infectivity rate of an infected agent[37]
Table 5 The proportion of infectivity
Days (119894) minus1 1 2 3 4 5 6 6Proportion of infectivity 03 05 1 08 04 02 01 005
to restrain the H1N1 transmissionThe settings of these inter-vention measures in IMCT are introduced in Section 522
52 Experiment Settings
521 H1N1 Settings by EECT EECT is designed to initializethe emergent events models such as H1N1The settings comefrom the specified emergent event models As shown inFigure 20 in theH1N1 used case the settings are composed ofthe distribution of infection source infection rate infectionperiod and so on The infection rate changes with envi-ronments and infection periods The settings are initializedfrom both the research of medicine on H1N1 and the data ofH1N1 outbreak in Beijing in 2009The process of influenza inartificial Beijing is simulated to repeat the possible scenariosof H1N1 break During the computational experiments ofartificial Beijing theH1N1 settings such as the period of beingsymptomatic are updated by the dynamic data registrationsfrom OsdRT The updating corrects the H1N1 model duringthe experiments With the help of OsdRT artificial Beijing isable to approach actual society It is also an implementationof parallel execution
522 Intervention Measures Settings by IMCT In the H1N1case the settings of intervention measures are shown inFigure 21 in IMCT These measures are designed under theinstruction of Emergency Decision Organizations like ChinaDisease Center (CDC) The measures such as vaccinationrate the rate of closedworkspace and the starting time can beinitialized in the tool Emergency response plans are imple-mented in these settings As illustrated in Figure 7 all the pos-sible compositions within value settings in reasonable inter-vals are tested to find the most optimal emergency responseplan For example different compositions of measures suchas vaccination rates and rates of workspace closed are usedto design the plans of experiments By the analysis of theresults the best composition of interventionmeasures can beobtainedThe analysis will be discussed in case study in detail
Mathematical Problems in Engineering 19
Table 6 The intervention measures of influenza
Intervention measures ValuesInterventions activated Time 119879Intervention isin [0 80th day]Vaccination rate 119877Vaccination isin [0 1]Antibiotic rate 119877Antibiotic isin [0 1]Delay to hospital 119867ospital(119860 119894) 119879infected lt 119905 lt 119879recovered + 119879HDelay 119879HDelay isin (0 3th day]Isolation duration 119868solation(119860 119894) 119879Intervention lt 119905 lt 119879Intervention + 119879IDuration 119879IDuration isin (0 14th day]Close workspace duration 119862lose(119864119894) 119879Intervention lt 119905 lt 119879Intervention + 119879CWDuration 119864119894 isinWorkspace 119879CWDuration isin (0 7th day]Limitation of agent activity duration 119871 imitation(119860 119894) 119879Intervention lt 119905 lt 119879Intervention + 119879LAADuration 119879LAADuration isin (0 7th day]119877Vaccination means the vaccination rate of artificial Beijing while 119877Antibiotic means the antibiotic rate used for infected agents in hospital119867ospital (119860119894) means that the infected agent 119860119894 is sent to hospital he or she will receive the retreatment immediately119868solation (119860119894) means that the infected agent 119860119894 is isolated at home he or she can only contact with the other agents until isolation duration is over119862lose (119864119894) means that the environment 119864119894 typed of workspace is closed until the close workspace duration is over119871 imitation (119860119894) means that the activities of agent 119860119894 are limited only the workspace and home are allowed until limitation duration is over
Incubation Weibulldistribution
Symbolic LogNormaldistribution
Recovering LogNormaldistribution
H1N1influenza
Settings of H1N1 influenza model
Disease model
Transmissionmodel
Figure 20 The settings of H1N1 in EECT
It is worth notifying that the settings of interventionmeasures are also influenced by OsdRT For example theH1N1 outbreak areas are notified by OsdRTThe interventionmeasures such as close workspace are used only in theoutbreak areas Therefore the unnecessary computation ofintervention measures is avoided So the data injection byOsdRT not only makes artificial society approach actualsociety but also increases the performance of computationalexperiments
53 Experiments by EMT When the initialization of arti-ficial Beijing is ready EMT is used to do the computa-tional experiments EMT is composed of two parts thecentral controller in the central node and the residen-tial service deployed in the computational nodes Centralcontroller loads the experiment plans and controls the whole
experiments process the GUI of controller is shown inFigure 22 EMT manages the computational experimentsin four steps cluster configuration load experiment plansmodels deployment and running experiments
531 Cluster Configuration Before the configuration clusterinformation is collected from the residential services insidethe cluster The information includes nodes amount nodenames node IPs and CPU occupations It is used to quantifythe computing power of the clusterThe cluster configurationis based on this information The nodes are selected forexperiments Then the LP (local process) [51] number foreach node is set according to the number of the CPU kernelsAfter the configuration the experiment plans are loaded tocustomize the models and data
20 Mathematical Problems in Engineering
Streets of close workspaces
Limitations ofactivities
IsolationsRespirator
Closeworkspace
Antibiotic
VaccinationHospital
Surveillance
Figure 21 The settings of intervention measures in IMCT
Information ofexperiment plan
List of experiment plans
Start | Pause | Reset | Stop | Speedup | Slowdown | Back
NodeName | CPU occupation | Memory occupation |Network | Models | Status
Figure 22 The deployment and running of experiments in EMT
532 Load Experiment Plan As discussed in Section 35 thedescription of experiment plans is composed of the infor-mation in three aspects models and the initial data for theexperiments the deployment mapping tables from the mod-els to the nodes and the plan of experiment executionWhenthe experiment plan is loaded by the EMT the file namesand model descriptions of models and data are listed Withthe support of cluster configuration models can be mappedto the nodes in the GUI according to the computationrequirements EMT also provides the setting of experimentsitself including the running times start time and end time
533 Models and Data Deployment Models are stored inthe agent population and services repository According tothe settings in experiment plan the customized models aredownloaded from the repository first then the models suchas agent model environment model emergent events modeland intervention measures model are integrated to be thecompositionmodels of artificial BeijingThemodels and datalocations in the nodes are configured in the ldquomodel pathrdquoand ldquodata pathrdquo in the GUI of central controller Finallythe models are uploaded to the nodes in cluster with theconsideration of partition The partition of artificial society
Mathematical Problems in Engineering 21
models is different from the multiagent system in [52]Models in artificial society are divided into three levels agentsand environments grids and city Agents and environmentsare encapsulated in grids which compose the whole cityThus the partition of artificial society becomes the partitionof grids The parallelism degrees are found to support thepartition of models agent independence and grid indepen-dence According to the parallelisms agents are partitioned toLPs according to the population distribution in grids Agentmovements among grids are simulated by the movementevents among grids in different LPs As a result a two-tierparallel architecture is proposed to support the simulation ofartificial society The architecture is detailed in [53]
534 Running of Experiments When the preparation isready the experiments can be done by the central controllerof EMT The running and control commands of the experi-ments are grouped into three aspects start pause or stop theexperiments get the running information of the nodes suchas the snapshot of the desktop of node With the help of thesecommands users can run the computational experimentsto do the research on the emergency management It isworth notifying that the simulation engine is optimized inorder to support the ten million agent simulation UsingGPU as coprocessor a two-tier parallel simulation engine isdesigned with support of MPI and OpenCL through phasedsynchronization [54] One-sided communication is used forreflection of remote simulation objects and message passingbetween processes A general kernel function prototype iselaborately designed and conditionally compiled for execu-tion on both CPU and GPU Moreover the optimizationoperations like load balance is developed under the instruc-tion of activity-based simulation [55] The densely populatedgrids are given more CPU and memory resource accordingto the activity predictions
54 Results of H1N1 Experiments with KD-ACP On thebasis of artificial society models and the intervention mea-sures mentioned in Section 512 a series of computationalexperiments are performed to study the H1N1 influenza inBeijing In our case of artificial Beijing 19610000 agentsare simulated in the cluster within 48 CPU cores and 128Gmemories the nodes are connected by kilomega networks Ittakes 18 hours to simulate a 250-day disease spreading
In order to illustrate how the optimal plans are obtainedbyKD-ACP the experiments are divided into four groups theexperiments ofmodel validation the experiments of sensitiv-ity analysis of vaccination rate the experiments of sensitivityanalysis of isolation and the experiments of the combinationof interventionmeasuresThese four groups show the generalprocess of the research on emergency management in H1N1influenza Firstly the models of H1N1 are validated withthe support of historical data Secondly the traditionalmedical interventions of pandemic such as vaccination areanalyzed to find the optimal plan Thirdly the traditionalnonmedical interventions of pandemic such as isolation areanalyzed to find the optimal plan Finally the combination ofinterventions is tested to show the combined effect
It is worth mentioning that all the experiments in ourwork are performed 100 times These experiments are ini-tialized with different settings of interventions in IMCT Theresults shown in the figures are the mean values
541 The Experiments of Model Validation In our artifi-cial Beijing the models of H1N1 influenza are built basedon the existing researches Wang and her colleagues hadstudied the H1N1 influenza by SEIR (SusceptibleExposedInfectiveRecovered) models [56] With the support of theirwork we build the models of artificial Beijing in the mannerof multiagents
In Figure 23 the control group shown in red line isdrawn based on the realistic statistical data collected in H1N1influenza in Beijing in 2009 The influenza lasts more thansix months and more than 170000 people are infectedComparing with the control group the simulated resultswithout interventions are drawn in blue It is obvious that thesimulated case fits the control group well before 70 days Butthe ascending of control group is slowed after 70 days andthe peak value (17883) is much less than the simulated results(25135) It is because the vaccination is activated at day 67 incontrol group according to the intervention measures usedin 2009 The restrain by vaccination becomes effective eightdays later The change of infected number after vaccinationis shown by red spotted line The trend of control group isdifferent from the simulation data after 75 days
However the simulated results (blue line) in compu-tational experiments fit the control group (as shown inFigure 23) in the first 70 days when no intervention mea-sures are executed The key features such as effective basicreproduction number (119877
0) are also almost the same in both
cases Therefore it can be concluded that the models ofartificial society built in microview are validated by therealistic statistical data in influenza in Beijing in 2009
In this group of experiments no interventions are exe-cuted in order to validate the models Therefore the simu-lated results of this extreme case which cannot happen todaycould be obtained As shown by blue line the number ofinfected agents without interventions reaches peak value atday 91 with 25135 consisting of 8052 in symptomatic phaseand 28950 in recovery phase It is obvious that the totalnumber of infected agents grows slowly at the beginningtime Then the number increases quickly when the influenzabreaks and it reaches its maximum value with 383870 in theend This phenomenon can be explained as in the followingAt the beginning of influenza due to the limited numberof infected agents the propagation of the disease remainsslow but many agents become susceptible When the numberof infected agents increases rapidly the spreading becomessignificant However the number of infected agents decreasesafter day 91 It is because the disease spreading relies on thesocial networks and spatial contact networks [49] It meansthat the infected agents can only infect agents either in thesocial relationship networks or in the spatially contacted
542 The Experiments of Sensitivity Analysis of VaccinationRate According to Section 512 vaccination is a typical
22 Mathematical Problems in Engineering
0
Infe
cted
agen
ts
0 50 100 150 200 250
Statistical dataSimulation data
25000
20000
15000
10000
5000
Day d
Figure 23 The results with no intervention executed and the com-parison with realistic statistical data in Beijing in 2009
medical intervention in influenza Traditionally the effective-ness of vaccination can only be evaluated after the influenzaand the key settings such as vaccination rate are determinedby the experience It is always difficult for emergency man-agement organizations to find the most optimal vaccinationrate in advanceTherefore the sensitivity of vaccination rate isanalyzed in this group of experiments The settings are listedin Table 7 average results are shown in Figure 24
Vaccination rates are set linearly (10 30 50 and70) to test the changes of influenza The peak value ofinfected agents decreases from 25135 22392 15787 6907to 638 respectively It shows that the more the agents arevaccinated the smaller the peak value of the infected agentswill be In the case of 70 vaccination the total infected agentis only 13799 It means that the influenza in this case does nothappen due to the high vaccination rate
The change of peak values also shows an interestingphenomenon Vaccination rates increase linearly the peakvalues decrease slower than the ascending of vaccination ratesat beginning but the descending becomes more and morefaster with the linear increase of vaccination rate It meansthat the small scale of vaccination (less than 30) does notmake senseWith the increase of vaccination rate the numberof susceptible agents decreases significantly In sequencethe possibility of infections is lowered greatly Additionallyit could be found that the decrease of peak values comeslater than the increase of the vaccination rates This is quitereasonable according to Mei et alrsquos work in [57]
As a result it is concluded that high vaccination rate iseffective in the control of H1N1 influenza though costly
543 The Experiments of Sensitivity Analysis of IsolationAccording to Section 512 isolation is a typical nonmedicalintervention It is the most used intervention measure in
Table 7 The initial settings of experiments of vaccination
Parameters ValueAgent count 19610000Initial infected count 40Interventions activated time(119879Intervention)
0
Vaccination rate (119877Vaccination) 0 10 30 50 70
0
Infe
cted
agen
ts
Loading intervention of vaccination
No vaccination10 vaccination30 vaccination
50 vaccination70 vaccination
0 50 100 150 200 250
25000
20000
15000
10000
5000
Day d
Figure 24 The results of loading intervention of vaccination
the emergency management in influenza As mentionedbefore a problem also exists that the measure cannot bevalidated before the intervention Therefore the sensitivityof isolation is analyzed in this group of experiments It isworth mentioning that the isolation is usually combinedwith the hospital policy It means the infected individual issent to hospital meanwhile the people who contact withthe infected are all isolated As a result parameters delay tohospital and isolation duration are designed together in thesettings of experiment Delay to hospital stands for the periodfrom infection time to hospital time while Isolation durationrepresents the period when an agent cannot contact others ifhe has contact with an infected agent The settings of theseparameters are listed in Table 8 average results are shown inFigure 25
Firstly the parameter delay to hospital is simulated aloneto test the sensitivity The simulated results are shown insolid line in Figure 25 The peak value decreases from 6708to 2218 when the delay to hospital decreases from 15 daysto 1 day The half day change brings the 67 reductionMoreover the peak value almost approximates to zero inthe case of 05 days The influenza does not happen in thiscase It can be concluded that the earlier the infected agents
Mathematical Problems in Engineering 23
Table 8 The initial settings of experiments of isolation
Parameters ValueAgent count 19610000Initial infected count 40Interventions activated time (119879Intervention) 80Delay to hospital (119879HDelay) isolation duration (119879IDuration) (15 0) (1 0) (05 0) (15 3) (1 3) (05 3)
0 50 100 150 200 2500
1000
2000
3000
4000
5000
6000
7000
Infe
cted
agen
ts
Loading intervention of isolation
Delay to hospital = 15
Delay to hospital = 10
Delay to hospital = 05
Delay to hospital = 15 isolation duration = 3
Delay to hospital = 10 isolation duration = 3
Delay to hospital = 05 isolation duration = 3
Day d
Figure 25 The results of loading intervention of isolation
are sent to hospital the less the agents will be infectedSo the construction of effective mechanism to find infectedindividuals is important in the emergency management ofinfluenza
Secondly the parameter isolation duration is added tosimulate the isolation of interventionObviously the isolationmeasure decreases the infected agents greatly Compared tothe parameter delay to hospital of 15-day case the peak valuedecreases from 6708 to 362when the isolation is added in theinterventionThe reduction of 6346 agents comes from the 3-day isolation of agents who had contacted with the infectedagents However the decrease of parameter delay to hospitalcannot bring the similar reduction in the 3-day isolationcases According to the figure the peak value decreases from362 to 51 from 15-day case to 05-day case only 311 agents aresaved from infection As a result the sensitivity of simulatedresults does not change obviously with delay to hospital inthe case of isolation It can be concluded that isolation is amore effective measure Though it is not easy for emergencymanagement organizations to decrease the delay to hospitaldue to the current monitoring mechanism the influenza canstill be controlled by isolation
544 The Experiments of Combination of Intervention Mea-sures Isolation is not the only nonmedical intervention ininfluenza close workspace is another common measureWhen close workspace is activated the workspaces are allclosed agents can only stay home at the closed time Luckilyisolation and close workspace are independent from oneanother It is possible for emergency management organi-zations to use these two measures together So this groupof experiments is designed to obtain the effectiveness of thecombination of intervention measures The parameters areinitialized with different settings listed in Table 9
Parameters delay to hospital and isolation duration is setin the middle values (1 day 3 days) according to the lastsection According to the experience close duration is set asseven days It is known that close workspace is a really costintervention measure the activated time is a key parameterWith the fixed parameters mentioned before the changeof interventions activated time is analyzed in this group ofcomputational experimentsThe average results are shown inFigure 26
Compared with the simulated results of experimentsdiscussed above Figure 26 strongly illustrates that influenzais ceased when the intervention measures are combinedtogether The peak value of infected agents is 141 in the ldquo80thdays of close-workspace and isolation with 1 day delay tohospitalrdquo case while 25135 in ldquono interventionrdquo case 2218 inldquo1 day delay to hospitalrdquo case 203 in ldquo1 day delay to hospitaland isolationrdquo case The interventions bring a great reductionof infection the peak values decrease from ten thousandslevel to hundred level
On the other hand the peak value even decreases to 108when the execution time of close workspace is 10 days earlierThe date of maximum value of infected agents is advancedfrom the 135th day to the 108th day The whole circle of H1N1influenza is reducedThe results of this group of experimentsshow that intervention measures should be activated as soonas possible and the compositions of interventions are moreeffective in emergency management of H1N1 influenza
In summary based on the general process lots of exper-iments could be done by KD-ACP to study emergency man-agement in H1N1 influenza Many conclusions are obtainedbased on the analysis of the experiments in this sectionAccording to these conclusions themost optimal response inour case study is the case of ldquo70th days of close-workspace andisolation with 1 day delay to hospitalrdquo It is because this casehas the fewest infected agents relatively But the modelling ofH1N1 influenza is not sufficient Economic and social factorsof interventions such as the money cost by vaccination andisolation the influence of the close-workspace in society
24 Mathematical Problems in Engineering
Table 9 The initial settings of experiments of isolation and closeworkspace
Parameters ValueAgent count 19610000Initial infected count 40Interventions activated time (119879Intervention) 70th day 80th dayDelay to hospital (119879HDelay) 1 dayIsolation duration (119879IDuration) 3 daysClose workspace duration (119879CWDuration) 7 days
50 100 150 200 2500
50
100
150
200
Infe
cted
agen
ts
Loading intervention of isolation and close workspace
Delay to hospital = 1 isolation duration = 3Start close time = 70 close duration time = 7Delay to hospital = 1 isolation duration = 3Start close time = 80 close duration time = 7
Day d
Figure 26The results of loading interventions of isolation and closeworkspace
and even opinion transmission of H1N1 in internet are notconsidered in our case study Without the analysis of thesefactors it does not make sense to talk about the most optimalintervention measures However KD-ACP had provided asoftware framework and general process of the research onemergency management Supported by KD-ACP the expertsof emergency management are able to study emergent eventssuch as H1N1 influenza as detailed as possible Based on theexperiments and analysis again and again the most optimalintervention measures will be found ultimately
55 Parallel Execution by ASST and OsdRT As discussed inSection 35 ASST andOsdRT are used to support the parallelexecution of artificial Beijing The statistical data is collectedand stored in artificial society runtime database at runtimeASST downloads the infection data from the database anddisplays it in the map As shown in Figure 27 the infectedagents are distributed in the districts in BeijingThe snapshotgives the geospatial distribution of the H1N1 epidemic Thelegends of the agents in different status are listed in the left
right corner in Figure 27 Five statuses are shown in the figurehealth without vaccination health with vaccination incuba-tion being symptomatic and convalescence ASST providesa very direct feeling for the emergency experts Partly withthe situation from ASST Emergency Decision Organizationmakes the emergency response plans to the response of theemergency scenario With the help of plans the epidemicin actual society can be restrained In the meantime theinfluenza information is reflected in the Internet Accordingto Figure 8 the data acquisition data extraction and datastandardization of H1N1 influenza are processed in OsdRTOsdRT is implemented by Dr Caorsquos team from Institute ofAutomation from Chinese Academy of Sciences [58] Thetool provides not only the reports of H1N1 SARS hand-foot-mouth disease and intestinal diseases are all monitored Theweb page is shown in Figure 28 H1N1 reports from Internetare collected first Based on the data extraction of the reportsH1N1 case of distribution is obtained and shown in theincidencemap As a result H1N1 incidence data is regarded asthe knowledge which is sent to EECT and IMCTThe settingsof emergent events and intervention measures are modifiedin order to make artificial society approach actual society innext turn of computational experiments With the help ofOsdRT the loop of parallel execution process integrates ACPapproach
6 Conclusion
This work provides an integrated software framework forsocial Computing in emergencymanagement From a systemperspective KD-ACP provides a reliable flexible low costand effective platform for the scientists to do the research onemergency response problems The analysis and predictionof these problems with inherent complexity can be solvedby the repetition of possible alternative experiments onthe software framework Therefore KD-ACP which is anattempt to implement ACP approach seemed as an effectivecomputational framework to support the decision making inemergency response
Currently KD-ACP is used to study the H1N1 epidemicin Beijing in 2009 The most optimal compositions of inter-vention measures are testified to support the response tonext influenza However it is still a long way ahead beforethe automation systematization and practicalization of KD-ACP A lot of work will be carried out along several directionsas in the following
(i) The parallel execution process in KD-ACP is actuallynot strongly connected with the emergency responseorganization Traditional experience and theory stillplay a dominant role in the decision of emergencymanagement It is necessary to propose a mechanismof hall for workshop to facilitate the parallel executionloop in actual society
(ii) Section 33 introduces the principle and process ofmodelling artificial society Metamodels and modelsof artificial Beijing are built by GME Model transfor-mation and code generation are used to implementcodes from FSM based models But many details
Mathematical Problems in Engineering 25
Fangshan district Daxing district Fengtai district Xicheng district
Mentougou district Haidian districtShijingshan district Dongcheng district
Incubation
Symptomatic
Convalescence
Health withoutvaccination
Health withvaccinationLegend of agent
health status
Figure 27 The situation of artificial Beijing by ASST
H1N1
Districts of Beijing
H1N1 case ofdistribution
Incidence map
Distribution ofH1N1 reportsfrom internet
Incidence of H1N1
Fangshan districtLiangxiang town
Population 19000Incidence 135789
(1100000)
Figure 28 The implementation of OsdRT by Institute of Automation from Chinese Academy of Sciences
are still not illustrated due to the topic of thispaper The key techniques such as template of codeframework and model transformations from specificdomains to FSM will be focused and introduced indetail in our next paper
(iii) The performance of computational experiments inKD-ACP is still slower than expectation The opti-mization of simulation engine and models structureis necessary to improve the performance So it ispossible to run artificial society with the real worldin parallel
In summary KD-ACP paves a new way for scientists inboth emergencymanagement and computation simulation to
collaborate with each other in solving the emergency man-agement problems by social computing
Research Highlights
(i) KD-ACP is implemented based on ACP approach
(ii) Artificial Beijing is built with the help of KD-ACP
(iii) H1N1 is modeled to simulate the influenza in Beijingin 2009
(iv) Computational experiments testify the effectivenessof intervention measures
26 Mathematical Problems in Engineering
Conflict of Interests
The authors declare that there is no conflict of interestsregarding the publication of this paper
Acknowledgment
This work was supported in part by the National ScienceFoundation of China (nos 91024030 71303252 61403402and 91324013)
References
[1] W Mao and F-Y Wang Advances in Intelligence and SecurityInformatics Academic Press Oxford UK 2012
[2] E Bonabeau ldquoAgent-based modeling methods and techniquesfor simulating human systemsrdquo Proceedings of the NationalAcademy of Sciences of the United States of America vol 99 no3 pp 7280ndash7287 2002
[3] K M Carley D B Fridsma E Casman et al ldquoBioWar scalableagent-basedmodel of bioattacksrdquo IEEETransactions on SystemsMan and Cybernetics A Systems and Humans vol 36 no 2 pp252ndash265 2006
[4] J M Epstein ldquoModelling to contain pandemicsrdquo Nature vol460 no 7256 p 687 2009
[5] Los Alamos National Laboratory Reports about EPISIMS LosAlamos National Laboratory 2003
[6] C H Builder and S C Bankes ldquoArtificial societies a conceptfor basic research on the societal impacts of informationtechnologyrdquo RAND Report 149 RAND Corp 1991
[7] F-Y Wang ldquoComputational theory and method on complexsystemrdquo China Basic Science vol 6 no 5 pp 3ndash10 2004
[8] F-Y Wang ldquoToward a paradigm shift in social computing theACP approachrdquo IEEE Intelligent Systems vol 22 no 5 pp 65ndash67 2007
[9] F-Y Wang and S Tang ldquoArtificial societies for integratedand sustainable development of metropolitan systemsrdquo IEEEIntelligent Systems vol 19 no 4 pp 82ndash87 2004
[10] D Wen Y Yuan and X-R Li ldquoArtificial societies computa-tional experiments and parallel systems an investigation ona computational theory for complex socioeconomic systemsrdquoIEEE Transactions on Services Computing vol 6 no 2 pp 177ndash185 2013
[11] F Wang F Zeng and Y Yuan ldquoAn ACP-based approach forcomplexity analysis of E-commerce systemrdquo Complex Systemsand Complexity Science vol 5 no 3 pp 1ndash8 2008 (Chinese)
[12] F-Y Wang and P K Wong ldquoResearch commentary intelligentsystems and technology for integrative and predictivemedicinean ACP approachrdquoACMTransactions on Intelligent Systems andTechnology vol 4 no 2 pp 1ndash6 2013
[13] J Sifeng X Gang F Dong and H Chunpeng ldquoStudy on theemergency rescue decision support system of petrochemicalplant based on ACP theoryrdquo in Proceedings of the 6th Manage-ment Annual Meeting (MAM rsquo11) 2011 (Chinese)
[14] X Dong D Fan G Xiong F Zhu Z Zhang and Y Yao ldquoPar-allel Bus Rapid Transit (BRT) operation management systembased on ACP approachrdquo in Proceedings of the 6th ManagementAnnual Meeting (MAM rsquo11) Beijing China 2011
[15] B Ning F-Y Wang H-R Dong R-M Li D Wen and L LildquoParallel systems for Urban rail transportation based on ACP
approachrdquo Journal of Transportation Systems Engineering andInformation Technology vol 10 no 6 pp 23ndash28 2010 (Chinese)
[16] F-Y Wang J Zhao and S-X Lun ldquoArtificial power systemsfor the operation and management of complex power gridsrdquoSouthern Power System Technology vol 2 no 3 pp 1ndash11 2008(Chinese)
[17] M J North T R Howe N T Collier and J R Vos ldquoRepastsimphony development environmentrdquo in Proceedings of theAgent 2005 Conference on Generative Social Processes Modelsand Mechanisms 2005
[18] M J North T RHowe N T Collier and J R Vos ldquoRepast Sim-phony runtime systemrdquo in Proceedings of the Agent Conferenceon Generative Social Processes Models and Mechanisms 2005
[19] SWARM Development Group Swarm 211 Reference Guide2000 httpwwwswarmorg
[20] S Tisue and U Wilensky ldquoNetLogo a simple environment formodeling complexityrdquo in International Conference on ComplexSystems pp 16ndash21 2004
[21] S Luke C Cioffi-Revilla L Panait K Sullivan and G BalanldquoMASON a multiagent simulation environmentrdquo Simulationvol 81 no 7 pp 517ndash527 2005
[22] M Ichikawa H Tanuma Y Koyama andH Deguchi ldquoSOARSintroduction as a social microscope for simulations of socialinteractions and gamingrdquo in Organizing and Learning throughGaming and Simulation Proceedings of the 38th Conference ofthe International Simulation and Gaming Association (ISAGArsquo07) pp 149ndash158 2007
[23] D S Burke J M Epstein D A T Cummings et al ldquoIndividual-based computational modeling of smallpox epidemic controlstrategiesrdquo Academic Emergency Medicine vol 13 no 11 pp1142ndash1149 2006
[24] F-Y Wang ldquoParallel control and management for intelligenttransportation systems concepts architectures and applica-tionsrdquo IEEE Transactions on Intelligent Transportation Systemsvol 11 no 3 pp 630ndash638 2010
[25] Z Molnar D Balasubramanian and A Ledeczi ldquoAn introduc-tion to the genericmodeling environmentrdquo in Proceedings of theTOOLS Europe 2007 Workshop on Model-Driven DevelopmentTool Implementers Forum Zurich Switzerland 2007
[26] S Kelly and J-P Tolvanen Domain Framework in Domain-Specific Modeling Enabling Full Code Generation JohnWiley ampSons Hoboken NJ USA 2007
[27] B P Zeigler H Praehofer and T G Kim Theory of Modelingand Simulation Integrating Discrete Event and ContinuousComplex Dynamic Systems Academic Press San Diego CalifUSA 2000
[28] G Booch J Rumbaugh and I JacobsonThe Unified ModelingLanguage User Guide Addison-Wesley Reading Mass USA1999
[29] F Wagner R Schmuki T Wagne and P Wolstenholme Mod-eling Software with Finite State Machines A Practical ApproachCRC amp Taylor amp Francis Boca Raton Fla USA 2006
[30] T Espiner ldquoChina builds worldrsquos fastest supercomputerrdquo ZDNetUK October 2010
[31] B Chen and G Guo ldquoA two-tier parallel architecture for arti-ficial society simulationrdquo in Proceedings of the ACMIEEESCS26th Workshop on Principles of Advanced and DistributedSimulation (PADS rsquo12) pp 184ndash186 July 2012
[32] G Guo B Chen X G Qiu and Z Li ldquoParallel simulation oflarge-scale artificial society on CPUGPU mixed architecturerdquo
Mathematical Problems in Engineering 27
in Proceedings of the ACMIEEESCS 26th Workshop on Prin-ciples of Advanced and Distributed Simulation (PADS rsquo12) pp174ndash177 July 2012
[33] S-X Lun ldquoResearch on the classification of parallel executionmodes of ACP theoryrdquo Acta Automatica Sinica vol 38 no 10pp 1602ndash1608 2012
[34] F-YWang ldquoA framework for social signal processing and anal-ysis from social sensing networks to computational dialecticalanalyticsrdquo Science China vol 43 no 12 pp 1598ndash1611 2013
[35] G GuoOne Model Userrsquos Manual version 12 National Univer-sity of Defense University 2011
[36] Y Ge R Meng Z Cao X Qiu and K Huang ldquoVirtual citymdashanindividual-based digital environment for human mobility andinteractive behaviorrdquo Simulation Transactions of the Society forthe Modeling and Simulation International 2014
[37] M Kretzschmar and R T Mikolajczyk ldquoContact profiles ineight European countries and implications for modelling thespread of airborne infectious diseasesrdquo PLoS ONE vol 4 no6 Article ID e5931 2009
[38] W Duan and X Qiu ldquoFostering artificial societies using sociallearning and social control in parallel emergency managementsystemsrdquo Frontiers in Computer Science vol 6 no 5 pp 604ndash610 2012
[39] W J Edmunds C J OrsquoCallaghan and D J Nokes ldquoWhomixes with whom Amethod to determine the contact patternsof adults that may lead to the spread of airborne infectionsrdquoProceedings of the Royal Society B Biological Sciences vol 264no 1384 pp 949ndash957 1997
[40] J Banks J S Carson B L Neison and D M Nicol Discrete-Event System Simulation Prentice Hall Englewood Cliffs NJUSA 4th edition 2007
[41] H Yongxia ldquoInvestigation and analysis of college studentsrsquo workand restrdquo Value Engineering vol 31 no 3 2012 (Chinese)
[42] W Duan Z Cao Y Wang et al ldquoAn ACP approach to publichealth emergency management using a campus outbreak ofh1n1 influenza as a case studyrdquo IEEE Transactions on SystemsMan and Cybernetics Part A Systems and Humans vol 43 no5 pp 1028ndash1041 2013
[43] HWHethcote ldquoThemathematics of infectious diseasesrdquo SIAMReview vol 42 no 4 pp 599ndash653 2000
[44] G Chowell E Shim F Brauer and et al ldquoModelling thetransmission dynamics of acute haemorrhagic conjunctivitisapplication to the 2003 outbreak in Mexicordquo Statistics inMedicine vol 25 no 11 pp 1840ndash1857 2006
[45] Y Ge L Liu B Chen X Qiu and K Huang ldquoAgent-basedmodeling for InfluenzaH1N1 in an artificial classroomrdquo SystemsEngineering Procedia vol 2 pp 94ndash104 2011
[46] A R Tuite A L Greer MWhelan et al ldquoEstimated epidemio-logic parameters andmorbidity associatedwith pandemicH1N1influenzardquo CMAJ vol 182 no 2 pp 131ndash136 2010
[47] L Brouwers B Cakici M Camitz A Tegnell and MBoman ldquoEconomic consequences to society of pandemic H1N1influenza 2009mdashpreliminary results for Swedenrdquo Euro Surveil-lance vol 14 no 37 pp 1ndash7 2009
[48] Y Ge L Liu X Qiu H Song Y Wang and K Huang ldquoAframework of multilayer social networks for communicationbehavior with agent-based modelingrdquo Simulation Transactionsof the Society for the Modeling and Simulation International vol89 no 7 pp 810ndash828 2013
[49] W Duan X Qiu Z Cao X Zheng and K Cui ldquoHeterogeneousand stochastic agent-based models for analyzing infectious
diseasesrsquo super spreadersrdquo IEEE Intelligent Systems vol 28 no4 pp 18ndash25 2013
[50] The Preparation and Emergency Response Plans to PandemicPublished by Ministry of Health Ministry of Health 2005(Chinese)
[51] R M Fujimoto Parallel and Distributed Simulation SystemsWiley Interscience 2000
[52] F Cicirelli A Giordano and L Nigro ldquoDistributed simulationof situated multi-agent systemsrdquo in Proceedings of the 15thIEEEACM International Symposium on Distributed Simulationand Real Time Applications (DS-RT rsquo11) pp 28ndash35 September2011
[53] B Chen and G Guo ldquoA two-tier parallel architecturefor artificial society simulationrdquo in Proceedings of the 26thACMIEEESCS Workshop on Principles of Advanced and Dis-tributed Simulation (PADS rsquo12) pp 184ndash186 July 2012
[54] G Guo B Chen and X Qiu ldquoParallel simulation of large-scale artificial society with GPU as coprocessorrdquo InternationalJournal of Modeling Simulation and Scientific Computing vol4 no 2 Article ID 1350005 2013
[55] B Chen L-B Zhang X-C Liu and H Vangheluwe ldquoActivity-based simulation using DEVS increasing performance by anactivity model in parallel DEVS simulationrdquo Journal of ZhejiangUniversity Science C vol 15 no 1 pp 13ndash30 2014
[56] X-L Wang P Yang Z-D Cao et al ldquoQuantitative evaluationon the effectiveness of prevention and control measures againstpandemic influenza A (H1N1) in Beijing 2009rdquoChinese Journalof Epidemiology vol 31 no 12 pp 1374ndash1378 2010 (Chinese)
[57] S Mei A D van de Vijver L Xuan Y Zhu and P M ASloot ldquoQuantitatively evaluating interventions in the influenzaA (H1N1) epidemic on China campus grounded on individual-based simulationsrdquo Procedia Computer Science vol 1 no 1 pp1675ndash1682 2010
[58] X Qiu Z Cao and Z Li Mid-Term Reports of China NationalScience Foundation 91024030 2014
Submit your manuscripts athttpwwwhindawicom
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
MathematicsJournal of
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Mathematical Problems in Engineering
Hindawi Publishing Corporationhttpwwwhindawicom
Differential EquationsInternational Journal of
Volume 2014
Applied MathematicsJournal of
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Probability and StatisticsHindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Journal of
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Mathematical PhysicsAdvances in
Complex AnalysisJournal of
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
OptimizationJournal of
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
CombinatoricsHindawi Publishing Corporationhttpwwwhindawicom Volume 2014
International Journal of
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Operations ResearchAdvances in
Journal of
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Function Spaces
Abstract and Applied AnalysisHindawi Publishing Corporationhttpwwwhindawicom Volume 2014
International Journal of Mathematics and Mathematical Sciences
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
The Scientific World JournalHindawi Publishing Corporation httpwwwhindawicom Volume 2014
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Algebra
Discrete Dynamics in Nature and Society
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Decision SciencesAdvances in
Discrete MathematicsJournal of
Hindawi Publishing Corporationhttpwwwhindawicom
Volume 2014 Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Stochastic AnalysisInternational Journal of
Mathematical Problems in Engineering 11
Transportation EnvironmentsEnvironments
Agents
Artificial Beijing
Agent activityActivity schedule
Agent activity
Communication by socialrelationship networks
Communication by socialrelationship networks
Figure 9 The basic elements of artificial society
Syntax elements of DSM
Services
Models of agentenvironment andactivity schedule
Domain models of publichealth events
Intervention models
Modelingaspect ofdomainexperts
Controllermodels
Figure 10 The models of artificial Beijing in GME
all the actions with probability in one day for agents inboth normal and emergent situation [38] There are severaltypes of activity schedule in artificial Beijing student agentactivity schedule worker agent activity schedule emergentagent schedule and so on For example Table 2 gives an agentactivity schedule Upon the instruction of activity schedulestudent agent changes the actions by 119901
119894after state transitions
The119901119894in the tablemeans the action probability in the relevant
period In the duration from 0800 to 1200 a student agenteither goes to classroom to have class or goes to libraryto study The probability of class action is 119901
2while the
probability of study action is 1 minus 1199012
Agent behaviors are decided by the settings of activityschedule In addition to the daily activity schedulementionedbefore emergent activity schedules are also considered inour work Take public health events for instance an infectedagent changes schedule from a normal one to emergentone The workflow of a susceptible agent is illustratedin Figure 12 to show the change of behaviors After theinfection the agent is set in incubation phase Not all theincubation agents will become symptomatic Some of themturn back to being susceptible and some of them becomesymptomatic The symptomatic agents change their activityschedule from normal to emergent In the emergent case
12 Mathematical Problems in Engineering
Next actionNext environment (location)
Advance to next actiontime
Find next environment inactivity chain by action
Current actionCurrent environment (location)
Q
q0
120575 Find next action inactivity schedule by time
F
Σ
(a) The state transitions of agent
Q
q0
120575 F
Current agent list
Current pollution level
Next agents list
Next pollution level
Agents entered
Infected agents entered
Add agents to currentagents list
Compute pollution levelby infected agents
Σ
(b) The state transitions of environment
Figure 11 The state transitions of models
Table 2 The activity schedule of a student agent
Duration (Δ119905) Activity Location Probability 119879Location (minute)0000ndash0600 Sleep Dormitory 119875
0(100) 360
0600ndash0800 Breakfast Dormitoryrestaurant 1198751(068) 120
Sports-breakfast-travel Playground-dormitoryrestaurant 1 minus 1198751(032)
0800ndash1200 Class Classroom 1198752(077) 240
Study Library 1 minus 1198752(023)
1200ndash1400 Lunch Restaurant 1198753(090) 120
Lunch-shopping Restaurant-convenience store 1 minus 1198753(010)
1400ndash1800 Class Classroom 1198754(070) 240
Study Library 1 minus 1198754(030)
1800-1900Dinner Dormitoryrestaurant 119875
5(063)
60Sports-shopping Playgroundconvenience store 1198756(025)
Study Classroomlibrary 1 minus (1198755+ 1198756) (012)
1900ndash2200Rest Dormitory 119875
7(045)
180Dinner-rest Dormitoryrestaurant-dormitory 1198758(020)
Study Classroomlibrary 1 minus (1198757+ 1198758) (035)
2200ndash2400 Rest Dormitory 1198759(072) 120
Sleep Dormitory 1 minus 1198759(018)
infected agents go to hospital according to the treatmentschedule or stay in dormitory according to isolation scheduleAfter the treatment in hospital or self-immunoprocess agentsbecome healthy and immune of disease If the agents aretreated in the hospital they are not allowed to get out untilrecovered Moreover activity schedules are also influencedby the emergency response plans For example in the case ofisolation in emergency response plans the activity scheduleof agents who had contacted with infected is changed Onlythe locations including home and dormitory compose thisemergent schedule Likewise based on the statistical datafrom emergency response plans the additional possibilities
are added into activity schedulesThe behaviors of agents alsochange with activity schedules In the view of ACP approachthe injection of new emergency response plans implementsthe parallel execution of microagent behaviors
In another aspect contact frequency is also anothercrucial element to determine the infected rate in public healthevent It is important to model the contact frequency andcontact time of individuals Based on the studies on thecontact behavior of human being by questionnaire surveyEdmunds found that the contact frequency of individualcould be fitted approximately into a normal distributionand the mean and the standard deviation distributions are
Mathematical Problems in Engineering 13
Susceptible agent ImmuneYes
Healthy and immune
No
Choose schedule
Self-isolation Go to hospital
Staysymptomatic
No
Receive treatment
Contact withinfected agents
Symptomatic
No
Self-immunoprocess
Yes
Recovered
Yes
Incubation
Infection
No
Yes No
NoYes
Figure 12 The workflow of a susceptible agent
168 and 85 [39] So it is possible to apply normal randomvariable to model the contact frequency In our work Box-Muller method [40] is used to generate the random variableof contact frequency shown in the following
119865119862= 120583119865+ 120590119865(minus2 ln (120574
1))12 cos (2120587120574
2) (1)
in which 119865119862is the random variable of contact frequency 120583
119865
is the mean value of normal distribution 120590119865is the standard
deviation of normal distribution and 1205741and 120574
2are the
uniform random variables distributed in the interval [0 1]Based on (1) and survey data [41] the contact frequenciesof agent 119865contact(119860 119894) are discretized as in Table 3 within theconsideration of activity differences
Similarly the duration per contact between individualsis another key factor of AHC transmission The duration ofcontact could also bemodeled by the normal randomvariablein the following [42]
119879119862= 120583119879+ 120590119879(minus2 ln (120574
1))12 cos (2120587120574
2) (2)
in which 119879119862is the random variable of duration per contact
120583119879
is the mean value of normal distribution 120590119879
is thestandard deviation of normal distribution and 120574
1and 120574
2
are the uniform random variables distributed in the interval
[0 1] In addition because the time spent in specific location119879Location(119860 119894) also follows the random distribution like (3)in our work In order to make the duration of contact119879contact(119860 119894) shorter than the 119879Location(119860 119894) the mean andstandard deviation of (2) is set up in (3) and (4) [42]
120583119879=119879Location(120583119865+ 120590119865) (3)
120590119879=119879Location(10120583119865+ 10120590
119865) (4)
in which 119879Location is the time of an activity in a specificlocation set in Table 2 120583
119879and 120590
119879are the mean and standard
deviation of contact frequency As a result the duration ofagent 119879contact(119860 119894) in specific locations (listed in Table 2) isalso discretized in Table 3 within the consideration of activitydifferences
414 Modelling Transportation In artificial Beijing threekinds of travel models are considered walking inside thedistrict travelling by road networks and travelling by subwaynetworks According to the traveling models the agentmovements are implemented by the compositions of travelingmodels before So it is important to build the basic road
14 Mathematical Problems in Engineering
Table 3 The contact frequency and duration in different locations and activities by social relationships in case of student and teacher
Social relationship Activity Location 119879contact(119860 119894) (minute) 119865contact(119860 119894)
Student-student
Sleep Dormitory 0 0
Breakfast Restaurant 119873(52) 119873(32)Breakfast Dormitory 119873(52) 119873(21)Sports Playground 119873(21) 119873(53)Class Classroom 119873(21) 119873(21)Study Library 119873(21) 119873(21)Lunch Restaurant 119873(158) 119873(32)
Shopping Convenience Store 119873(203) 119873(21)Dinner Restaurant 119873(108) 119873(21)Dinner Dormitory 119873(158) 119873(32)Rest Dormitory 119873(63) 119873(52)
Teacher-student Class Classroom 119873(21) 119873(21)
Subway network of Beijing city (red lines) Road network of Beijing city (green lines)
Figure 13 The implementation of transportation for artificial Beijing
and subway networks With the help of Google map thekey points of road are sampled first and the road networksare generated by the links among point Similarly subwaynetworks are generated by the links between subway stationsmapped from the real stations in Google map Figure 13shows the road and subway networks in artificial Beijingwith the road networks in green and the subway networks inred Most of the districts in Beijing city are covered by thesetwo networks The algorithm of hierarchical route planningis proposed as follows
(i) Find the next position where agent will be located innext action
(ii) Obtain the travelingmodels from the statistical trans-portation data
(iii) Search the path from the starting position to thenearest road entry or subway stations
(iv) Search the optimal path from the starting station orroad entry to the nearest station or road exit of targetposition
(v) Search the path from the target subway station or roadexit to the target position
(vi) Attach the subway train number or the bus numberincluding the transfer information to the path accord-ing to the timetable of subway or bus
(vii) Connect all the paths obtained before Generate thepath from the starting position to the target position
Based on the algorithm for each agent in artificial Beijingthe transportation is simulated during the computationalexperiments It is worth noting that the activities insidethe buses or train cars are considered because the agentcommunication during travelling is also necessary to besimulated
Mathematical Problems in Engineering 15
(a) The implementation of agent in MDT (b) The code framework of agent
Figure 14 The agent model implementation in MDT [35]
415 The Implementation of Models by MDT All the modelsare developed in the MDT Figure 14 shows the GUI of thetool MDT is used to implement the state space and statetransitions of FSM models The code framework is shownin Figure 14(b) Using the C++ inheritance features and thetechniques of dynamic link library (dll) the models areencapsulated in the dll componentsThe flexible compositionmechanism is designed to support the models evolution incomputational experiments
416 The Description of Artificial Beijing by ASE As men-tioned in Section 34 ASE describes the artificial Beijingin macroview The statistical information of Beijing city iscollected by ASE Table 4 lists part of the description ofartificial Beijing generated by ASE
42 How to Obtain the Data of Beijing City
421 Population and Geospatial Environment Database ofBeijing City Based on the models developed for ArtificialBeijing it is necessary to generate the initial data to do thecomputational experiments Because the individual level dataare not available it is necessary to construct an individual-based population database for both accurate computationalexperiment and determining optimal decisions
According to the state spaces of models discussed inSection 412 the geospatial and population database of Arti-ficial Beijing is designed as shown in Figure 15 The kernelpart of database consists of the tables such as agent list tablegeospatial environment list table household list table andagent distribution table Agent list table is an individual leveltable used to store the data to initialize the state space of agentmodel such as id and gender age Geospatial environmentlist table is also an individual level table which stores thedata of environment attributes such as id street id andtype Household list table and agent distribution table storethe statistical data for the data generation for agent andenvironment models With the help of the geospatial andpopulation database the data generating of artificial Beijingis proposed in the next section
422 Generating Geospatial and Social Environment Datafor Artificial Beijing by PGET The algorithm of generatinggeospatial and social environment data for artificial Beijingis used to quantify the spatial distribution of population andformalize geospatial behavior of each agent The algorithmshown in Figure 16 is capable of generating a syntheticartificial society which allowsmultiresolution statistical datasocial interactive behavior and multilayer social networksto be integrated together The synthetic population canrepresent individual agents in the form of households andhousehold members and the synthetic population is statis-tically equivalent to a real population For each householdcharacteristics such as address household size family typesand relationships are generated Each person is described bycharacteristics such as age gender social role and correlatedlocations The algorithm provides an effective methodologyto reconstruct the computing environment in high resolutionby using statistical data in low resolution leading to betterprediction and management of emergencies The details ofalgorithm are illustrated in [36]
As shown in Figure 17 the implementation of the algo-rithm is embedded inside the GEPTThe collective input datasuch as the data files of population distribution is collected inthe tool first then the algorithm is activated to generate thedatabase The generation lasts almost twelve hours one timewith the specified parameters settings When the generationis finished the data of artificial Beijing is obtained in thedatabase with 19610000 agents and 16000 environments
5 H1N1 Experiments in ArtificialBeijing with KD-ACP
With the support of the artificial Beijing modeled before thecomputational experiments and parallel execution are alsoimplemented by KD-ACP The experiments are designed tofind the most optimal emergency decision response plansEpidemic in city is a typical scenario in emergency manage-ment H1N1 epidemic in Beijing in 2009 is used to be the usedcase to test KD-ACP According to Section 34 it is necessarytomodel emergencymanagement first in order to support thecomputational experiments
16 Mathematical Problems in Engineering
Table 4 The description of artificial Beijing generated by ASE
Statistical features ofartificial Beijing Descriptions of statistical features Data sourcedefault values
Agent roles Baby child primary student middle school student college student worker retired
Demographic data file ofBeijing
Environment types Super market store restaurant park stadium hospital school cliniccommercial building dormitory building residential building
Demographic data file ofBeijing
Demographic data Age distribution of population sex distribution of population householddistribution of population retired age distribution children distribution
Demographic data file ofBeijing
Environment data Distribution of all types of environments Environment distributiondata file of Beijing
Relationships Classmate relative college friendship family Demographic data file ofBeijing
Varying parameters
Age difference of couples119873(120583120590) 119873(03) isin [0 12]Minimummarriage age (male female) (22 20)Age difference of children119873(120583120590) 119873(13) isin [1 10]Enrollment rate of kindergarten 120572 08Enrollment rate of primary school 120573 1Enrollment rate of middle school 120574 1Enrollment rate of college 120575 08Entrance age of kindergarten 1205721015840 3Entrance age of primary school 1205731015840 7Entrance age of middle school 1205741015840 13Entrance age of middle school 1205751015840 19
Age distributioninformation table
Agent settable
Geospatialenvironment set table
Household settable
Figure 15 The geospatial and population database of Beijing city
51 Modelling Emergency Management The modelling ofemergency management consists of two parts the modellingof emergent events and the modelling of intervention mea-sures H1N1 and the intervention measures for H1N1 aremodeled in our case
511 Modelling H1N1 by MDT H1N1 Model in agent simu-lates the states transitions of health status and the relevantactions Referring to SIR (SusceptibleInfectiveRecovered)[43 44] models agent has three health statuses susceptibleinfected and healthy with immunity Only susceptible agentcan be infected by the contact with infected agents After
the three infected phases (incubation being symptomaticand recovery) of H1N1 model agent is healthy again It isworthy mentioning that the recovered agents can be infectediteratively when the immunity disappears gradually by timeThe state transitions are illustrated in Figure 18
According to [45] the latent period of H1N1 was in aWeibull distribution within people from one to seven days[46 47] The distribution was usually centered in the rangebetween one and three days AWeibull random variable [40]is used to model the latent period as described in
119879Lat = 120572 times [minus ln (1 minus 120574)]1120573+ 120592 (119879Lat ge 120592) (5)
Mathematical Problems in Engineering 17
Censusdata
Generate GUID
Assign individualhousehold
Generate familystructure
Assign family roleto every member
Endow individualwith age gender
social role
Distribute environmentlocations on the map
School Workplace Entertainmentlocations
Consumptionlocations
Assign school orworkplace to
individual
Correlatedindividual withother locations
Generate friendshipin school
Generate coworkerrelations inworkplace
Assign individualadministrative
region
Generate housebuildings
Locate householdto house building
Generate familyrelation
Generateneighborhood
relation
PopulationEnvironment
Environment listSocial network
Figure 16 The algorithm of generating geospatial and social environment data [36]
Edit
Edit
Edit
Edit
Edit
Edit
BrowseStatistical characteristic of population
Geospatial population data
Population data of street
Population types
Geospatial and social environment generation tool
Workspaces
Parameters distribution settings
Household types
Roles in household
Household structure
Social roles
BrowseBrowseBrowseBrowseBrowseBrowse
Population distribution data
File path
Retired distributionPop distributionHousehold numHousehold distributionSex distributionAge distributionData type
Figure 17 The GUI of Population and Geospatial Environment generation Tool (PGET)
18 Mathematical Problems in Engineering
Infected
SusceptibleHealth
withimmunity
Infected
by contac
t
with in
fected
agents
Recovery
Increase before beingsymptomatic symptomatic
Decrease after being
Decrease by time
Infectivity
Immunity
Figure 18 The transitions of agent health status
119879Lat denotes the duration of latent period 120592 120572 and 120573 arethe location parameter scale parameter and shape parameterof Weibull distribution respectively 120574 is a uniform randomnumber in the range [0 1] According to statistics [47] 120592 120572and 120573 are set as 0 18 and 121 respectively Then the mean(standard deviation) of latent period is calculated as 159 days(058 day2) Furthermore infectious period is set as 7 days
A susceptible studentrsquos probability of getting infecteddepended on the infectivity of infectious agents his ownimmunity level the duration of the contact action and soon The infectivity of an infected agent evolved with timeduring the infectious period According to the statistics ofinfectivity in the chart with red bars in Figure 19 an infectedstudent had the highest level of infectivity in the second dayafter he had the first symptom The infectivity levels in otherdays in comparison with the second day are in Table 5 Theday labeled as ldquominus1rdquo means the day before the starting of thesymptomatic period
Based on the algorithm discussed before theH1N1modelis developed with the help of MDT The settings of influenzamodel will be discussed in Section 521
512 Modelling Intervention Measures by MDT In corre-spondence intervention measures model is used to ceasethe outbreak of H1N1 As discussed in [42 48 49] if theappropriate measures are taken when the infectious diseaseemerges the transmission of the disease could be sloweddown and the damage could be decreased Therefore it isnecessary and important to design emergency interventionmeasures in the artificial Beijing According to the MinistryofHealth inChina the intervention policies [50] are designedto control the spread of influenza such as H1N1 The inter-ventionmeasures of these policies including are interventionsactivated time vaccination rate antibiotic delay to hospitalisolation duration close workspace duration and limitationof activities listed in Table 6 Similarly the models of theseintervention measures are also developed with the help ofMDT Using the reasonable data ranges listed in Table 6intervention models can be initialized to do the computa-tional experiments These experiments are used to show how
Symptomatic1 2 3 4 5 6 7minus1
Recovered
Tim
e (da
y)
Infe
ctiv
ity
Figure 19 The distribution of infectivity rate of an infected agent[37]
Table 5 The proportion of infectivity
Days (119894) minus1 1 2 3 4 5 6 6Proportion of infectivity 03 05 1 08 04 02 01 005
to restrain the H1N1 transmissionThe settings of these inter-vention measures in IMCT are introduced in Section 522
52 Experiment Settings
521 H1N1 Settings by EECT EECT is designed to initializethe emergent events models such as H1N1The settings comefrom the specified emergent event models As shown inFigure 20 in theH1N1 used case the settings are composed ofthe distribution of infection source infection rate infectionperiod and so on The infection rate changes with envi-ronments and infection periods The settings are initializedfrom both the research of medicine on H1N1 and the data ofH1N1 outbreak in Beijing in 2009The process of influenza inartificial Beijing is simulated to repeat the possible scenariosof H1N1 break During the computational experiments ofartificial Beijing theH1N1 settings such as the period of beingsymptomatic are updated by the dynamic data registrationsfrom OsdRT The updating corrects the H1N1 model duringthe experiments With the help of OsdRT artificial Beijing isable to approach actual society It is also an implementationof parallel execution
522 Intervention Measures Settings by IMCT In the H1N1case the settings of intervention measures are shown inFigure 21 in IMCT These measures are designed under theinstruction of Emergency Decision Organizations like ChinaDisease Center (CDC) The measures such as vaccinationrate the rate of closedworkspace and the starting time can beinitialized in the tool Emergency response plans are imple-mented in these settings As illustrated in Figure 7 all the pos-sible compositions within value settings in reasonable inter-vals are tested to find the most optimal emergency responseplan For example different compositions of measures suchas vaccination rates and rates of workspace closed are usedto design the plans of experiments By the analysis of theresults the best composition of interventionmeasures can beobtainedThe analysis will be discussed in case study in detail
Mathematical Problems in Engineering 19
Table 6 The intervention measures of influenza
Intervention measures ValuesInterventions activated Time 119879Intervention isin [0 80th day]Vaccination rate 119877Vaccination isin [0 1]Antibiotic rate 119877Antibiotic isin [0 1]Delay to hospital 119867ospital(119860 119894) 119879infected lt 119905 lt 119879recovered + 119879HDelay 119879HDelay isin (0 3th day]Isolation duration 119868solation(119860 119894) 119879Intervention lt 119905 lt 119879Intervention + 119879IDuration 119879IDuration isin (0 14th day]Close workspace duration 119862lose(119864119894) 119879Intervention lt 119905 lt 119879Intervention + 119879CWDuration 119864119894 isinWorkspace 119879CWDuration isin (0 7th day]Limitation of agent activity duration 119871 imitation(119860 119894) 119879Intervention lt 119905 lt 119879Intervention + 119879LAADuration 119879LAADuration isin (0 7th day]119877Vaccination means the vaccination rate of artificial Beijing while 119877Antibiotic means the antibiotic rate used for infected agents in hospital119867ospital (119860119894) means that the infected agent 119860119894 is sent to hospital he or she will receive the retreatment immediately119868solation (119860119894) means that the infected agent 119860119894 is isolated at home he or she can only contact with the other agents until isolation duration is over119862lose (119864119894) means that the environment 119864119894 typed of workspace is closed until the close workspace duration is over119871 imitation (119860119894) means that the activities of agent 119860119894 are limited only the workspace and home are allowed until limitation duration is over
Incubation Weibulldistribution
Symbolic LogNormaldistribution
Recovering LogNormaldistribution
H1N1influenza
Settings of H1N1 influenza model
Disease model
Transmissionmodel
Figure 20 The settings of H1N1 in EECT
It is worth notifying that the settings of interventionmeasures are also influenced by OsdRT For example theH1N1 outbreak areas are notified by OsdRTThe interventionmeasures such as close workspace are used only in theoutbreak areas Therefore the unnecessary computation ofintervention measures is avoided So the data injection byOsdRT not only makes artificial society approach actualsociety but also increases the performance of computationalexperiments
53 Experiments by EMT When the initialization of arti-ficial Beijing is ready EMT is used to do the computa-tional experiments EMT is composed of two parts thecentral controller in the central node and the residen-tial service deployed in the computational nodes Centralcontroller loads the experiment plans and controls the whole
experiments process the GUI of controller is shown inFigure 22 EMT manages the computational experimentsin four steps cluster configuration load experiment plansmodels deployment and running experiments
531 Cluster Configuration Before the configuration clusterinformation is collected from the residential services insidethe cluster The information includes nodes amount nodenames node IPs and CPU occupations It is used to quantifythe computing power of the clusterThe cluster configurationis based on this information The nodes are selected forexperiments Then the LP (local process) [51] number foreach node is set according to the number of the CPU kernelsAfter the configuration the experiment plans are loaded tocustomize the models and data
20 Mathematical Problems in Engineering
Streets of close workspaces
Limitations ofactivities
IsolationsRespirator
Closeworkspace
Antibiotic
VaccinationHospital
Surveillance
Figure 21 The settings of intervention measures in IMCT
Information ofexperiment plan
List of experiment plans
Start | Pause | Reset | Stop | Speedup | Slowdown | Back
NodeName | CPU occupation | Memory occupation |Network | Models | Status
Figure 22 The deployment and running of experiments in EMT
532 Load Experiment Plan As discussed in Section 35 thedescription of experiment plans is composed of the infor-mation in three aspects models and the initial data for theexperiments the deployment mapping tables from the mod-els to the nodes and the plan of experiment executionWhenthe experiment plan is loaded by the EMT the file namesand model descriptions of models and data are listed Withthe support of cluster configuration models can be mappedto the nodes in the GUI according to the computationrequirements EMT also provides the setting of experimentsitself including the running times start time and end time
533 Models and Data Deployment Models are stored inthe agent population and services repository According tothe settings in experiment plan the customized models aredownloaded from the repository first then the models suchas agent model environment model emergent events modeland intervention measures model are integrated to be thecompositionmodels of artificial BeijingThemodels and datalocations in the nodes are configured in the ldquomodel pathrdquoand ldquodata pathrdquo in the GUI of central controller Finallythe models are uploaded to the nodes in cluster with theconsideration of partition The partition of artificial society
Mathematical Problems in Engineering 21
models is different from the multiagent system in [52]Models in artificial society are divided into three levels agentsand environments grids and city Agents and environmentsare encapsulated in grids which compose the whole cityThus the partition of artificial society becomes the partitionof grids The parallelism degrees are found to support thepartition of models agent independence and grid indepen-dence According to the parallelisms agents are partitioned toLPs according to the population distribution in grids Agentmovements among grids are simulated by the movementevents among grids in different LPs As a result a two-tierparallel architecture is proposed to support the simulation ofartificial society The architecture is detailed in [53]
534 Running of Experiments When the preparation isready the experiments can be done by the central controllerof EMT The running and control commands of the experi-ments are grouped into three aspects start pause or stop theexperiments get the running information of the nodes suchas the snapshot of the desktop of node With the help of thesecommands users can run the computational experimentsto do the research on the emergency management It isworth notifying that the simulation engine is optimized inorder to support the ten million agent simulation UsingGPU as coprocessor a two-tier parallel simulation engine isdesigned with support of MPI and OpenCL through phasedsynchronization [54] One-sided communication is used forreflection of remote simulation objects and message passingbetween processes A general kernel function prototype iselaborately designed and conditionally compiled for execu-tion on both CPU and GPU Moreover the optimizationoperations like load balance is developed under the instruc-tion of activity-based simulation [55] The densely populatedgrids are given more CPU and memory resource accordingto the activity predictions
54 Results of H1N1 Experiments with KD-ACP On thebasis of artificial society models and the intervention mea-sures mentioned in Section 512 a series of computationalexperiments are performed to study the H1N1 influenza inBeijing In our case of artificial Beijing 19610000 agentsare simulated in the cluster within 48 CPU cores and 128Gmemories the nodes are connected by kilomega networks Ittakes 18 hours to simulate a 250-day disease spreading
In order to illustrate how the optimal plans are obtainedbyKD-ACP the experiments are divided into four groups theexperiments ofmodel validation the experiments of sensitiv-ity analysis of vaccination rate the experiments of sensitivityanalysis of isolation and the experiments of the combinationof interventionmeasuresThese four groups show the generalprocess of the research on emergency management in H1N1influenza Firstly the models of H1N1 are validated withthe support of historical data Secondly the traditionalmedical interventions of pandemic such as vaccination areanalyzed to find the optimal plan Thirdly the traditionalnonmedical interventions of pandemic such as isolation areanalyzed to find the optimal plan Finally the combination ofinterventions is tested to show the combined effect
It is worth mentioning that all the experiments in ourwork are performed 100 times These experiments are ini-tialized with different settings of interventions in IMCT Theresults shown in the figures are the mean values
541 The Experiments of Model Validation In our artifi-cial Beijing the models of H1N1 influenza are built basedon the existing researches Wang and her colleagues hadstudied the H1N1 influenza by SEIR (SusceptibleExposedInfectiveRecovered) models [56] With the support of theirwork we build the models of artificial Beijing in the mannerof multiagents
In Figure 23 the control group shown in red line isdrawn based on the realistic statistical data collected in H1N1influenza in Beijing in 2009 The influenza lasts more thansix months and more than 170000 people are infectedComparing with the control group the simulated resultswithout interventions are drawn in blue It is obvious that thesimulated case fits the control group well before 70 days Butthe ascending of control group is slowed after 70 days andthe peak value (17883) is much less than the simulated results(25135) It is because the vaccination is activated at day 67 incontrol group according to the intervention measures usedin 2009 The restrain by vaccination becomes effective eightdays later The change of infected number after vaccinationis shown by red spotted line The trend of control group isdifferent from the simulation data after 75 days
However the simulated results (blue line) in compu-tational experiments fit the control group (as shown inFigure 23) in the first 70 days when no intervention mea-sures are executed The key features such as effective basicreproduction number (119877
0) are also almost the same in both
cases Therefore it can be concluded that the models ofartificial society built in microview are validated by therealistic statistical data in influenza in Beijing in 2009
In this group of experiments no interventions are exe-cuted in order to validate the models Therefore the simu-lated results of this extreme case which cannot happen todaycould be obtained As shown by blue line the number ofinfected agents without interventions reaches peak value atday 91 with 25135 consisting of 8052 in symptomatic phaseand 28950 in recovery phase It is obvious that the totalnumber of infected agents grows slowly at the beginningtime Then the number increases quickly when the influenzabreaks and it reaches its maximum value with 383870 in theend This phenomenon can be explained as in the followingAt the beginning of influenza due to the limited numberof infected agents the propagation of the disease remainsslow but many agents become susceptible When the numberof infected agents increases rapidly the spreading becomessignificant However the number of infected agents decreasesafter day 91 It is because the disease spreading relies on thesocial networks and spatial contact networks [49] It meansthat the infected agents can only infect agents either in thesocial relationship networks or in the spatially contacted
542 The Experiments of Sensitivity Analysis of VaccinationRate According to Section 512 vaccination is a typical
22 Mathematical Problems in Engineering
0
Infe
cted
agen
ts
0 50 100 150 200 250
Statistical dataSimulation data
25000
20000
15000
10000
5000
Day d
Figure 23 The results with no intervention executed and the com-parison with realistic statistical data in Beijing in 2009
medical intervention in influenza Traditionally the effective-ness of vaccination can only be evaluated after the influenzaand the key settings such as vaccination rate are determinedby the experience It is always difficult for emergency man-agement organizations to find the most optimal vaccinationrate in advanceTherefore the sensitivity of vaccination rate isanalyzed in this group of experiments The settings are listedin Table 7 average results are shown in Figure 24
Vaccination rates are set linearly (10 30 50 and70) to test the changes of influenza The peak value ofinfected agents decreases from 25135 22392 15787 6907to 638 respectively It shows that the more the agents arevaccinated the smaller the peak value of the infected agentswill be In the case of 70 vaccination the total infected agentis only 13799 It means that the influenza in this case does nothappen due to the high vaccination rate
The change of peak values also shows an interestingphenomenon Vaccination rates increase linearly the peakvalues decrease slower than the ascending of vaccination ratesat beginning but the descending becomes more and morefaster with the linear increase of vaccination rate It meansthat the small scale of vaccination (less than 30) does notmake senseWith the increase of vaccination rate the numberof susceptible agents decreases significantly In sequencethe possibility of infections is lowered greatly Additionallyit could be found that the decrease of peak values comeslater than the increase of the vaccination rates This is quitereasonable according to Mei et alrsquos work in [57]
As a result it is concluded that high vaccination rate iseffective in the control of H1N1 influenza though costly
543 The Experiments of Sensitivity Analysis of IsolationAccording to Section 512 isolation is a typical nonmedicalintervention It is the most used intervention measure in
Table 7 The initial settings of experiments of vaccination
Parameters ValueAgent count 19610000Initial infected count 40Interventions activated time(119879Intervention)
0
Vaccination rate (119877Vaccination) 0 10 30 50 70
0
Infe
cted
agen
ts
Loading intervention of vaccination
No vaccination10 vaccination30 vaccination
50 vaccination70 vaccination
0 50 100 150 200 250
25000
20000
15000
10000
5000
Day d
Figure 24 The results of loading intervention of vaccination
the emergency management in influenza As mentionedbefore a problem also exists that the measure cannot bevalidated before the intervention Therefore the sensitivityof isolation is analyzed in this group of experiments It isworth mentioning that the isolation is usually combinedwith the hospital policy It means the infected individual issent to hospital meanwhile the people who contact withthe infected are all isolated As a result parameters delay tohospital and isolation duration are designed together in thesettings of experiment Delay to hospital stands for the periodfrom infection time to hospital time while Isolation durationrepresents the period when an agent cannot contact others ifhe has contact with an infected agent The settings of theseparameters are listed in Table 8 average results are shown inFigure 25
Firstly the parameter delay to hospital is simulated aloneto test the sensitivity The simulated results are shown insolid line in Figure 25 The peak value decreases from 6708to 2218 when the delay to hospital decreases from 15 daysto 1 day The half day change brings the 67 reductionMoreover the peak value almost approximates to zero inthe case of 05 days The influenza does not happen in thiscase It can be concluded that the earlier the infected agents
Mathematical Problems in Engineering 23
Table 8 The initial settings of experiments of isolation
Parameters ValueAgent count 19610000Initial infected count 40Interventions activated time (119879Intervention) 80Delay to hospital (119879HDelay) isolation duration (119879IDuration) (15 0) (1 0) (05 0) (15 3) (1 3) (05 3)
0 50 100 150 200 2500
1000
2000
3000
4000
5000
6000
7000
Infe
cted
agen
ts
Loading intervention of isolation
Delay to hospital = 15
Delay to hospital = 10
Delay to hospital = 05
Delay to hospital = 15 isolation duration = 3
Delay to hospital = 10 isolation duration = 3
Delay to hospital = 05 isolation duration = 3
Day d
Figure 25 The results of loading intervention of isolation
are sent to hospital the less the agents will be infectedSo the construction of effective mechanism to find infectedindividuals is important in the emergency management ofinfluenza
Secondly the parameter isolation duration is added tosimulate the isolation of interventionObviously the isolationmeasure decreases the infected agents greatly Compared tothe parameter delay to hospital of 15-day case the peak valuedecreases from 6708 to 362when the isolation is added in theinterventionThe reduction of 6346 agents comes from the 3-day isolation of agents who had contacted with the infectedagents However the decrease of parameter delay to hospitalcannot bring the similar reduction in the 3-day isolationcases According to the figure the peak value decreases from362 to 51 from 15-day case to 05-day case only 311 agents aresaved from infection As a result the sensitivity of simulatedresults does not change obviously with delay to hospital inthe case of isolation It can be concluded that isolation is amore effective measure Though it is not easy for emergencymanagement organizations to decrease the delay to hospitaldue to the current monitoring mechanism the influenza canstill be controlled by isolation
544 The Experiments of Combination of Intervention Mea-sures Isolation is not the only nonmedical intervention ininfluenza close workspace is another common measureWhen close workspace is activated the workspaces are allclosed agents can only stay home at the closed time Luckilyisolation and close workspace are independent from oneanother It is possible for emergency management organi-zations to use these two measures together So this groupof experiments is designed to obtain the effectiveness of thecombination of intervention measures The parameters areinitialized with different settings listed in Table 9
Parameters delay to hospital and isolation duration is setin the middle values (1 day 3 days) according to the lastsection According to the experience close duration is set asseven days It is known that close workspace is a really costintervention measure the activated time is a key parameterWith the fixed parameters mentioned before the changeof interventions activated time is analyzed in this group ofcomputational experimentsThe average results are shown inFigure 26
Compared with the simulated results of experimentsdiscussed above Figure 26 strongly illustrates that influenzais ceased when the intervention measures are combinedtogether The peak value of infected agents is 141 in the ldquo80thdays of close-workspace and isolation with 1 day delay tohospitalrdquo case while 25135 in ldquono interventionrdquo case 2218 inldquo1 day delay to hospitalrdquo case 203 in ldquo1 day delay to hospitaland isolationrdquo case The interventions bring a great reductionof infection the peak values decrease from ten thousandslevel to hundred level
On the other hand the peak value even decreases to 108when the execution time of close workspace is 10 days earlierThe date of maximum value of infected agents is advancedfrom the 135th day to the 108th day The whole circle of H1N1influenza is reducedThe results of this group of experimentsshow that intervention measures should be activated as soonas possible and the compositions of interventions are moreeffective in emergency management of H1N1 influenza
In summary based on the general process lots of exper-iments could be done by KD-ACP to study emergency man-agement in H1N1 influenza Many conclusions are obtainedbased on the analysis of the experiments in this sectionAccording to these conclusions themost optimal response inour case study is the case of ldquo70th days of close-workspace andisolation with 1 day delay to hospitalrdquo It is because this casehas the fewest infected agents relatively But the modelling ofH1N1 influenza is not sufficient Economic and social factorsof interventions such as the money cost by vaccination andisolation the influence of the close-workspace in society
24 Mathematical Problems in Engineering
Table 9 The initial settings of experiments of isolation and closeworkspace
Parameters ValueAgent count 19610000Initial infected count 40Interventions activated time (119879Intervention) 70th day 80th dayDelay to hospital (119879HDelay) 1 dayIsolation duration (119879IDuration) 3 daysClose workspace duration (119879CWDuration) 7 days
50 100 150 200 2500
50
100
150
200
Infe
cted
agen
ts
Loading intervention of isolation and close workspace
Delay to hospital = 1 isolation duration = 3Start close time = 70 close duration time = 7Delay to hospital = 1 isolation duration = 3Start close time = 80 close duration time = 7
Day d
Figure 26The results of loading interventions of isolation and closeworkspace
and even opinion transmission of H1N1 in internet are notconsidered in our case study Without the analysis of thesefactors it does not make sense to talk about the most optimalintervention measures However KD-ACP had provided asoftware framework and general process of the research onemergency management Supported by KD-ACP the expertsof emergency management are able to study emergent eventssuch as H1N1 influenza as detailed as possible Based on theexperiments and analysis again and again the most optimalintervention measures will be found ultimately
55 Parallel Execution by ASST and OsdRT As discussed inSection 35 ASST andOsdRT are used to support the parallelexecution of artificial Beijing The statistical data is collectedand stored in artificial society runtime database at runtimeASST downloads the infection data from the database anddisplays it in the map As shown in Figure 27 the infectedagents are distributed in the districts in BeijingThe snapshotgives the geospatial distribution of the H1N1 epidemic Thelegends of the agents in different status are listed in the left
right corner in Figure 27 Five statuses are shown in the figurehealth without vaccination health with vaccination incuba-tion being symptomatic and convalescence ASST providesa very direct feeling for the emergency experts Partly withthe situation from ASST Emergency Decision Organizationmakes the emergency response plans to the response of theemergency scenario With the help of plans the epidemicin actual society can be restrained In the meantime theinfluenza information is reflected in the Internet Accordingto Figure 8 the data acquisition data extraction and datastandardization of H1N1 influenza are processed in OsdRTOsdRT is implemented by Dr Caorsquos team from Institute ofAutomation from Chinese Academy of Sciences [58] Thetool provides not only the reports of H1N1 SARS hand-foot-mouth disease and intestinal diseases are all monitored Theweb page is shown in Figure 28 H1N1 reports from Internetare collected first Based on the data extraction of the reportsH1N1 case of distribution is obtained and shown in theincidencemap As a result H1N1 incidence data is regarded asthe knowledge which is sent to EECT and IMCTThe settingsof emergent events and intervention measures are modifiedin order to make artificial society approach actual society innext turn of computational experiments With the help ofOsdRT the loop of parallel execution process integrates ACPapproach
6 Conclusion
This work provides an integrated software framework forsocial Computing in emergencymanagement From a systemperspective KD-ACP provides a reliable flexible low costand effective platform for the scientists to do the research onemergency response problems The analysis and predictionof these problems with inherent complexity can be solvedby the repetition of possible alternative experiments onthe software framework Therefore KD-ACP which is anattempt to implement ACP approach seemed as an effectivecomputational framework to support the decision making inemergency response
Currently KD-ACP is used to study the H1N1 epidemicin Beijing in 2009 The most optimal compositions of inter-vention measures are testified to support the response tonext influenza However it is still a long way ahead beforethe automation systematization and practicalization of KD-ACP A lot of work will be carried out along several directionsas in the following
(i) The parallel execution process in KD-ACP is actuallynot strongly connected with the emergency responseorganization Traditional experience and theory stillplay a dominant role in the decision of emergencymanagement It is necessary to propose a mechanismof hall for workshop to facilitate the parallel executionloop in actual society
(ii) Section 33 introduces the principle and process ofmodelling artificial society Metamodels and modelsof artificial Beijing are built by GME Model transfor-mation and code generation are used to implementcodes from FSM based models But many details
Mathematical Problems in Engineering 25
Fangshan district Daxing district Fengtai district Xicheng district
Mentougou district Haidian districtShijingshan district Dongcheng district
Incubation
Symptomatic
Convalescence
Health withoutvaccination
Health withvaccinationLegend of agent
health status
Figure 27 The situation of artificial Beijing by ASST
H1N1
Districts of Beijing
H1N1 case ofdistribution
Incidence map
Distribution ofH1N1 reportsfrom internet
Incidence of H1N1
Fangshan districtLiangxiang town
Population 19000Incidence 135789
(1100000)
Figure 28 The implementation of OsdRT by Institute of Automation from Chinese Academy of Sciences
are still not illustrated due to the topic of thispaper The key techniques such as template of codeframework and model transformations from specificdomains to FSM will be focused and introduced indetail in our next paper
(iii) The performance of computational experiments inKD-ACP is still slower than expectation The opti-mization of simulation engine and models structureis necessary to improve the performance So it ispossible to run artificial society with the real worldin parallel
In summary KD-ACP paves a new way for scientists inboth emergencymanagement and computation simulation to
collaborate with each other in solving the emergency man-agement problems by social computing
Research Highlights
(i) KD-ACP is implemented based on ACP approach
(ii) Artificial Beijing is built with the help of KD-ACP
(iii) H1N1 is modeled to simulate the influenza in Beijingin 2009
(iv) Computational experiments testify the effectivenessof intervention measures
26 Mathematical Problems in Engineering
Conflict of Interests
The authors declare that there is no conflict of interestsregarding the publication of this paper
Acknowledgment
This work was supported in part by the National ScienceFoundation of China (nos 91024030 71303252 61403402and 91324013)
References
[1] W Mao and F-Y Wang Advances in Intelligence and SecurityInformatics Academic Press Oxford UK 2012
[2] E Bonabeau ldquoAgent-based modeling methods and techniquesfor simulating human systemsrdquo Proceedings of the NationalAcademy of Sciences of the United States of America vol 99 no3 pp 7280ndash7287 2002
[3] K M Carley D B Fridsma E Casman et al ldquoBioWar scalableagent-basedmodel of bioattacksrdquo IEEETransactions on SystemsMan and Cybernetics A Systems and Humans vol 36 no 2 pp252ndash265 2006
[4] J M Epstein ldquoModelling to contain pandemicsrdquo Nature vol460 no 7256 p 687 2009
[5] Los Alamos National Laboratory Reports about EPISIMS LosAlamos National Laboratory 2003
[6] C H Builder and S C Bankes ldquoArtificial societies a conceptfor basic research on the societal impacts of informationtechnologyrdquo RAND Report 149 RAND Corp 1991
[7] F-Y Wang ldquoComputational theory and method on complexsystemrdquo China Basic Science vol 6 no 5 pp 3ndash10 2004
[8] F-Y Wang ldquoToward a paradigm shift in social computing theACP approachrdquo IEEE Intelligent Systems vol 22 no 5 pp 65ndash67 2007
[9] F-Y Wang and S Tang ldquoArtificial societies for integratedand sustainable development of metropolitan systemsrdquo IEEEIntelligent Systems vol 19 no 4 pp 82ndash87 2004
[10] D Wen Y Yuan and X-R Li ldquoArtificial societies computa-tional experiments and parallel systems an investigation ona computational theory for complex socioeconomic systemsrdquoIEEE Transactions on Services Computing vol 6 no 2 pp 177ndash185 2013
[11] F Wang F Zeng and Y Yuan ldquoAn ACP-based approach forcomplexity analysis of E-commerce systemrdquo Complex Systemsand Complexity Science vol 5 no 3 pp 1ndash8 2008 (Chinese)
[12] F-Y Wang and P K Wong ldquoResearch commentary intelligentsystems and technology for integrative and predictivemedicinean ACP approachrdquoACMTransactions on Intelligent Systems andTechnology vol 4 no 2 pp 1ndash6 2013
[13] J Sifeng X Gang F Dong and H Chunpeng ldquoStudy on theemergency rescue decision support system of petrochemicalplant based on ACP theoryrdquo in Proceedings of the 6th Manage-ment Annual Meeting (MAM rsquo11) 2011 (Chinese)
[14] X Dong D Fan G Xiong F Zhu Z Zhang and Y Yao ldquoPar-allel Bus Rapid Transit (BRT) operation management systembased on ACP approachrdquo in Proceedings of the 6th ManagementAnnual Meeting (MAM rsquo11) Beijing China 2011
[15] B Ning F-Y Wang H-R Dong R-M Li D Wen and L LildquoParallel systems for Urban rail transportation based on ACP
approachrdquo Journal of Transportation Systems Engineering andInformation Technology vol 10 no 6 pp 23ndash28 2010 (Chinese)
[16] F-Y Wang J Zhao and S-X Lun ldquoArtificial power systemsfor the operation and management of complex power gridsrdquoSouthern Power System Technology vol 2 no 3 pp 1ndash11 2008(Chinese)
[17] M J North T R Howe N T Collier and J R Vos ldquoRepastsimphony development environmentrdquo in Proceedings of theAgent 2005 Conference on Generative Social Processes Modelsand Mechanisms 2005
[18] M J North T RHowe N T Collier and J R Vos ldquoRepast Sim-phony runtime systemrdquo in Proceedings of the Agent Conferenceon Generative Social Processes Models and Mechanisms 2005
[19] SWARM Development Group Swarm 211 Reference Guide2000 httpwwwswarmorg
[20] S Tisue and U Wilensky ldquoNetLogo a simple environment formodeling complexityrdquo in International Conference on ComplexSystems pp 16ndash21 2004
[21] S Luke C Cioffi-Revilla L Panait K Sullivan and G BalanldquoMASON a multiagent simulation environmentrdquo Simulationvol 81 no 7 pp 517ndash527 2005
[22] M Ichikawa H Tanuma Y Koyama andH Deguchi ldquoSOARSintroduction as a social microscope for simulations of socialinteractions and gamingrdquo in Organizing and Learning throughGaming and Simulation Proceedings of the 38th Conference ofthe International Simulation and Gaming Association (ISAGArsquo07) pp 149ndash158 2007
[23] D S Burke J M Epstein D A T Cummings et al ldquoIndividual-based computational modeling of smallpox epidemic controlstrategiesrdquo Academic Emergency Medicine vol 13 no 11 pp1142ndash1149 2006
[24] F-Y Wang ldquoParallel control and management for intelligenttransportation systems concepts architectures and applica-tionsrdquo IEEE Transactions on Intelligent Transportation Systemsvol 11 no 3 pp 630ndash638 2010
[25] Z Molnar D Balasubramanian and A Ledeczi ldquoAn introduc-tion to the genericmodeling environmentrdquo in Proceedings of theTOOLS Europe 2007 Workshop on Model-Driven DevelopmentTool Implementers Forum Zurich Switzerland 2007
[26] S Kelly and J-P Tolvanen Domain Framework in Domain-Specific Modeling Enabling Full Code Generation JohnWiley ampSons Hoboken NJ USA 2007
[27] B P Zeigler H Praehofer and T G Kim Theory of Modelingand Simulation Integrating Discrete Event and ContinuousComplex Dynamic Systems Academic Press San Diego CalifUSA 2000
[28] G Booch J Rumbaugh and I JacobsonThe Unified ModelingLanguage User Guide Addison-Wesley Reading Mass USA1999
[29] F Wagner R Schmuki T Wagne and P Wolstenholme Mod-eling Software with Finite State Machines A Practical ApproachCRC amp Taylor amp Francis Boca Raton Fla USA 2006
[30] T Espiner ldquoChina builds worldrsquos fastest supercomputerrdquo ZDNetUK October 2010
[31] B Chen and G Guo ldquoA two-tier parallel architecture for arti-ficial society simulationrdquo in Proceedings of the ACMIEEESCS26th Workshop on Principles of Advanced and DistributedSimulation (PADS rsquo12) pp 184ndash186 July 2012
[32] G Guo B Chen X G Qiu and Z Li ldquoParallel simulation oflarge-scale artificial society on CPUGPU mixed architecturerdquo
Mathematical Problems in Engineering 27
in Proceedings of the ACMIEEESCS 26th Workshop on Prin-ciples of Advanced and Distributed Simulation (PADS rsquo12) pp174ndash177 July 2012
[33] S-X Lun ldquoResearch on the classification of parallel executionmodes of ACP theoryrdquo Acta Automatica Sinica vol 38 no 10pp 1602ndash1608 2012
[34] F-YWang ldquoA framework for social signal processing and anal-ysis from social sensing networks to computational dialecticalanalyticsrdquo Science China vol 43 no 12 pp 1598ndash1611 2013
[35] G GuoOne Model Userrsquos Manual version 12 National Univer-sity of Defense University 2011
[36] Y Ge R Meng Z Cao X Qiu and K Huang ldquoVirtual citymdashanindividual-based digital environment for human mobility andinteractive behaviorrdquo Simulation Transactions of the Society forthe Modeling and Simulation International 2014
[37] M Kretzschmar and R T Mikolajczyk ldquoContact profiles ineight European countries and implications for modelling thespread of airborne infectious diseasesrdquo PLoS ONE vol 4 no6 Article ID e5931 2009
[38] W Duan and X Qiu ldquoFostering artificial societies using sociallearning and social control in parallel emergency managementsystemsrdquo Frontiers in Computer Science vol 6 no 5 pp 604ndash610 2012
[39] W J Edmunds C J OrsquoCallaghan and D J Nokes ldquoWhomixes with whom Amethod to determine the contact patternsof adults that may lead to the spread of airborne infectionsrdquoProceedings of the Royal Society B Biological Sciences vol 264no 1384 pp 949ndash957 1997
[40] J Banks J S Carson B L Neison and D M Nicol Discrete-Event System Simulation Prentice Hall Englewood Cliffs NJUSA 4th edition 2007
[41] H Yongxia ldquoInvestigation and analysis of college studentsrsquo workand restrdquo Value Engineering vol 31 no 3 2012 (Chinese)
[42] W Duan Z Cao Y Wang et al ldquoAn ACP approach to publichealth emergency management using a campus outbreak ofh1n1 influenza as a case studyrdquo IEEE Transactions on SystemsMan and Cybernetics Part A Systems and Humans vol 43 no5 pp 1028ndash1041 2013
[43] HWHethcote ldquoThemathematics of infectious diseasesrdquo SIAMReview vol 42 no 4 pp 599ndash653 2000
[44] G Chowell E Shim F Brauer and et al ldquoModelling thetransmission dynamics of acute haemorrhagic conjunctivitisapplication to the 2003 outbreak in Mexicordquo Statistics inMedicine vol 25 no 11 pp 1840ndash1857 2006
[45] Y Ge L Liu B Chen X Qiu and K Huang ldquoAgent-basedmodeling for InfluenzaH1N1 in an artificial classroomrdquo SystemsEngineering Procedia vol 2 pp 94ndash104 2011
[46] A R Tuite A L Greer MWhelan et al ldquoEstimated epidemio-logic parameters andmorbidity associatedwith pandemicH1N1influenzardquo CMAJ vol 182 no 2 pp 131ndash136 2010
[47] L Brouwers B Cakici M Camitz A Tegnell and MBoman ldquoEconomic consequences to society of pandemic H1N1influenza 2009mdashpreliminary results for Swedenrdquo Euro Surveil-lance vol 14 no 37 pp 1ndash7 2009
[48] Y Ge L Liu X Qiu H Song Y Wang and K Huang ldquoAframework of multilayer social networks for communicationbehavior with agent-based modelingrdquo Simulation Transactionsof the Society for the Modeling and Simulation International vol89 no 7 pp 810ndash828 2013
[49] W Duan X Qiu Z Cao X Zheng and K Cui ldquoHeterogeneousand stochastic agent-based models for analyzing infectious
diseasesrsquo super spreadersrdquo IEEE Intelligent Systems vol 28 no4 pp 18ndash25 2013
[50] The Preparation and Emergency Response Plans to PandemicPublished by Ministry of Health Ministry of Health 2005(Chinese)
[51] R M Fujimoto Parallel and Distributed Simulation SystemsWiley Interscience 2000
[52] F Cicirelli A Giordano and L Nigro ldquoDistributed simulationof situated multi-agent systemsrdquo in Proceedings of the 15thIEEEACM International Symposium on Distributed Simulationand Real Time Applications (DS-RT rsquo11) pp 28ndash35 September2011
[53] B Chen and G Guo ldquoA two-tier parallel architecturefor artificial society simulationrdquo in Proceedings of the 26thACMIEEESCS Workshop on Principles of Advanced and Dis-tributed Simulation (PADS rsquo12) pp 184ndash186 July 2012
[54] G Guo B Chen and X Qiu ldquoParallel simulation of large-scale artificial society with GPU as coprocessorrdquo InternationalJournal of Modeling Simulation and Scientific Computing vol4 no 2 Article ID 1350005 2013
[55] B Chen L-B Zhang X-C Liu and H Vangheluwe ldquoActivity-based simulation using DEVS increasing performance by anactivity model in parallel DEVS simulationrdquo Journal of ZhejiangUniversity Science C vol 15 no 1 pp 13ndash30 2014
[56] X-L Wang P Yang Z-D Cao et al ldquoQuantitative evaluationon the effectiveness of prevention and control measures againstpandemic influenza A (H1N1) in Beijing 2009rdquoChinese Journalof Epidemiology vol 31 no 12 pp 1374ndash1378 2010 (Chinese)
[57] S Mei A D van de Vijver L Xuan Y Zhu and P M ASloot ldquoQuantitatively evaluating interventions in the influenzaA (H1N1) epidemic on China campus grounded on individual-based simulationsrdquo Procedia Computer Science vol 1 no 1 pp1675ndash1682 2010
[58] X Qiu Z Cao and Z Li Mid-Term Reports of China NationalScience Foundation 91024030 2014
Submit your manuscripts athttpwwwhindawicom
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
MathematicsJournal of
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Mathematical Problems in Engineering
Hindawi Publishing Corporationhttpwwwhindawicom
Differential EquationsInternational Journal of
Volume 2014
Applied MathematicsJournal of
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Probability and StatisticsHindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Journal of
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Mathematical PhysicsAdvances in
Complex AnalysisJournal of
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
OptimizationJournal of
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
CombinatoricsHindawi Publishing Corporationhttpwwwhindawicom Volume 2014
International Journal of
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Operations ResearchAdvances in
Journal of
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Function Spaces
Abstract and Applied AnalysisHindawi Publishing Corporationhttpwwwhindawicom Volume 2014
International Journal of Mathematics and Mathematical Sciences
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
The Scientific World JournalHindawi Publishing Corporation httpwwwhindawicom Volume 2014
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Algebra
Discrete Dynamics in Nature and Society
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Decision SciencesAdvances in
Discrete MathematicsJournal of
Hindawi Publishing Corporationhttpwwwhindawicom
Volume 2014 Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Stochastic AnalysisInternational Journal of
12 Mathematical Problems in Engineering
Next actionNext environment (location)
Advance to next actiontime
Find next environment inactivity chain by action
Current actionCurrent environment (location)
Q
q0
120575 Find next action inactivity schedule by time
F
Σ
(a) The state transitions of agent
Q
q0
120575 F
Current agent list
Current pollution level
Next agents list
Next pollution level
Agents entered
Infected agents entered
Add agents to currentagents list
Compute pollution levelby infected agents
Σ
(b) The state transitions of environment
Figure 11 The state transitions of models
Table 2 The activity schedule of a student agent
Duration (Δ119905) Activity Location Probability 119879Location (minute)0000ndash0600 Sleep Dormitory 119875
0(100) 360
0600ndash0800 Breakfast Dormitoryrestaurant 1198751(068) 120
Sports-breakfast-travel Playground-dormitoryrestaurant 1 minus 1198751(032)
0800ndash1200 Class Classroom 1198752(077) 240
Study Library 1 minus 1198752(023)
1200ndash1400 Lunch Restaurant 1198753(090) 120
Lunch-shopping Restaurant-convenience store 1 minus 1198753(010)
1400ndash1800 Class Classroom 1198754(070) 240
Study Library 1 minus 1198754(030)
1800-1900Dinner Dormitoryrestaurant 119875
5(063)
60Sports-shopping Playgroundconvenience store 1198756(025)
Study Classroomlibrary 1 minus (1198755+ 1198756) (012)
1900ndash2200Rest Dormitory 119875
7(045)
180Dinner-rest Dormitoryrestaurant-dormitory 1198758(020)
Study Classroomlibrary 1 minus (1198757+ 1198758) (035)
2200ndash2400 Rest Dormitory 1198759(072) 120
Sleep Dormitory 1 minus 1198759(018)
infected agents go to hospital according to the treatmentschedule or stay in dormitory according to isolation scheduleAfter the treatment in hospital or self-immunoprocess agentsbecome healthy and immune of disease If the agents aretreated in the hospital they are not allowed to get out untilrecovered Moreover activity schedules are also influencedby the emergency response plans For example in the case ofisolation in emergency response plans the activity scheduleof agents who had contacted with infected is changed Onlythe locations including home and dormitory compose thisemergent schedule Likewise based on the statistical datafrom emergency response plans the additional possibilities
are added into activity schedulesThe behaviors of agents alsochange with activity schedules In the view of ACP approachthe injection of new emergency response plans implementsthe parallel execution of microagent behaviors
In another aspect contact frequency is also anothercrucial element to determine the infected rate in public healthevent It is important to model the contact frequency andcontact time of individuals Based on the studies on thecontact behavior of human being by questionnaire surveyEdmunds found that the contact frequency of individualcould be fitted approximately into a normal distributionand the mean and the standard deviation distributions are
Mathematical Problems in Engineering 13
Susceptible agent ImmuneYes
Healthy and immune
No
Choose schedule
Self-isolation Go to hospital
Staysymptomatic
No
Receive treatment
Contact withinfected agents
Symptomatic
No
Self-immunoprocess
Yes
Recovered
Yes
Incubation
Infection
No
Yes No
NoYes
Figure 12 The workflow of a susceptible agent
168 and 85 [39] So it is possible to apply normal randomvariable to model the contact frequency In our work Box-Muller method [40] is used to generate the random variableof contact frequency shown in the following
119865119862= 120583119865+ 120590119865(minus2 ln (120574
1))12 cos (2120587120574
2) (1)
in which 119865119862is the random variable of contact frequency 120583
119865
is the mean value of normal distribution 120590119865is the standard
deviation of normal distribution and 1205741and 120574
2are the
uniform random variables distributed in the interval [0 1]Based on (1) and survey data [41] the contact frequenciesof agent 119865contact(119860 119894) are discretized as in Table 3 within theconsideration of activity differences
Similarly the duration per contact between individualsis another key factor of AHC transmission The duration ofcontact could also bemodeled by the normal randomvariablein the following [42]
119879119862= 120583119879+ 120590119879(minus2 ln (120574
1))12 cos (2120587120574
2) (2)
in which 119879119862is the random variable of duration per contact
120583119879
is the mean value of normal distribution 120590119879
is thestandard deviation of normal distribution and 120574
1and 120574
2
are the uniform random variables distributed in the interval
[0 1] In addition because the time spent in specific location119879Location(119860 119894) also follows the random distribution like (3)in our work In order to make the duration of contact119879contact(119860 119894) shorter than the 119879Location(119860 119894) the mean andstandard deviation of (2) is set up in (3) and (4) [42]
120583119879=119879Location(120583119865+ 120590119865) (3)
120590119879=119879Location(10120583119865+ 10120590
119865) (4)
in which 119879Location is the time of an activity in a specificlocation set in Table 2 120583
119879and 120590
119879are the mean and standard
deviation of contact frequency As a result the duration ofagent 119879contact(119860 119894) in specific locations (listed in Table 2) isalso discretized in Table 3 within the consideration of activitydifferences
414 Modelling Transportation In artificial Beijing threekinds of travel models are considered walking inside thedistrict travelling by road networks and travelling by subwaynetworks According to the traveling models the agentmovements are implemented by the compositions of travelingmodels before So it is important to build the basic road
14 Mathematical Problems in Engineering
Table 3 The contact frequency and duration in different locations and activities by social relationships in case of student and teacher
Social relationship Activity Location 119879contact(119860 119894) (minute) 119865contact(119860 119894)
Student-student
Sleep Dormitory 0 0
Breakfast Restaurant 119873(52) 119873(32)Breakfast Dormitory 119873(52) 119873(21)Sports Playground 119873(21) 119873(53)Class Classroom 119873(21) 119873(21)Study Library 119873(21) 119873(21)Lunch Restaurant 119873(158) 119873(32)
Shopping Convenience Store 119873(203) 119873(21)Dinner Restaurant 119873(108) 119873(21)Dinner Dormitory 119873(158) 119873(32)Rest Dormitory 119873(63) 119873(52)
Teacher-student Class Classroom 119873(21) 119873(21)
Subway network of Beijing city (red lines) Road network of Beijing city (green lines)
Figure 13 The implementation of transportation for artificial Beijing
and subway networks With the help of Google map thekey points of road are sampled first and the road networksare generated by the links among point Similarly subwaynetworks are generated by the links between subway stationsmapped from the real stations in Google map Figure 13shows the road and subway networks in artificial Beijingwith the road networks in green and the subway networks inred Most of the districts in Beijing city are covered by thesetwo networks The algorithm of hierarchical route planningis proposed as follows
(i) Find the next position where agent will be located innext action
(ii) Obtain the travelingmodels from the statistical trans-portation data
(iii) Search the path from the starting position to thenearest road entry or subway stations
(iv) Search the optimal path from the starting station orroad entry to the nearest station or road exit of targetposition
(v) Search the path from the target subway station or roadexit to the target position
(vi) Attach the subway train number or the bus numberincluding the transfer information to the path accord-ing to the timetable of subway or bus
(vii) Connect all the paths obtained before Generate thepath from the starting position to the target position
Based on the algorithm for each agent in artificial Beijingthe transportation is simulated during the computationalexperiments It is worth noting that the activities insidethe buses or train cars are considered because the agentcommunication during travelling is also necessary to besimulated
Mathematical Problems in Engineering 15
(a) The implementation of agent in MDT (b) The code framework of agent
Figure 14 The agent model implementation in MDT [35]
415 The Implementation of Models by MDT All the modelsare developed in the MDT Figure 14 shows the GUI of thetool MDT is used to implement the state space and statetransitions of FSM models The code framework is shownin Figure 14(b) Using the C++ inheritance features and thetechniques of dynamic link library (dll) the models areencapsulated in the dll componentsThe flexible compositionmechanism is designed to support the models evolution incomputational experiments
416 The Description of Artificial Beijing by ASE As men-tioned in Section 34 ASE describes the artificial Beijingin macroview The statistical information of Beijing city iscollected by ASE Table 4 lists part of the description ofartificial Beijing generated by ASE
42 How to Obtain the Data of Beijing City
421 Population and Geospatial Environment Database ofBeijing City Based on the models developed for ArtificialBeijing it is necessary to generate the initial data to do thecomputational experiments Because the individual level dataare not available it is necessary to construct an individual-based population database for both accurate computationalexperiment and determining optimal decisions
According to the state spaces of models discussed inSection 412 the geospatial and population database of Arti-ficial Beijing is designed as shown in Figure 15 The kernelpart of database consists of the tables such as agent list tablegeospatial environment list table household list table andagent distribution table Agent list table is an individual leveltable used to store the data to initialize the state space of agentmodel such as id and gender age Geospatial environmentlist table is also an individual level table which stores thedata of environment attributes such as id street id andtype Household list table and agent distribution table storethe statistical data for the data generation for agent andenvironment models With the help of the geospatial andpopulation database the data generating of artificial Beijingis proposed in the next section
422 Generating Geospatial and Social Environment Datafor Artificial Beijing by PGET The algorithm of generatinggeospatial and social environment data for artificial Beijingis used to quantify the spatial distribution of population andformalize geospatial behavior of each agent The algorithmshown in Figure 16 is capable of generating a syntheticartificial society which allowsmultiresolution statistical datasocial interactive behavior and multilayer social networksto be integrated together The synthetic population canrepresent individual agents in the form of households andhousehold members and the synthetic population is statis-tically equivalent to a real population For each householdcharacteristics such as address household size family typesand relationships are generated Each person is described bycharacteristics such as age gender social role and correlatedlocations The algorithm provides an effective methodologyto reconstruct the computing environment in high resolutionby using statistical data in low resolution leading to betterprediction and management of emergencies The details ofalgorithm are illustrated in [36]
As shown in Figure 17 the implementation of the algo-rithm is embedded inside the GEPTThe collective input datasuch as the data files of population distribution is collected inthe tool first then the algorithm is activated to generate thedatabase The generation lasts almost twelve hours one timewith the specified parameters settings When the generationis finished the data of artificial Beijing is obtained in thedatabase with 19610000 agents and 16000 environments
5 H1N1 Experiments in ArtificialBeijing with KD-ACP
With the support of the artificial Beijing modeled before thecomputational experiments and parallel execution are alsoimplemented by KD-ACP The experiments are designed tofind the most optimal emergency decision response plansEpidemic in city is a typical scenario in emergency manage-ment H1N1 epidemic in Beijing in 2009 is used to be the usedcase to test KD-ACP According to Section 34 it is necessarytomodel emergencymanagement first in order to support thecomputational experiments
16 Mathematical Problems in Engineering
Table 4 The description of artificial Beijing generated by ASE
Statistical features ofartificial Beijing Descriptions of statistical features Data sourcedefault values
Agent roles Baby child primary student middle school student college student worker retired
Demographic data file ofBeijing
Environment types Super market store restaurant park stadium hospital school cliniccommercial building dormitory building residential building
Demographic data file ofBeijing
Demographic data Age distribution of population sex distribution of population householddistribution of population retired age distribution children distribution
Demographic data file ofBeijing
Environment data Distribution of all types of environments Environment distributiondata file of Beijing
Relationships Classmate relative college friendship family Demographic data file ofBeijing
Varying parameters
Age difference of couples119873(120583120590) 119873(03) isin [0 12]Minimummarriage age (male female) (22 20)Age difference of children119873(120583120590) 119873(13) isin [1 10]Enrollment rate of kindergarten 120572 08Enrollment rate of primary school 120573 1Enrollment rate of middle school 120574 1Enrollment rate of college 120575 08Entrance age of kindergarten 1205721015840 3Entrance age of primary school 1205731015840 7Entrance age of middle school 1205741015840 13Entrance age of middle school 1205751015840 19
Age distributioninformation table
Agent settable
Geospatialenvironment set table
Household settable
Figure 15 The geospatial and population database of Beijing city
51 Modelling Emergency Management The modelling ofemergency management consists of two parts the modellingof emergent events and the modelling of intervention mea-sures H1N1 and the intervention measures for H1N1 aremodeled in our case
511 Modelling H1N1 by MDT H1N1 Model in agent simu-lates the states transitions of health status and the relevantactions Referring to SIR (SusceptibleInfectiveRecovered)[43 44] models agent has three health statuses susceptibleinfected and healthy with immunity Only susceptible agentcan be infected by the contact with infected agents After
the three infected phases (incubation being symptomaticand recovery) of H1N1 model agent is healthy again It isworthy mentioning that the recovered agents can be infectediteratively when the immunity disappears gradually by timeThe state transitions are illustrated in Figure 18
According to [45] the latent period of H1N1 was in aWeibull distribution within people from one to seven days[46 47] The distribution was usually centered in the rangebetween one and three days AWeibull random variable [40]is used to model the latent period as described in
119879Lat = 120572 times [minus ln (1 minus 120574)]1120573+ 120592 (119879Lat ge 120592) (5)
Mathematical Problems in Engineering 17
Censusdata
Generate GUID
Assign individualhousehold
Generate familystructure
Assign family roleto every member
Endow individualwith age gender
social role
Distribute environmentlocations on the map
School Workplace Entertainmentlocations
Consumptionlocations
Assign school orworkplace to
individual
Correlatedindividual withother locations
Generate friendshipin school
Generate coworkerrelations inworkplace
Assign individualadministrative
region
Generate housebuildings
Locate householdto house building
Generate familyrelation
Generateneighborhood
relation
PopulationEnvironment
Environment listSocial network
Figure 16 The algorithm of generating geospatial and social environment data [36]
Edit
Edit
Edit
Edit
Edit
Edit
BrowseStatistical characteristic of population
Geospatial population data
Population data of street
Population types
Geospatial and social environment generation tool
Workspaces
Parameters distribution settings
Household types
Roles in household
Household structure
Social roles
BrowseBrowseBrowseBrowseBrowseBrowse
Population distribution data
File path
Retired distributionPop distributionHousehold numHousehold distributionSex distributionAge distributionData type
Figure 17 The GUI of Population and Geospatial Environment generation Tool (PGET)
18 Mathematical Problems in Engineering
Infected
SusceptibleHealth
withimmunity
Infected
by contac
t
with in
fected
agents
Recovery
Increase before beingsymptomatic symptomatic
Decrease after being
Decrease by time
Infectivity
Immunity
Figure 18 The transitions of agent health status
119879Lat denotes the duration of latent period 120592 120572 and 120573 arethe location parameter scale parameter and shape parameterof Weibull distribution respectively 120574 is a uniform randomnumber in the range [0 1] According to statistics [47] 120592 120572and 120573 are set as 0 18 and 121 respectively Then the mean(standard deviation) of latent period is calculated as 159 days(058 day2) Furthermore infectious period is set as 7 days
A susceptible studentrsquos probability of getting infecteddepended on the infectivity of infectious agents his ownimmunity level the duration of the contact action and soon The infectivity of an infected agent evolved with timeduring the infectious period According to the statistics ofinfectivity in the chart with red bars in Figure 19 an infectedstudent had the highest level of infectivity in the second dayafter he had the first symptom The infectivity levels in otherdays in comparison with the second day are in Table 5 Theday labeled as ldquominus1rdquo means the day before the starting of thesymptomatic period
Based on the algorithm discussed before theH1N1modelis developed with the help of MDT The settings of influenzamodel will be discussed in Section 521
512 Modelling Intervention Measures by MDT In corre-spondence intervention measures model is used to ceasethe outbreak of H1N1 As discussed in [42 48 49] if theappropriate measures are taken when the infectious diseaseemerges the transmission of the disease could be sloweddown and the damage could be decreased Therefore it isnecessary and important to design emergency interventionmeasures in the artificial Beijing According to the MinistryofHealth inChina the intervention policies [50] are designedto control the spread of influenza such as H1N1 The inter-ventionmeasures of these policies including are interventionsactivated time vaccination rate antibiotic delay to hospitalisolation duration close workspace duration and limitationof activities listed in Table 6 Similarly the models of theseintervention measures are also developed with the help ofMDT Using the reasonable data ranges listed in Table 6intervention models can be initialized to do the computa-tional experiments These experiments are used to show how
Symptomatic1 2 3 4 5 6 7minus1
Recovered
Tim
e (da
y)
Infe
ctiv
ity
Figure 19 The distribution of infectivity rate of an infected agent[37]
Table 5 The proportion of infectivity
Days (119894) minus1 1 2 3 4 5 6 6Proportion of infectivity 03 05 1 08 04 02 01 005
to restrain the H1N1 transmissionThe settings of these inter-vention measures in IMCT are introduced in Section 522
52 Experiment Settings
521 H1N1 Settings by EECT EECT is designed to initializethe emergent events models such as H1N1The settings comefrom the specified emergent event models As shown inFigure 20 in theH1N1 used case the settings are composed ofthe distribution of infection source infection rate infectionperiod and so on The infection rate changes with envi-ronments and infection periods The settings are initializedfrom both the research of medicine on H1N1 and the data ofH1N1 outbreak in Beijing in 2009The process of influenza inartificial Beijing is simulated to repeat the possible scenariosof H1N1 break During the computational experiments ofartificial Beijing theH1N1 settings such as the period of beingsymptomatic are updated by the dynamic data registrationsfrom OsdRT The updating corrects the H1N1 model duringthe experiments With the help of OsdRT artificial Beijing isable to approach actual society It is also an implementationof parallel execution
522 Intervention Measures Settings by IMCT In the H1N1case the settings of intervention measures are shown inFigure 21 in IMCT These measures are designed under theinstruction of Emergency Decision Organizations like ChinaDisease Center (CDC) The measures such as vaccinationrate the rate of closedworkspace and the starting time can beinitialized in the tool Emergency response plans are imple-mented in these settings As illustrated in Figure 7 all the pos-sible compositions within value settings in reasonable inter-vals are tested to find the most optimal emergency responseplan For example different compositions of measures suchas vaccination rates and rates of workspace closed are usedto design the plans of experiments By the analysis of theresults the best composition of interventionmeasures can beobtainedThe analysis will be discussed in case study in detail
Mathematical Problems in Engineering 19
Table 6 The intervention measures of influenza
Intervention measures ValuesInterventions activated Time 119879Intervention isin [0 80th day]Vaccination rate 119877Vaccination isin [0 1]Antibiotic rate 119877Antibiotic isin [0 1]Delay to hospital 119867ospital(119860 119894) 119879infected lt 119905 lt 119879recovered + 119879HDelay 119879HDelay isin (0 3th day]Isolation duration 119868solation(119860 119894) 119879Intervention lt 119905 lt 119879Intervention + 119879IDuration 119879IDuration isin (0 14th day]Close workspace duration 119862lose(119864119894) 119879Intervention lt 119905 lt 119879Intervention + 119879CWDuration 119864119894 isinWorkspace 119879CWDuration isin (0 7th day]Limitation of agent activity duration 119871 imitation(119860 119894) 119879Intervention lt 119905 lt 119879Intervention + 119879LAADuration 119879LAADuration isin (0 7th day]119877Vaccination means the vaccination rate of artificial Beijing while 119877Antibiotic means the antibiotic rate used for infected agents in hospital119867ospital (119860119894) means that the infected agent 119860119894 is sent to hospital he or she will receive the retreatment immediately119868solation (119860119894) means that the infected agent 119860119894 is isolated at home he or she can only contact with the other agents until isolation duration is over119862lose (119864119894) means that the environment 119864119894 typed of workspace is closed until the close workspace duration is over119871 imitation (119860119894) means that the activities of agent 119860119894 are limited only the workspace and home are allowed until limitation duration is over
Incubation Weibulldistribution
Symbolic LogNormaldistribution
Recovering LogNormaldistribution
H1N1influenza
Settings of H1N1 influenza model
Disease model
Transmissionmodel
Figure 20 The settings of H1N1 in EECT
It is worth notifying that the settings of interventionmeasures are also influenced by OsdRT For example theH1N1 outbreak areas are notified by OsdRTThe interventionmeasures such as close workspace are used only in theoutbreak areas Therefore the unnecessary computation ofintervention measures is avoided So the data injection byOsdRT not only makes artificial society approach actualsociety but also increases the performance of computationalexperiments
53 Experiments by EMT When the initialization of arti-ficial Beijing is ready EMT is used to do the computa-tional experiments EMT is composed of two parts thecentral controller in the central node and the residen-tial service deployed in the computational nodes Centralcontroller loads the experiment plans and controls the whole
experiments process the GUI of controller is shown inFigure 22 EMT manages the computational experimentsin four steps cluster configuration load experiment plansmodels deployment and running experiments
531 Cluster Configuration Before the configuration clusterinformation is collected from the residential services insidethe cluster The information includes nodes amount nodenames node IPs and CPU occupations It is used to quantifythe computing power of the clusterThe cluster configurationis based on this information The nodes are selected forexperiments Then the LP (local process) [51] number foreach node is set according to the number of the CPU kernelsAfter the configuration the experiment plans are loaded tocustomize the models and data
20 Mathematical Problems in Engineering
Streets of close workspaces
Limitations ofactivities
IsolationsRespirator
Closeworkspace
Antibiotic
VaccinationHospital
Surveillance
Figure 21 The settings of intervention measures in IMCT
Information ofexperiment plan
List of experiment plans
Start | Pause | Reset | Stop | Speedup | Slowdown | Back
NodeName | CPU occupation | Memory occupation |Network | Models | Status
Figure 22 The deployment and running of experiments in EMT
532 Load Experiment Plan As discussed in Section 35 thedescription of experiment plans is composed of the infor-mation in three aspects models and the initial data for theexperiments the deployment mapping tables from the mod-els to the nodes and the plan of experiment executionWhenthe experiment plan is loaded by the EMT the file namesand model descriptions of models and data are listed Withthe support of cluster configuration models can be mappedto the nodes in the GUI according to the computationrequirements EMT also provides the setting of experimentsitself including the running times start time and end time
533 Models and Data Deployment Models are stored inthe agent population and services repository According tothe settings in experiment plan the customized models aredownloaded from the repository first then the models suchas agent model environment model emergent events modeland intervention measures model are integrated to be thecompositionmodels of artificial BeijingThemodels and datalocations in the nodes are configured in the ldquomodel pathrdquoand ldquodata pathrdquo in the GUI of central controller Finallythe models are uploaded to the nodes in cluster with theconsideration of partition The partition of artificial society
Mathematical Problems in Engineering 21
models is different from the multiagent system in [52]Models in artificial society are divided into three levels agentsand environments grids and city Agents and environmentsare encapsulated in grids which compose the whole cityThus the partition of artificial society becomes the partitionof grids The parallelism degrees are found to support thepartition of models agent independence and grid indepen-dence According to the parallelisms agents are partitioned toLPs according to the population distribution in grids Agentmovements among grids are simulated by the movementevents among grids in different LPs As a result a two-tierparallel architecture is proposed to support the simulation ofartificial society The architecture is detailed in [53]
534 Running of Experiments When the preparation isready the experiments can be done by the central controllerof EMT The running and control commands of the experi-ments are grouped into three aspects start pause or stop theexperiments get the running information of the nodes suchas the snapshot of the desktop of node With the help of thesecommands users can run the computational experimentsto do the research on the emergency management It isworth notifying that the simulation engine is optimized inorder to support the ten million agent simulation UsingGPU as coprocessor a two-tier parallel simulation engine isdesigned with support of MPI and OpenCL through phasedsynchronization [54] One-sided communication is used forreflection of remote simulation objects and message passingbetween processes A general kernel function prototype iselaborately designed and conditionally compiled for execu-tion on both CPU and GPU Moreover the optimizationoperations like load balance is developed under the instruc-tion of activity-based simulation [55] The densely populatedgrids are given more CPU and memory resource accordingto the activity predictions
54 Results of H1N1 Experiments with KD-ACP On thebasis of artificial society models and the intervention mea-sures mentioned in Section 512 a series of computationalexperiments are performed to study the H1N1 influenza inBeijing In our case of artificial Beijing 19610000 agentsare simulated in the cluster within 48 CPU cores and 128Gmemories the nodes are connected by kilomega networks Ittakes 18 hours to simulate a 250-day disease spreading
In order to illustrate how the optimal plans are obtainedbyKD-ACP the experiments are divided into four groups theexperiments ofmodel validation the experiments of sensitiv-ity analysis of vaccination rate the experiments of sensitivityanalysis of isolation and the experiments of the combinationof interventionmeasuresThese four groups show the generalprocess of the research on emergency management in H1N1influenza Firstly the models of H1N1 are validated withthe support of historical data Secondly the traditionalmedical interventions of pandemic such as vaccination areanalyzed to find the optimal plan Thirdly the traditionalnonmedical interventions of pandemic such as isolation areanalyzed to find the optimal plan Finally the combination ofinterventions is tested to show the combined effect
It is worth mentioning that all the experiments in ourwork are performed 100 times These experiments are ini-tialized with different settings of interventions in IMCT Theresults shown in the figures are the mean values
541 The Experiments of Model Validation In our artifi-cial Beijing the models of H1N1 influenza are built basedon the existing researches Wang and her colleagues hadstudied the H1N1 influenza by SEIR (SusceptibleExposedInfectiveRecovered) models [56] With the support of theirwork we build the models of artificial Beijing in the mannerof multiagents
In Figure 23 the control group shown in red line isdrawn based on the realistic statistical data collected in H1N1influenza in Beijing in 2009 The influenza lasts more thansix months and more than 170000 people are infectedComparing with the control group the simulated resultswithout interventions are drawn in blue It is obvious that thesimulated case fits the control group well before 70 days Butthe ascending of control group is slowed after 70 days andthe peak value (17883) is much less than the simulated results(25135) It is because the vaccination is activated at day 67 incontrol group according to the intervention measures usedin 2009 The restrain by vaccination becomes effective eightdays later The change of infected number after vaccinationis shown by red spotted line The trend of control group isdifferent from the simulation data after 75 days
However the simulated results (blue line) in compu-tational experiments fit the control group (as shown inFigure 23) in the first 70 days when no intervention mea-sures are executed The key features such as effective basicreproduction number (119877
0) are also almost the same in both
cases Therefore it can be concluded that the models ofartificial society built in microview are validated by therealistic statistical data in influenza in Beijing in 2009
In this group of experiments no interventions are exe-cuted in order to validate the models Therefore the simu-lated results of this extreme case which cannot happen todaycould be obtained As shown by blue line the number ofinfected agents without interventions reaches peak value atday 91 with 25135 consisting of 8052 in symptomatic phaseand 28950 in recovery phase It is obvious that the totalnumber of infected agents grows slowly at the beginningtime Then the number increases quickly when the influenzabreaks and it reaches its maximum value with 383870 in theend This phenomenon can be explained as in the followingAt the beginning of influenza due to the limited numberof infected agents the propagation of the disease remainsslow but many agents become susceptible When the numberof infected agents increases rapidly the spreading becomessignificant However the number of infected agents decreasesafter day 91 It is because the disease spreading relies on thesocial networks and spatial contact networks [49] It meansthat the infected agents can only infect agents either in thesocial relationship networks or in the spatially contacted
542 The Experiments of Sensitivity Analysis of VaccinationRate According to Section 512 vaccination is a typical
22 Mathematical Problems in Engineering
0
Infe
cted
agen
ts
0 50 100 150 200 250
Statistical dataSimulation data
25000
20000
15000
10000
5000
Day d
Figure 23 The results with no intervention executed and the com-parison with realistic statistical data in Beijing in 2009
medical intervention in influenza Traditionally the effective-ness of vaccination can only be evaluated after the influenzaand the key settings such as vaccination rate are determinedby the experience It is always difficult for emergency man-agement organizations to find the most optimal vaccinationrate in advanceTherefore the sensitivity of vaccination rate isanalyzed in this group of experiments The settings are listedin Table 7 average results are shown in Figure 24
Vaccination rates are set linearly (10 30 50 and70) to test the changes of influenza The peak value ofinfected agents decreases from 25135 22392 15787 6907to 638 respectively It shows that the more the agents arevaccinated the smaller the peak value of the infected agentswill be In the case of 70 vaccination the total infected agentis only 13799 It means that the influenza in this case does nothappen due to the high vaccination rate
The change of peak values also shows an interestingphenomenon Vaccination rates increase linearly the peakvalues decrease slower than the ascending of vaccination ratesat beginning but the descending becomes more and morefaster with the linear increase of vaccination rate It meansthat the small scale of vaccination (less than 30) does notmake senseWith the increase of vaccination rate the numberof susceptible agents decreases significantly In sequencethe possibility of infections is lowered greatly Additionallyit could be found that the decrease of peak values comeslater than the increase of the vaccination rates This is quitereasonable according to Mei et alrsquos work in [57]
As a result it is concluded that high vaccination rate iseffective in the control of H1N1 influenza though costly
543 The Experiments of Sensitivity Analysis of IsolationAccording to Section 512 isolation is a typical nonmedicalintervention It is the most used intervention measure in
Table 7 The initial settings of experiments of vaccination
Parameters ValueAgent count 19610000Initial infected count 40Interventions activated time(119879Intervention)
0
Vaccination rate (119877Vaccination) 0 10 30 50 70
0
Infe
cted
agen
ts
Loading intervention of vaccination
No vaccination10 vaccination30 vaccination
50 vaccination70 vaccination
0 50 100 150 200 250
25000
20000
15000
10000
5000
Day d
Figure 24 The results of loading intervention of vaccination
the emergency management in influenza As mentionedbefore a problem also exists that the measure cannot bevalidated before the intervention Therefore the sensitivityof isolation is analyzed in this group of experiments It isworth mentioning that the isolation is usually combinedwith the hospital policy It means the infected individual issent to hospital meanwhile the people who contact withthe infected are all isolated As a result parameters delay tohospital and isolation duration are designed together in thesettings of experiment Delay to hospital stands for the periodfrom infection time to hospital time while Isolation durationrepresents the period when an agent cannot contact others ifhe has contact with an infected agent The settings of theseparameters are listed in Table 8 average results are shown inFigure 25
Firstly the parameter delay to hospital is simulated aloneto test the sensitivity The simulated results are shown insolid line in Figure 25 The peak value decreases from 6708to 2218 when the delay to hospital decreases from 15 daysto 1 day The half day change brings the 67 reductionMoreover the peak value almost approximates to zero inthe case of 05 days The influenza does not happen in thiscase It can be concluded that the earlier the infected agents
Mathematical Problems in Engineering 23
Table 8 The initial settings of experiments of isolation
Parameters ValueAgent count 19610000Initial infected count 40Interventions activated time (119879Intervention) 80Delay to hospital (119879HDelay) isolation duration (119879IDuration) (15 0) (1 0) (05 0) (15 3) (1 3) (05 3)
0 50 100 150 200 2500
1000
2000
3000
4000
5000
6000
7000
Infe
cted
agen
ts
Loading intervention of isolation
Delay to hospital = 15
Delay to hospital = 10
Delay to hospital = 05
Delay to hospital = 15 isolation duration = 3
Delay to hospital = 10 isolation duration = 3
Delay to hospital = 05 isolation duration = 3
Day d
Figure 25 The results of loading intervention of isolation
are sent to hospital the less the agents will be infectedSo the construction of effective mechanism to find infectedindividuals is important in the emergency management ofinfluenza
Secondly the parameter isolation duration is added tosimulate the isolation of interventionObviously the isolationmeasure decreases the infected agents greatly Compared tothe parameter delay to hospital of 15-day case the peak valuedecreases from 6708 to 362when the isolation is added in theinterventionThe reduction of 6346 agents comes from the 3-day isolation of agents who had contacted with the infectedagents However the decrease of parameter delay to hospitalcannot bring the similar reduction in the 3-day isolationcases According to the figure the peak value decreases from362 to 51 from 15-day case to 05-day case only 311 agents aresaved from infection As a result the sensitivity of simulatedresults does not change obviously with delay to hospital inthe case of isolation It can be concluded that isolation is amore effective measure Though it is not easy for emergencymanagement organizations to decrease the delay to hospitaldue to the current monitoring mechanism the influenza canstill be controlled by isolation
544 The Experiments of Combination of Intervention Mea-sures Isolation is not the only nonmedical intervention ininfluenza close workspace is another common measureWhen close workspace is activated the workspaces are allclosed agents can only stay home at the closed time Luckilyisolation and close workspace are independent from oneanother It is possible for emergency management organi-zations to use these two measures together So this groupof experiments is designed to obtain the effectiveness of thecombination of intervention measures The parameters areinitialized with different settings listed in Table 9
Parameters delay to hospital and isolation duration is setin the middle values (1 day 3 days) according to the lastsection According to the experience close duration is set asseven days It is known that close workspace is a really costintervention measure the activated time is a key parameterWith the fixed parameters mentioned before the changeof interventions activated time is analyzed in this group ofcomputational experimentsThe average results are shown inFigure 26
Compared with the simulated results of experimentsdiscussed above Figure 26 strongly illustrates that influenzais ceased when the intervention measures are combinedtogether The peak value of infected agents is 141 in the ldquo80thdays of close-workspace and isolation with 1 day delay tohospitalrdquo case while 25135 in ldquono interventionrdquo case 2218 inldquo1 day delay to hospitalrdquo case 203 in ldquo1 day delay to hospitaland isolationrdquo case The interventions bring a great reductionof infection the peak values decrease from ten thousandslevel to hundred level
On the other hand the peak value even decreases to 108when the execution time of close workspace is 10 days earlierThe date of maximum value of infected agents is advancedfrom the 135th day to the 108th day The whole circle of H1N1influenza is reducedThe results of this group of experimentsshow that intervention measures should be activated as soonas possible and the compositions of interventions are moreeffective in emergency management of H1N1 influenza
In summary based on the general process lots of exper-iments could be done by KD-ACP to study emergency man-agement in H1N1 influenza Many conclusions are obtainedbased on the analysis of the experiments in this sectionAccording to these conclusions themost optimal response inour case study is the case of ldquo70th days of close-workspace andisolation with 1 day delay to hospitalrdquo It is because this casehas the fewest infected agents relatively But the modelling ofH1N1 influenza is not sufficient Economic and social factorsof interventions such as the money cost by vaccination andisolation the influence of the close-workspace in society
24 Mathematical Problems in Engineering
Table 9 The initial settings of experiments of isolation and closeworkspace
Parameters ValueAgent count 19610000Initial infected count 40Interventions activated time (119879Intervention) 70th day 80th dayDelay to hospital (119879HDelay) 1 dayIsolation duration (119879IDuration) 3 daysClose workspace duration (119879CWDuration) 7 days
50 100 150 200 2500
50
100
150
200
Infe
cted
agen
ts
Loading intervention of isolation and close workspace
Delay to hospital = 1 isolation duration = 3Start close time = 70 close duration time = 7Delay to hospital = 1 isolation duration = 3Start close time = 80 close duration time = 7
Day d
Figure 26The results of loading interventions of isolation and closeworkspace
and even opinion transmission of H1N1 in internet are notconsidered in our case study Without the analysis of thesefactors it does not make sense to talk about the most optimalintervention measures However KD-ACP had provided asoftware framework and general process of the research onemergency management Supported by KD-ACP the expertsof emergency management are able to study emergent eventssuch as H1N1 influenza as detailed as possible Based on theexperiments and analysis again and again the most optimalintervention measures will be found ultimately
55 Parallel Execution by ASST and OsdRT As discussed inSection 35 ASST andOsdRT are used to support the parallelexecution of artificial Beijing The statistical data is collectedand stored in artificial society runtime database at runtimeASST downloads the infection data from the database anddisplays it in the map As shown in Figure 27 the infectedagents are distributed in the districts in BeijingThe snapshotgives the geospatial distribution of the H1N1 epidemic Thelegends of the agents in different status are listed in the left
right corner in Figure 27 Five statuses are shown in the figurehealth without vaccination health with vaccination incuba-tion being symptomatic and convalescence ASST providesa very direct feeling for the emergency experts Partly withthe situation from ASST Emergency Decision Organizationmakes the emergency response plans to the response of theemergency scenario With the help of plans the epidemicin actual society can be restrained In the meantime theinfluenza information is reflected in the Internet Accordingto Figure 8 the data acquisition data extraction and datastandardization of H1N1 influenza are processed in OsdRTOsdRT is implemented by Dr Caorsquos team from Institute ofAutomation from Chinese Academy of Sciences [58] Thetool provides not only the reports of H1N1 SARS hand-foot-mouth disease and intestinal diseases are all monitored Theweb page is shown in Figure 28 H1N1 reports from Internetare collected first Based on the data extraction of the reportsH1N1 case of distribution is obtained and shown in theincidencemap As a result H1N1 incidence data is regarded asthe knowledge which is sent to EECT and IMCTThe settingsof emergent events and intervention measures are modifiedin order to make artificial society approach actual society innext turn of computational experiments With the help ofOsdRT the loop of parallel execution process integrates ACPapproach
6 Conclusion
This work provides an integrated software framework forsocial Computing in emergencymanagement From a systemperspective KD-ACP provides a reliable flexible low costand effective platform for the scientists to do the research onemergency response problems The analysis and predictionof these problems with inherent complexity can be solvedby the repetition of possible alternative experiments onthe software framework Therefore KD-ACP which is anattempt to implement ACP approach seemed as an effectivecomputational framework to support the decision making inemergency response
Currently KD-ACP is used to study the H1N1 epidemicin Beijing in 2009 The most optimal compositions of inter-vention measures are testified to support the response tonext influenza However it is still a long way ahead beforethe automation systematization and practicalization of KD-ACP A lot of work will be carried out along several directionsas in the following
(i) The parallel execution process in KD-ACP is actuallynot strongly connected with the emergency responseorganization Traditional experience and theory stillplay a dominant role in the decision of emergencymanagement It is necessary to propose a mechanismof hall for workshop to facilitate the parallel executionloop in actual society
(ii) Section 33 introduces the principle and process ofmodelling artificial society Metamodels and modelsof artificial Beijing are built by GME Model transfor-mation and code generation are used to implementcodes from FSM based models But many details
Mathematical Problems in Engineering 25
Fangshan district Daxing district Fengtai district Xicheng district
Mentougou district Haidian districtShijingshan district Dongcheng district
Incubation
Symptomatic
Convalescence
Health withoutvaccination
Health withvaccinationLegend of agent
health status
Figure 27 The situation of artificial Beijing by ASST
H1N1
Districts of Beijing
H1N1 case ofdistribution
Incidence map
Distribution ofH1N1 reportsfrom internet
Incidence of H1N1
Fangshan districtLiangxiang town
Population 19000Incidence 135789
(1100000)
Figure 28 The implementation of OsdRT by Institute of Automation from Chinese Academy of Sciences
are still not illustrated due to the topic of thispaper The key techniques such as template of codeframework and model transformations from specificdomains to FSM will be focused and introduced indetail in our next paper
(iii) The performance of computational experiments inKD-ACP is still slower than expectation The opti-mization of simulation engine and models structureis necessary to improve the performance So it ispossible to run artificial society with the real worldin parallel
In summary KD-ACP paves a new way for scientists inboth emergencymanagement and computation simulation to
collaborate with each other in solving the emergency man-agement problems by social computing
Research Highlights
(i) KD-ACP is implemented based on ACP approach
(ii) Artificial Beijing is built with the help of KD-ACP
(iii) H1N1 is modeled to simulate the influenza in Beijingin 2009
(iv) Computational experiments testify the effectivenessof intervention measures
26 Mathematical Problems in Engineering
Conflict of Interests
The authors declare that there is no conflict of interestsregarding the publication of this paper
Acknowledgment
This work was supported in part by the National ScienceFoundation of China (nos 91024030 71303252 61403402and 91324013)
References
[1] W Mao and F-Y Wang Advances in Intelligence and SecurityInformatics Academic Press Oxford UK 2012
[2] E Bonabeau ldquoAgent-based modeling methods and techniquesfor simulating human systemsrdquo Proceedings of the NationalAcademy of Sciences of the United States of America vol 99 no3 pp 7280ndash7287 2002
[3] K M Carley D B Fridsma E Casman et al ldquoBioWar scalableagent-basedmodel of bioattacksrdquo IEEETransactions on SystemsMan and Cybernetics A Systems and Humans vol 36 no 2 pp252ndash265 2006
[4] J M Epstein ldquoModelling to contain pandemicsrdquo Nature vol460 no 7256 p 687 2009
[5] Los Alamos National Laboratory Reports about EPISIMS LosAlamos National Laboratory 2003
[6] C H Builder and S C Bankes ldquoArtificial societies a conceptfor basic research on the societal impacts of informationtechnologyrdquo RAND Report 149 RAND Corp 1991
[7] F-Y Wang ldquoComputational theory and method on complexsystemrdquo China Basic Science vol 6 no 5 pp 3ndash10 2004
[8] F-Y Wang ldquoToward a paradigm shift in social computing theACP approachrdquo IEEE Intelligent Systems vol 22 no 5 pp 65ndash67 2007
[9] F-Y Wang and S Tang ldquoArtificial societies for integratedand sustainable development of metropolitan systemsrdquo IEEEIntelligent Systems vol 19 no 4 pp 82ndash87 2004
[10] D Wen Y Yuan and X-R Li ldquoArtificial societies computa-tional experiments and parallel systems an investigation ona computational theory for complex socioeconomic systemsrdquoIEEE Transactions on Services Computing vol 6 no 2 pp 177ndash185 2013
[11] F Wang F Zeng and Y Yuan ldquoAn ACP-based approach forcomplexity analysis of E-commerce systemrdquo Complex Systemsand Complexity Science vol 5 no 3 pp 1ndash8 2008 (Chinese)
[12] F-Y Wang and P K Wong ldquoResearch commentary intelligentsystems and technology for integrative and predictivemedicinean ACP approachrdquoACMTransactions on Intelligent Systems andTechnology vol 4 no 2 pp 1ndash6 2013
[13] J Sifeng X Gang F Dong and H Chunpeng ldquoStudy on theemergency rescue decision support system of petrochemicalplant based on ACP theoryrdquo in Proceedings of the 6th Manage-ment Annual Meeting (MAM rsquo11) 2011 (Chinese)
[14] X Dong D Fan G Xiong F Zhu Z Zhang and Y Yao ldquoPar-allel Bus Rapid Transit (BRT) operation management systembased on ACP approachrdquo in Proceedings of the 6th ManagementAnnual Meeting (MAM rsquo11) Beijing China 2011
[15] B Ning F-Y Wang H-R Dong R-M Li D Wen and L LildquoParallel systems for Urban rail transportation based on ACP
approachrdquo Journal of Transportation Systems Engineering andInformation Technology vol 10 no 6 pp 23ndash28 2010 (Chinese)
[16] F-Y Wang J Zhao and S-X Lun ldquoArtificial power systemsfor the operation and management of complex power gridsrdquoSouthern Power System Technology vol 2 no 3 pp 1ndash11 2008(Chinese)
[17] M J North T R Howe N T Collier and J R Vos ldquoRepastsimphony development environmentrdquo in Proceedings of theAgent 2005 Conference on Generative Social Processes Modelsand Mechanisms 2005
[18] M J North T RHowe N T Collier and J R Vos ldquoRepast Sim-phony runtime systemrdquo in Proceedings of the Agent Conferenceon Generative Social Processes Models and Mechanisms 2005
[19] SWARM Development Group Swarm 211 Reference Guide2000 httpwwwswarmorg
[20] S Tisue and U Wilensky ldquoNetLogo a simple environment formodeling complexityrdquo in International Conference on ComplexSystems pp 16ndash21 2004
[21] S Luke C Cioffi-Revilla L Panait K Sullivan and G BalanldquoMASON a multiagent simulation environmentrdquo Simulationvol 81 no 7 pp 517ndash527 2005
[22] M Ichikawa H Tanuma Y Koyama andH Deguchi ldquoSOARSintroduction as a social microscope for simulations of socialinteractions and gamingrdquo in Organizing and Learning throughGaming and Simulation Proceedings of the 38th Conference ofthe International Simulation and Gaming Association (ISAGArsquo07) pp 149ndash158 2007
[23] D S Burke J M Epstein D A T Cummings et al ldquoIndividual-based computational modeling of smallpox epidemic controlstrategiesrdquo Academic Emergency Medicine vol 13 no 11 pp1142ndash1149 2006
[24] F-Y Wang ldquoParallel control and management for intelligenttransportation systems concepts architectures and applica-tionsrdquo IEEE Transactions on Intelligent Transportation Systemsvol 11 no 3 pp 630ndash638 2010
[25] Z Molnar D Balasubramanian and A Ledeczi ldquoAn introduc-tion to the genericmodeling environmentrdquo in Proceedings of theTOOLS Europe 2007 Workshop on Model-Driven DevelopmentTool Implementers Forum Zurich Switzerland 2007
[26] S Kelly and J-P Tolvanen Domain Framework in Domain-Specific Modeling Enabling Full Code Generation JohnWiley ampSons Hoboken NJ USA 2007
[27] B P Zeigler H Praehofer and T G Kim Theory of Modelingand Simulation Integrating Discrete Event and ContinuousComplex Dynamic Systems Academic Press San Diego CalifUSA 2000
[28] G Booch J Rumbaugh and I JacobsonThe Unified ModelingLanguage User Guide Addison-Wesley Reading Mass USA1999
[29] F Wagner R Schmuki T Wagne and P Wolstenholme Mod-eling Software with Finite State Machines A Practical ApproachCRC amp Taylor amp Francis Boca Raton Fla USA 2006
[30] T Espiner ldquoChina builds worldrsquos fastest supercomputerrdquo ZDNetUK October 2010
[31] B Chen and G Guo ldquoA two-tier parallel architecture for arti-ficial society simulationrdquo in Proceedings of the ACMIEEESCS26th Workshop on Principles of Advanced and DistributedSimulation (PADS rsquo12) pp 184ndash186 July 2012
[32] G Guo B Chen X G Qiu and Z Li ldquoParallel simulation oflarge-scale artificial society on CPUGPU mixed architecturerdquo
Mathematical Problems in Engineering 27
in Proceedings of the ACMIEEESCS 26th Workshop on Prin-ciples of Advanced and Distributed Simulation (PADS rsquo12) pp174ndash177 July 2012
[33] S-X Lun ldquoResearch on the classification of parallel executionmodes of ACP theoryrdquo Acta Automatica Sinica vol 38 no 10pp 1602ndash1608 2012
[34] F-YWang ldquoA framework for social signal processing and anal-ysis from social sensing networks to computational dialecticalanalyticsrdquo Science China vol 43 no 12 pp 1598ndash1611 2013
[35] G GuoOne Model Userrsquos Manual version 12 National Univer-sity of Defense University 2011
[36] Y Ge R Meng Z Cao X Qiu and K Huang ldquoVirtual citymdashanindividual-based digital environment for human mobility andinteractive behaviorrdquo Simulation Transactions of the Society forthe Modeling and Simulation International 2014
[37] M Kretzschmar and R T Mikolajczyk ldquoContact profiles ineight European countries and implications for modelling thespread of airborne infectious diseasesrdquo PLoS ONE vol 4 no6 Article ID e5931 2009
[38] W Duan and X Qiu ldquoFostering artificial societies using sociallearning and social control in parallel emergency managementsystemsrdquo Frontiers in Computer Science vol 6 no 5 pp 604ndash610 2012
[39] W J Edmunds C J OrsquoCallaghan and D J Nokes ldquoWhomixes with whom Amethod to determine the contact patternsof adults that may lead to the spread of airborne infectionsrdquoProceedings of the Royal Society B Biological Sciences vol 264no 1384 pp 949ndash957 1997
[40] J Banks J S Carson B L Neison and D M Nicol Discrete-Event System Simulation Prentice Hall Englewood Cliffs NJUSA 4th edition 2007
[41] H Yongxia ldquoInvestigation and analysis of college studentsrsquo workand restrdquo Value Engineering vol 31 no 3 2012 (Chinese)
[42] W Duan Z Cao Y Wang et al ldquoAn ACP approach to publichealth emergency management using a campus outbreak ofh1n1 influenza as a case studyrdquo IEEE Transactions on SystemsMan and Cybernetics Part A Systems and Humans vol 43 no5 pp 1028ndash1041 2013
[43] HWHethcote ldquoThemathematics of infectious diseasesrdquo SIAMReview vol 42 no 4 pp 599ndash653 2000
[44] G Chowell E Shim F Brauer and et al ldquoModelling thetransmission dynamics of acute haemorrhagic conjunctivitisapplication to the 2003 outbreak in Mexicordquo Statistics inMedicine vol 25 no 11 pp 1840ndash1857 2006
[45] Y Ge L Liu B Chen X Qiu and K Huang ldquoAgent-basedmodeling for InfluenzaH1N1 in an artificial classroomrdquo SystemsEngineering Procedia vol 2 pp 94ndash104 2011
[46] A R Tuite A L Greer MWhelan et al ldquoEstimated epidemio-logic parameters andmorbidity associatedwith pandemicH1N1influenzardquo CMAJ vol 182 no 2 pp 131ndash136 2010
[47] L Brouwers B Cakici M Camitz A Tegnell and MBoman ldquoEconomic consequences to society of pandemic H1N1influenza 2009mdashpreliminary results for Swedenrdquo Euro Surveil-lance vol 14 no 37 pp 1ndash7 2009
[48] Y Ge L Liu X Qiu H Song Y Wang and K Huang ldquoAframework of multilayer social networks for communicationbehavior with agent-based modelingrdquo Simulation Transactionsof the Society for the Modeling and Simulation International vol89 no 7 pp 810ndash828 2013
[49] W Duan X Qiu Z Cao X Zheng and K Cui ldquoHeterogeneousand stochastic agent-based models for analyzing infectious
diseasesrsquo super spreadersrdquo IEEE Intelligent Systems vol 28 no4 pp 18ndash25 2013
[50] The Preparation and Emergency Response Plans to PandemicPublished by Ministry of Health Ministry of Health 2005(Chinese)
[51] R M Fujimoto Parallel and Distributed Simulation SystemsWiley Interscience 2000
[52] F Cicirelli A Giordano and L Nigro ldquoDistributed simulationof situated multi-agent systemsrdquo in Proceedings of the 15thIEEEACM International Symposium on Distributed Simulationand Real Time Applications (DS-RT rsquo11) pp 28ndash35 September2011
[53] B Chen and G Guo ldquoA two-tier parallel architecturefor artificial society simulationrdquo in Proceedings of the 26thACMIEEESCS Workshop on Principles of Advanced and Dis-tributed Simulation (PADS rsquo12) pp 184ndash186 July 2012
[54] G Guo B Chen and X Qiu ldquoParallel simulation of large-scale artificial society with GPU as coprocessorrdquo InternationalJournal of Modeling Simulation and Scientific Computing vol4 no 2 Article ID 1350005 2013
[55] B Chen L-B Zhang X-C Liu and H Vangheluwe ldquoActivity-based simulation using DEVS increasing performance by anactivity model in parallel DEVS simulationrdquo Journal of ZhejiangUniversity Science C vol 15 no 1 pp 13ndash30 2014
[56] X-L Wang P Yang Z-D Cao et al ldquoQuantitative evaluationon the effectiveness of prevention and control measures againstpandemic influenza A (H1N1) in Beijing 2009rdquoChinese Journalof Epidemiology vol 31 no 12 pp 1374ndash1378 2010 (Chinese)
[57] S Mei A D van de Vijver L Xuan Y Zhu and P M ASloot ldquoQuantitatively evaluating interventions in the influenzaA (H1N1) epidemic on China campus grounded on individual-based simulationsrdquo Procedia Computer Science vol 1 no 1 pp1675ndash1682 2010
[58] X Qiu Z Cao and Z Li Mid-Term Reports of China NationalScience Foundation 91024030 2014
Submit your manuscripts athttpwwwhindawicom
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
MathematicsJournal of
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Mathematical Problems in Engineering
Hindawi Publishing Corporationhttpwwwhindawicom
Differential EquationsInternational Journal of
Volume 2014
Applied MathematicsJournal of
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Probability and StatisticsHindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Journal of
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Mathematical PhysicsAdvances in
Complex AnalysisJournal of
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
OptimizationJournal of
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
CombinatoricsHindawi Publishing Corporationhttpwwwhindawicom Volume 2014
International Journal of
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Operations ResearchAdvances in
Journal of
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Function Spaces
Abstract and Applied AnalysisHindawi Publishing Corporationhttpwwwhindawicom Volume 2014
International Journal of Mathematics and Mathematical Sciences
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
The Scientific World JournalHindawi Publishing Corporation httpwwwhindawicom Volume 2014
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Algebra
Discrete Dynamics in Nature and Society
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Decision SciencesAdvances in
Discrete MathematicsJournal of
Hindawi Publishing Corporationhttpwwwhindawicom
Volume 2014 Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Stochastic AnalysisInternational Journal of
Mathematical Problems in Engineering 13
Susceptible agent ImmuneYes
Healthy and immune
No
Choose schedule
Self-isolation Go to hospital
Staysymptomatic
No
Receive treatment
Contact withinfected agents
Symptomatic
No
Self-immunoprocess
Yes
Recovered
Yes
Incubation
Infection
No
Yes No
NoYes
Figure 12 The workflow of a susceptible agent
168 and 85 [39] So it is possible to apply normal randomvariable to model the contact frequency In our work Box-Muller method [40] is used to generate the random variableof contact frequency shown in the following
119865119862= 120583119865+ 120590119865(minus2 ln (120574
1))12 cos (2120587120574
2) (1)
in which 119865119862is the random variable of contact frequency 120583
119865
is the mean value of normal distribution 120590119865is the standard
deviation of normal distribution and 1205741and 120574
2are the
uniform random variables distributed in the interval [0 1]Based on (1) and survey data [41] the contact frequenciesof agent 119865contact(119860 119894) are discretized as in Table 3 within theconsideration of activity differences
Similarly the duration per contact between individualsis another key factor of AHC transmission The duration ofcontact could also bemodeled by the normal randomvariablein the following [42]
119879119862= 120583119879+ 120590119879(minus2 ln (120574
1))12 cos (2120587120574
2) (2)
in which 119879119862is the random variable of duration per contact
120583119879
is the mean value of normal distribution 120590119879
is thestandard deviation of normal distribution and 120574
1and 120574
2
are the uniform random variables distributed in the interval
[0 1] In addition because the time spent in specific location119879Location(119860 119894) also follows the random distribution like (3)in our work In order to make the duration of contact119879contact(119860 119894) shorter than the 119879Location(119860 119894) the mean andstandard deviation of (2) is set up in (3) and (4) [42]
120583119879=119879Location(120583119865+ 120590119865) (3)
120590119879=119879Location(10120583119865+ 10120590
119865) (4)
in which 119879Location is the time of an activity in a specificlocation set in Table 2 120583
119879and 120590
119879are the mean and standard
deviation of contact frequency As a result the duration ofagent 119879contact(119860 119894) in specific locations (listed in Table 2) isalso discretized in Table 3 within the consideration of activitydifferences
414 Modelling Transportation In artificial Beijing threekinds of travel models are considered walking inside thedistrict travelling by road networks and travelling by subwaynetworks According to the traveling models the agentmovements are implemented by the compositions of travelingmodels before So it is important to build the basic road
14 Mathematical Problems in Engineering
Table 3 The contact frequency and duration in different locations and activities by social relationships in case of student and teacher
Social relationship Activity Location 119879contact(119860 119894) (minute) 119865contact(119860 119894)
Student-student
Sleep Dormitory 0 0
Breakfast Restaurant 119873(52) 119873(32)Breakfast Dormitory 119873(52) 119873(21)Sports Playground 119873(21) 119873(53)Class Classroom 119873(21) 119873(21)Study Library 119873(21) 119873(21)Lunch Restaurant 119873(158) 119873(32)
Shopping Convenience Store 119873(203) 119873(21)Dinner Restaurant 119873(108) 119873(21)Dinner Dormitory 119873(158) 119873(32)Rest Dormitory 119873(63) 119873(52)
Teacher-student Class Classroom 119873(21) 119873(21)
Subway network of Beijing city (red lines) Road network of Beijing city (green lines)
Figure 13 The implementation of transportation for artificial Beijing
and subway networks With the help of Google map thekey points of road are sampled first and the road networksare generated by the links among point Similarly subwaynetworks are generated by the links between subway stationsmapped from the real stations in Google map Figure 13shows the road and subway networks in artificial Beijingwith the road networks in green and the subway networks inred Most of the districts in Beijing city are covered by thesetwo networks The algorithm of hierarchical route planningis proposed as follows
(i) Find the next position where agent will be located innext action
(ii) Obtain the travelingmodels from the statistical trans-portation data
(iii) Search the path from the starting position to thenearest road entry or subway stations
(iv) Search the optimal path from the starting station orroad entry to the nearest station or road exit of targetposition
(v) Search the path from the target subway station or roadexit to the target position
(vi) Attach the subway train number or the bus numberincluding the transfer information to the path accord-ing to the timetable of subway or bus
(vii) Connect all the paths obtained before Generate thepath from the starting position to the target position
Based on the algorithm for each agent in artificial Beijingthe transportation is simulated during the computationalexperiments It is worth noting that the activities insidethe buses or train cars are considered because the agentcommunication during travelling is also necessary to besimulated
Mathematical Problems in Engineering 15
(a) The implementation of agent in MDT (b) The code framework of agent
Figure 14 The agent model implementation in MDT [35]
415 The Implementation of Models by MDT All the modelsare developed in the MDT Figure 14 shows the GUI of thetool MDT is used to implement the state space and statetransitions of FSM models The code framework is shownin Figure 14(b) Using the C++ inheritance features and thetechniques of dynamic link library (dll) the models areencapsulated in the dll componentsThe flexible compositionmechanism is designed to support the models evolution incomputational experiments
416 The Description of Artificial Beijing by ASE As men-tioned in Section 34 ASE describes the artificial Beijingin macroview The statistical information of Beijing city iscollected by ASE Table 4 lists part of the description ofartificial Beijing generated by ASE
42 How to Obtain the Data of Beijing City
421 Population and Geospatial Environment Database ofBeijing City Based on the models developed for ArtificialBeijing it is necessary to generate the initial data to do thecomputational experiments Because the individual level dataare not available it is necessary to construct an individual-based population database for both accurate computationalexperiment and determining optimal decisions
According to the state spaces of models discussed inSection 412 the geospatial and population database of Arti-ficial Beijing is designed as shown in Figure 15 The kernelpart of database consists of the tables such as agent list tablegeospatial environment list table household list table andagent distribution table Agent list table is an individual leveltable used to store the data to initialize the state space of agentmodel such as id and gender age Geospatial environmentlist table is also an individual level table which stores thedata of environment attributes such as id street id andtype Household list table and agent distribution table storethe statistical data for the data generation for agent andenvironment models With the help of the geospatial andpopulation database the data generating of artificial Beijingis proposed in the next section
422 Generating Geospatial and Social Environment Datafor Artificial Beijing by PGET The algorithm of generatinggeospatial and social environment data for artificial Beijingis used to quantify the spatial distribution of population andformalize geospatial behavior of each agent The algorithmshown in Figure 16 is capable of generating a syntheticartificial society which allowsmultiresolution statistical datasocial interactive behavior and multilayer social networksto be integrated together The synthetic population canrepresent individual agents in the form of households andhousehold members and the synthetic population is statis-tically equivalent to a real population For each householdcharacteristics such as address household size family typesand relationships are generated Each person is described bycharacteristics such as age gender social role and correlatedlocations The algorithm provides an effective methodologyto reconstruct the computing environment in high resolutionby using statistical data in low resolution leading to betterprediction and management of emergencies The details ofalgorithm are illustrated in [36]
As shown in Figure 17 the implementation of the algo-rithm is embedded inside the GEPTThe collective input datasuch as the data files of population distribution is collected inthe tool first then the algorithm is activated to generate thedatabase The generation lasts almost twelve hours one timewith the specified parameters settings When the generationis finished the data of artificial Beijing is obtained in thedatabase with 19610000 agents and 16000 environments
5 H1N1 Experiments in ArtificialBeijing with KD-ACP
With the support of the artificial Beijing modeled before thecomputational experiments and parallel execution are alsoimplemented by KD-ACP The experiments are designed tofind the most optimal emergency decision response plansEpidemic in city is a typical scenario in emergency manage-ment H1N1 epidemic in Beijing in 2009 is used to be the usedcase to test KD-ACP According to Section 34 it is necessarytomodel emergencymanagement first in order to support thecomputational experiments
16 Mathematical Problems in Engineering
Table 4 The description of artificial Beijing generated by ASE
Statistical features ofartificial Beijing Descriptions of statistical features Data sourcedefault values
Agent roles Baby child primary student middle school student college student worker retired
Demographic data file ofBeijing
Environment types Super market store restaurant park stadium hospital school cliniccommercial building dormitory building residential building
Demographic data file ofBeijing
Demographic data Age distribution of population sex distribution of population householddistribution of population retired age distribution children distribution
Demographic data file ofBeijing
Environment data Distribution of all types of environments Environment distributiondata file of Beijing
Relationships Classmate relative college friendship family Demographic data file ofBeijing
Varying parameters
Age difference of couples119873(120583120590) 119873(03) isin [0 12]Minimummarriage age (male female) (22 20)Age difference of children119873(120583120590) 119873(13) isin [1 10]Enrollment rate of kindergarten 120572 08Enrollment rate of primary school 120573 1Enrollment rate of middle school 120574 1Enrollment rate of college 120575 08Entrance age of kindergarten 1205721015840 3Entrance age of primary school 1205731015840 7Entrance age of middle school 1205741015840 13Entrance age of middle school 1205751015840 19
Age distributioninformation table
Agent settable
Geospatialenvironment set table
Household settable
Figure 15 The geospatial and population database of Beijing city
51 Modelling Emergency Management The modelling ofemergency management consists of two parts the modellingof emergent events and the modelling of intervention mea-sures H1N1 and the intervention measures for H1N1 aremodeled in our case
511 Modelling H1N1 by MDT H1N1 Model in agent simu-lates the states transitions of health status and the relevantactions Referring to SIR (SusceptibleInfectiveRecovered)[43 44] models agent has three health statuses susceptibleinfected and healthy with immunity Only susceptible agentcan be infected by the contact with infected agents After
the three infected phases (incubation being symptomaticand recovery) of H1N1 model agent is healthy again It isworthy mentioning that the recovered agents can be infectediteratively when the immunity disappears gradually by timeThe state transitions are illustrated in Figure 18
According to [45] the latent period of H1N1 was in aWeibull distribution within people from one to seven days[46 47] The distribution was usually centered in the rangebetween one and three days AWeibull random variable [40]is used to model the latent period as described in
119879Lat = 120572 times [minus ln (1 minus 120574)]1120573+ 120592 (119879Lat ge 120592) (5)
Mathematical Problems in Engineering 17
Censusdata
Generate GUID
Assign individualhousehold
Generate familystructure
Assign family roleto every member
Endow individualwith age gender
social role
Distribute environmentlocations on the map
School Workplace Entertainmentlocations
Consumptionlocations
Assign school orworkplace to
individual
Correlatedindividual withother locations
Generate friendshipin school
Generate coworkerrelations inworkplace
Assign individualadministrative
region
Generate housebuildings
Locate householdto house building
Generate familyrelation
Generateneighborhood
relation
PopulationEnvironment
Environment listSocial network
Figure 16 The algorithm of generating geospatial and social environment data [36]
Edit
Edit
Edit
Edit
Edit
Edit
BrowseStatistical characteristic of population
Geospatial population data
Population data of street
Population types
Geospatial and social environment generation tool
Workspaces
Parameters distribution settings
Household types
Roles in household
Household structure
Social roles
BrowseBrowseBrowseBrowseBrowseBrowse
Population distribution data
File path
Retired distributionPop distributionHousehold numHousehold distributionSex distributionAge distributionData type
Figure 17 The GUI of Population and Geospatial Environment generation Tool (PGET)
18 Mathematical Problems in Engineering
Infected
SusceptibleHealth
withimmunity
Infected
by contac
t
with in
fected
agents
Recovery
Increase before beingsymptomatic symptomatic
Decrease after being
Decrease by time
Infectivity
Immunity
Figure 18 The transitions of agent health status
119879Lat denotes the duration of latent period 120592 120572 and 120573 arethe location parameter scale parameter and shape parameterof Weibull distribution respectively 120574 is a uniform randomnumber in the range [0 1] According to statistics [47] 120592 120572and 120573 are set as 0 18 and 121 respectively Then the mean(standard deviation) of latent period is calculated as 159 days(058 day2) Furthermore infectious period is set as 7 days
A susceptible studentrsquos probability of getting infecteddepended on the infectivity of infectious agents his ownimmunity level the duration of the contact action and soon The infectivity of an infected agent evolved with timeduring the infectious period According to the statistics ofinfectivity in the chart with red bars in Figure 19 an infectedstudent had the highest level of infectivity in the second dayafter he had the first symptom The infectivity levels in otherdays in comparison with the second day are in Table 5 Theday labeled as ldquominus1rdquo means the day before the starting of thesymptomatic period
Based on the algorithm discussed before theH1N1modelis developed with the help of MDT The settings of influenzamodel will be discussed in Section 521
512 Modelling Intervention Measures by MDT In corre-spondence intervention measures model is used to ceasethe outbreak of H1N1 As discussed in [42 48 49] if theappropriate measures are taken when the infectious diseaseemerges the transmission of the disease could be sloweddown and the damage could be decreased Therefore it isnecessary and important to design emergency interventionmeasures in the artificial Beijing According to the MinistryofHealth inChina the intervention policies [50] are designedto control the spread of influenza such as H1N1 The inter-ventionmeasures of these policies including are interventionsactivated time vaccination rate antibiotic delay to hospitalisolation duration close workspace duration and limitationof activities listed in Table 6 Similarly the models of theseintervention measures are also developed with the help ofMDT Using the reasonable data ranges listed in Table 6intervention models can be initialized to do the computa-tional experiments These experiments are used to show how
Symptomatic1 2 3 4 5 6 7minus1
Recovered
Tim
e (da
y)
Infe
ctiv
ity
Figure 19 The distribution of infectivity rate of an infected agent[37]
Table 5 The proportion of infectivity
Days (119894) minus1 1 2 3 4 5 6 6Proportion of infectivity 03 05 1 08 04 02 01 005
to restrain the H1N1 transmissionThe settings of these inter-vention measures in IMCT are introduced in Section 522
52 Experiment Settings
521 H1N1 Settings by EECT EECT is designed to initializethe emergent events models such as H1N1The settings comefrom the specified emergent event models As shown inFigure 20 in theH1N1 used case the settings are composed ofthe distribution of infection source infection rate infectionperiod and so on The infection rate changes with envi-ronments and infection periods The settings are initializedfrom both the research of medicine on H1N1 and the data ofH1N1 outbreak in Beijing in 2009The process of influenza inartificial Beijing is simulated to repeat the possible scenariosof H1N1 break During the computational experiments ofartificial Beijing theH1N1 settings such as the period of beingsymptomatic are updated by the dynamic data registrationsfrom OsdRT The updating corrects the H1N1 model duringthe experiments With the help of OsdRT artificial Beijing isable to approach actual society It is also an implementationof parallel execution
522 Intervention Measures Settings by IMCT In the H1N1case the settings of intervention measures are shown inFigure 21 in IMCT These measures are designed under theinstruction of Emergency Decision Organizations like ChinaDisease Center (CDC) The measures such as vaccinationrate the rate of closedworkspace and the starting time can beinitialized in the tool Emergency response plans are imple-mented in these settings As illustrated in Figure 7 all the pos-sible compositions within value settings in reasonable inter-vals are tested to find the most optimal emergency responseplan For example different compositions of measures suchas vaccination rates and rates of workspace closed are usedto design the plans of experiments By the analysis of theresults the best composition of interventionmeasures can beobtainedThe analysis will be discussed in case study in detail
Mathematical Problems in Engineering 19
Table 6 The intervention measures of influenza
Intervention measures ValuesInterventions activated Time 119879Intervention isin [0 80th day]Vaccination rate 119877Vaccination isin [0 1]Antibiotic rate 119877Antibiotic isin [0 1]Delay to hospital 119867ospital(119860 119894) 119879infected lt 119905 lt 119879recovered + 119879HDelay 119879HDelay isin (0 3th day]Isolation duration 119868solation(119860 119894) 119879Intervention lt 119905 lt 119879Intervention + 119879IDuration 119879IDuration isin (0 14th day]Close workspace duration 119862lose(119864119894) 119879Intervention lt 119905 lt 119879Intervention + 119879CWDuration 119864119894 isinWorkspace 119879CWDuration isin (0 7th day]Limitation of agent activity duration 119871 imitation(119860 119894) 119879Intervention lt 119905 lt 119879Intervention + 119879LAADuration 119879LAADuration isin (0 7th day]119877Vaccination means the vaccination rate of artificial Beijing while 119877Antibiotic means the antibiotic rate used for infected agents in hospital119867ospital (119860119894) means that the infected agent 119860119894 is sent to hospital he or she will receive the retreatment immediately119868solation (119860119894) means that the infected agent 119860119894 is isolated at home he or she can only contact with the other agents until isolation duration is over119862lose (119864119894) means that the environment 119864119894 typed of workspace is closed until the close workspace duration is over119871 imitation (119860119894) means that the activities of agent 119860119894 are limited only the workspace and home are allowed until limitation duration is over
Incubation Weibulldistribution
Symbolic LogNormaldistribution
Recovering LogNormaldistribution
H1N1influenza
Settings of H1N1 influenza model
Disease model
Transmissionmodel
Figure 20 The settings of H1N1 in EECT
It is worth notifying that the settings of interventionmeasures are also influenced by OsdRT For example theH1N1 outbreak areas are notified by OsdRTThe interventionmeasures such as close workspace are used only in theoutbreak areas Therefore the unnecessary computation ofintervention measures is avoided So the data injection byOsdRT not only makes artificial society approach actualsociety but also increases the performance of computationalexperiments
53 Experiments by EMT When the initialization of arti-ficial Beijing is ready EMT is used to do the computa-tional experiments EMT is composed of two parts thecentral controller in the central node and the residen-tial service deployed in the computational nodes Centralcontroller loads the experiment plans and controls the whole
experiments process the GUI of controller is shown inFigure 22 EMT manages the computational experimentsin four steps cluster configuration load experiment plansmodels deployment and running experiments
531 Cluster Configuration Before the configuration clusterinformation is collected from the residential services insidethe cluster The information includes nodes amount nodenames node IPs and CPU occupations It is used to quantifythe computing power of the clusterThe cluster configurationis based on this information The nodes are selected forexperiments Then the LP (local process) [51] number foreach node is set according to the number of the CPU kernelsAfter the configuration the experiment plans are loaded tocustomize the models and data
20 Mathematical Problems in Engineering
Streets of close workspaces
Limitations ofactivities
IsolationsRespirator
Closeworkspace
Antibiotic
VaccinationHospital
Surveillance
Figure 21 The settings of intervention measures in IMCT
Information ofexperiment plan
List of experiment plans
Start | Pause | Reset | Stop | Speedup | Slowdown | Back
NodeName | CPU occupation | Memory occupation |Network | Models | Status
Figure 22 The deployment and running of experiments in EMT
532 Load Experiment Plan As discussed in Section 35 thedescription of experiment plans is composed of the infor-mation in three aspects models and the initial data for theexperiments the deployment mapping tables from the mod-els to the nodes and the plan of experiment executionWhenthe experiment plan is loaded by the EMT the file namesand model descriptions of models and data are listed Withthe support of cluster configuration models can be mappedto the nodes in the GUI according to the computationrequirements EMT also provides the setting of experimentsitself including the running times start time and end time
533 Models and Data Deployment Models are stored inthe agent population and services repository According tothe settings in experiment plan the customized models aredownloaded from the repository first then the models suchas agent model environment model emergent events modeland intervention measures model are integrated to be thecompositionmodels of artificial BeijingThemodels and datalocations in the nodes are configured in the ldquomodel pathrdquoand ldquodata pathrdquo in the GUI of central controller Finallythe models are uploaded to the nodes in cluster with theconsideration of partition The partition of artificial society
Mathematical Problems in Engineering 21
models is different from the multiagent system in [52]Models in artificial society are divided into three levels agentsand environments grids and city Agents and environmentsare encapsulated in grids which compose the whole cityThus the partition of artificial society becomes the partitionof grids The parallelism degrees are found to support thepartition of models agent independence and grid indepen-dence According to the parallelisms agents are partitioned toLPs according to the population distribution in grids Agentmovements among grids are simulated by the movementevents among grids in different LPs As a result a two-tierparallel architecture is proposed to support the simulation ofartificial society The architecture is detailed in [53]
534 Running of Experiments When the preparation isready the experiments can be done by the central controllerof EMT The running and control commands of the experi-ments are grouped into three aspects start pause or stop theexperiments get the running information of the nodes suchas the snapshot of the desktop of node With the help of thesecommands users can run the computational experimentsto do the research on the emergency management It isworth notifying that the simulation engine is optimized inorder to support the ten million agent simulation UsingGPU as coprocessor a two-tier parallel simulation engine isdesigned with support of MPI and OpenCL through phasedsynchronization [54] One-sided communication is used forreflection of remote simulation objects and message passingbetween processes A general kernel function prototype iselaborately designed and conditionally compiled for execu-tion on both CPU and GPU Moreover the optimizationoperations like load balance is developed under the instruc-tion of activity-based simulation [55] The densely populatedgrids are given more CPU and memory resource accordingto the activity predictions
54 Results of H1N1 Experiments with KD-ACP On thebasis of artificial society models and the intervention mea-sures mentioned in Section 512 a series of computationalexperiments are performed to study the H1N1 influenza inBeijing In our case of artificial Beijing 19610000 agentsare simulated in the cluster within 48 CPU cores and 128Gmemories the nodes are connected by kilomega networks Ittakes 18 hours to simulate a 250-day disease spreading
In order to illustrate how the optimal plans are obtainedbyKD-ACP the experiments are divided into four groups theexperiments ofmodel validation the experiments of sensitiv-ity analysis of vaccination rate the experiments of sensitivityanalysis of isolation and the experiments of the combinationof interventionmeasuresThese four groups show the generalprocess of the research on emergency management in H1N1influenza Firstly the models of H1N1 are validated withthe support of historical data Secondly the traditionalmedical interventions of pandemic such as vaccination areanalyzed to find the optimal plan Thirdly the traditionalnonmedical interventions of pandemic such as isolation areanalyzed to find the optimal plan Finally the combination ofinterventions is tested to show the combined effect
It is worth mentioning that all the experiments in ourwork are performed 100 times These experiments are ini-tialized with different settings of interventions in IMCT Theresults shown in the figures are the mean values
541 The Experiments of Model Validation In our artifi-cial Beijing the models of H1N1 influenza are built basedon the existing researches Wang and her colleagues hadstudied the H1N1 influenza by SEIR (SusceptibleExposedInfectiveRecovered) models [56] With the support of theirwork we build the models of artificial Beijing in the mannerof multiagents
In Figure 23 the control group shown in red line isdrawn based on the realistic statistical data collected in H1N1influenza in Beijing in 2009 The influenza lasts more thansix months and more than 170000 people are infectedComparing with the control group the simulated resultswithout interventions are drawn in blue It is obvious that thesimulated case fits the control group well before 70 days Butthe ascending of control group is slowed after 70 days andthe peak value (17883) is much less than the simulated results(25135) It is because the vaccination is activated at day 67 incontrol group according to the intervention measures usedin 2009 The restrain by vaccination becomes effective eightdays later The change of infected number after vaccinationis shown by red spotted line The trend of control group isdifferent from the simulation data after 75 days
However the simulated results (blue line) in compu-tational experiments fit the control group (as shown inFigure 23) in the first 70 days when no intervention mea-sures are executed The key features such as effective basicreproduction number (119877
0) are also almost the same in both
cases Therefore it can be concluded that the models ofartificial society built in microview are validated by therealistic statistical data in influenza in Beijing in 2009
In this group of experiments no interventions are exe-cuted in order to validate the models Therefore the simu-lated results of this extreme case which cannot happen todaycould be obtained As shown by blue line the number ofinfected agents without interventions reaches peak value atday 91 with 25135 consisting of 8052 in symptomatic phaseand 28950 in recovery phase It is obvious that the totalnumber of infected agents grows slowly at the beginningtime Then the number increases quickly when the influenzabreaks and it reaches its maximum value with 383870 in theend This phenomenon can be explained as in the followingAt the beginning of influenza due to the limited numberof infected agents the propagation of the disease remainsslow but many agents become susceptible When the numberof infected agents increases rapidly the spreading becomessignificant However the number of infected agents decreasesafter day 91 It is because the disease spreading relies on thesocial networks and spatial contact networks [49] It meansthat the infected agents can only infect agents either in thesocial relationship networks or in the spatially contacted
542 The Experiments of Sensitivity Analysis of VaccinationRate According to Section 512 vaccination is a typical
22 Mathematical Problems in Engineering
0
Infe
cted
agen
ts
0 50 100 150 200 250
Statistical dataSimulation data
25000
20000
15000
10000
5000
Day d
Figure 23 The results with no intervention executed and the com-parison with realistic statistical data in Beijing in 2009
medical intervention in influenza Traditionally the effective-ness of vaccination can only be evaluated after the influenzaand the key settings such as vaccination rate are determinedby the experience It is always difficult for emergency man-agement organizations to find the most optimal vaccinationrate in advanceTherefore the sensitivity of vaccination rate isanalyzed in this group of experiments The settings are listedin Table 7 average results are shown in Figure 24
Vaccination rates are set linearly (10 30 50 and70) to test the changes of influenza The peak value ofinfected agents decreases from 25135 22392 15787 6907to 638 respectively It shows that the more the agents arevaccinated the smaller the peak value of the infected agentswill be In the case of 70 vaccination the total infected agentis only 13799 It means that the influenza in this case does nothappen due to the high vaccination rate
The change of peak values also shows an interestingphenomenon Vaccination rates increase linearly the peakvalues decrease slower than the ascending of vaccination ratesat beginning but the descending becomes more and morefaster with the linear increase of vaccination rate It meansthat the small scale of vaccination (less than 30) does notmake senseWith the increase of vaccination rate the numberof susceptible agents decreases significantly In sequencethe possibility of infections is lowered greatly Additionallyit could be found that the decrease of peak values comeslater than the increase of the vaccination rates This is quitereasonable according to Mei et alrsquos work in [57]
As a result it is concluded that high vaccination rate iseffective in the control of H1N1 influenza though costly
543 The Experiments of Sensitivity Analysis of IsolationAccording to Section 512 isolation is a typical nonmedicalintervention It is the most used intervention measure in
Table 7 The initial settings of experiments of vaccination
Parameters ValueAgent count 19610000Initial infected count 40Interventions activated time(119879Intervention)
0
Vaccination rate (119877Vaccination) 0 10 30 50 70
0
Infe
cted
agen
ts
Loading intervention of vaccination
No vaccination10 vaccination30 vaccination
50 vaccination70 vaccination
0 50 100 150 200 250
25000
20000
15000
10000
5000
Day d
Figure 24 The results of loading intervention of vaccination
the emergency management in influenza As mentionedbefore a problem also exists that the measure cannot bevalidated before the intervention Therefore the sensitivityof isolation is analyzed in this group of experiments It isworth mentioning that the isolation is usually combinedwith the hospital policy It means the infected individual issent to hospital meanwhile the people who contact withthe infected are all isolated As a result parameters delay tohospital and isolation duration are designed together in thesettings of experiment Delay to hospital stands for the periodfrom infection time to hospital time while Isolation durationrepresents the period when an agent cannot contact others ifhe has contact with an infected agent The settings of theseparameters are listed in Table 8 average results are shown inFigure 25
Firstly the parameter delay to hospital is simulated aloneto test the sensitivity The simulated results are shown insolid line in Figure 25 The peak value decreases from 6708to 2218 when the delay to hospital decreases from 15 daysto 1 day The half day change brings the 67 reductionMoreover the peak value almost approximates to zero inthe case of 05 days The influenza does not happen in thiscase It can be concluded that the earlier the infected agents
Mathematical Problems in Engineering 23
Table 8 The initial settings of experiments of isolation
Parameters ValueAgent count 19610000Initial infected count 40Interventions activated time (119879Intervention) 80Delay to hospital (119879HDelay) isolation duration (119879IDuration) (15 0) (1 0) (05 0) (15 3) (1 3) (05 3)
0 50 100 150 200 2500
1000
2000
3000
4000
5000
6000
7000
Infe
cted
agen
ts
Loading intervention of isolation
Delay to hospital = 15
Delay to hospital = 10
Delay to hospital = 05
Delay to hospital = 15 isolation duration = 3
Delay to hospital = 10 isolation duration = 3
Delay to hospital = 05 isolation duration = 3
Day d
Figure 25 The results of loading intervention of isolation
are sent to hospital the less the agents will be infectedSo the construction of effective mechanism to find infectedindividuals is important in the emergency management ofinfluenza
Secondly the parameter isolation duration is added tosimulate the isolation of interventionObviously the isolationmeasure decreases the infected agents greatly Compared tothe parameter delay to hospital of 15-day case the peak valuedecreases from 6708 to 362when the isolation is added in theinterventionThe reduction of 6346 agents comes from the 3-day isolation of agents who had contacted with the infectedagents However the decrease of parameter delay to hospitalcannot bring the similar reduction in the 3-day isolationcases According to the figure the peak value decreases from362 to 51 from 15-day case to 05-day case only 311 agents aresaved from infection As a result the sensitivity of simulatedresults does not change obviously with delay to hospital inthe case of isolation It can be concluded that isolation is amore effective measure Though it is not easy for emergencymanagement organizations to decrease the delay to hospitaldue to the current monitoring mechanism the influenza canstill be controlled by isolation
544 The Experiments of Combination of Intervention Mea-sures Isolation is not the only nonmedical intervention ininfluenza close workspace is another common measureWhen close workspace is activated the workspaces are allclosed agents can only stay home at the closed time Luckilyisolation and close workspace are independent from oneanother It is possible for emergency management organi-zations to use these two measures together So this groupof experiments is designed to obtain the effectiveness of thecombination of intervention measures The parameters areinitialized with different settings listed in Table 9
Parameters delay to hospital and isolation duration is setin the middle values (1 day 3 days) according to the lastsection According to the experience close duration is set asseven days It is known that close workspace is a really costintervention measure the activated time is a key parameterWith the fixed parameters mentioned before the changeof interventions activated time is analyzed in this group ofcomputational experimentsThe average results are shown inFigure 26
Compared with the simulated results of experimentsdiscussed above Figure 26 strongly illustrates that influenzais ceased when the intervention measures are combinedtogether The peak value of infected agents is 141 in the ldquo80thdays of close-workspace and isolation with 1 day delay tohospitalrdquo case while 25135 in ldquono interventionrdquo case 2218 inldquo1 day delay to hospitalrdquo case 203 in ldquo1 day delay to hospitaland isolationrdquo case The interventions bring a great reductionof infection the peak values decrease from ten thousandslevel to hundred level
On the other hand the peak value even decreases to 108when the execution time of close workspace is 10 days earlierThe date of maximum value of infected agents is advancedfrom the 135th day to the 108th day The whole circle of H1N1influenza is reducedThe results of this group of experimentsshow that intervention measures should be activated as soonas possible and the compositions of interventions are moreeffective in emergency management of H1N1 influenza
In summary based on the general process lots of exper-iments could be done by KD-ACP to study emergency man-agement in H1N1 influenza Many conclusions are obtainedbased on the analysis of the experiments in this sectionAccording to these conclusions themost optimal response inour case study is the case of ldquo70th days of close-workspace andisolation with 1 day delay to hospitalrdquo It is because this casehas the fewest infected agents relatively But the modelling ofH1N1 influenza is not sufficient Economic and social factorsof interventions such as the money cost by vaccination andisolation the influence of the close-workspace in society
24 Mathematical Problems in Engineering
Table 9 The initial settings of experiments of isolation and closeworkspace
Parameters ValueAgent count 19610000Initial infected count 40Interventions activated time (119879Intervention) 70th day 80th dayDelay to hospital (119879HDelay) 1 dayIsolation duration (119879IDuration) 3 daysClose workspace duration (119879CWDuration) 7 days
50 100 150 200 2500
50
100
150
200
Infe
cted
agen
ts
Loading intervention of isolation and close workspace
Delay to hospital = 1 isolation duration = 3Start close time = 70 close duration time = 7Delay to hospital = 1 isolation duration = 3Start close time = 80 close duration time = 7
Day d
Figure 26The results of loading interventions of isolation and closeworkspace
and even opinion transmission of H1N1 in internet are notconsidered in our case study Without the analysis of thesefactors it does not make sense to talk about the most optimalintervention measures However KD-ACP had provided asoftware framework and general process of the research onemergency management Supported by KD-ACP the expertsof emergency management are able to study emergent eventssuch as H1N1 influenza as detailed as possible Based on theexperiments and analysis again and again the most optimalintervention measures will be found ultimately
55 Parallel Execution by ASST and OsdRT As discussed inSection 35 ASST andOsdRT are used to support the parallelexecution of artificial Beijing The statistical data is collectedand stored in artificial society runtime database at runtimeASST downloads the infection data from the database anddisplays it in the map As shown in Figure 27 the infectedagents are distributed in the districts in BeijingThe snapshotgives the geospatial distribution of the H1N1 epidemic Thelegends of the agents in different status are listed in the left
right corner in Figure 27 Five statuses are shown in the figurehealth without vaccination health with vaccination incuba-tion being symptomatic and convalescence ASST providesa very direct feeling for the emergency experts Partly withthe situation from ASST Emergency Decision Organizationmakes the emergency response plans to the response of theemergency scenario With the help of plans the epidemicin actual society can be restrained In the meantime theinfluenza information is reflected in the Internet Accordingto Figure 8 the data acquisition data extraction and datastandardization of H1N1 influenza are processed in OsdRTOsdRT is implemented by Dr Caorsquos team from Institute ofAutomation from Chinese Academy of Sciences [58] Thetool provides not only the reports of H1N1 SARS hand-foot-mouth disease and intestinal diseases are all monitored Theweb page is shown in Figure 28 H1N1 reports from Internetare collected first Based on the data extraction of the reportsH1N1 case of distribution is obtained and shown in theincidencemap As a result H1N1 incidence data is regarded asthe knowledge which is sent to EECT and IMCTThe settingsof emergent events and intervention measures are modifiedin order to make artificial society approach actual society innext turn of computational experiments With the help ofOsdRT the loop of parallel execution process integrates ACPapproach
6 Conclusion
This work provides an integrated software framework forsocial Computing in emergencymanagement From a systemperspective KD-ACP provides a reliable flexible low costand effective platform for the scientists to do the research onemergency response problems The analysis and predictionof these problems with inherent complexity can be solvedby the repetition of possible alternative experiments onthe software framework Therefore KD-ACP which is anattempt to implement ACP approach seemed as an effectivecomputational framework to support the decision making inemergency response
Currently KD-ACP is used to study the H1N1 epidemicin Beijing in 2009 The most optimal compositions of inter-vention measures are testified to support the response tonext influenza However it is still a long way ahead beforethe automation systematization and practicalization of KD-ACP A lot of work will be carried out along several directionsas in the following
(i) The parallel execution process in KD-ACP is actuallynot strongly connected with the emergency responseorganization Traditional experience and theory stillplay a dominant role in the decision of emergencymanagement It is necessary to propose a mechanismof hall for workshop to facilitate the parallel executionloop in actual society
(ii) Section 33 introduces the principle and process ofmodelling artificial society Metamodels and modelsof artificial Beijing are built by GME Model transfor-mation and code generation are used to implementcodes from FSM based models But many details
Mathematical Problems in Engineering 25
Fangshan district Daxing district Fengtai district Xicheng district
Mentougou district Haidian districtShijingshan district Dongcheng district
Incubation
Symptomatic
Convalescence
Health withoutvaccination
Health withvaccinationLegend of agent
health status
Figure 27 The situation of artificial Beijing by ASST
H1N1
Districts of Beijing
H1N1 case ofdistribution
Incidence map
Distribution ofH1N1 reportsfrom internet
Incidence of H1N1
Fangshan districtLiangxiang town
Population 19000Incidence 135789
(1100000)
Figure 28 The implementation of OsdRT by Institute of Automation from Chinese Academy of Sciences
are still not illustrated due to the topic of thispaper The key techniques such as template of codeframework and model transformations from specificdomains to FSM will be focused and introduced indetail in our next paper
(iii) The performance of computational experiments inKD-ACP is still slower than expectation The opti-mization of simulation engine and models structureis necessary to improve the performance So it ispossible to run artificial society with the real worldin parallel
In summary KD-ACP paves a new way for scientists inboth emergencymanagement and computation simulation to
collaborate with each other in solving the emergency man-agement problems by social computing
Research Highlights
(i) KD-ACP is implemented based on ACP approach
(ii) Artificial Beijing is built with the help of KD-ACP
(iii) H1N1 is modeled to simulate the influenza in Beijingin 2009
(iv) Computational experiments testify the effectivenessof intervention measures
26 Mathematical Problems in Engineering
Conflict of Interests
The authors declare that there is no conflict of interestsregarding the publication of this paper
Acknowledgment
This work was supported in part by the National ScienceFoundation of China (nos 91024030 71303252 61403402and 91324013)
References
[1] W Mao and F-Y Wang Advances in Intelligence and SecurityInformatics Academic Press Oxford UK 2012
[2] E Bonabeau ldquoAgent-based modeling methods and techniquesfor simulating human systemsrdquo Proceedings of the NationalAcademy of Sciences of the United States of America vol 99 no3 pp 7280ndash7287 2002
[3] K M Carley D B Fridsma E Casman et al ldquoBioWar scalableagent-basedmodel of bioattacksrdquo IEEETransactions on SystemsMan and Cybernetics A Systems and Humans vol 36 no 2 pp252ndash265 2006
[4] J M Epstein ldquoModelling to contain pandemicsrdquo Nature vol460 no 7256 p 687 2009
[5] Los Alamos National Laboratory Reports about EPISIMS LosAlamos National Laboratory 2003
[6] C H Builder and S C Bankes ldquoArtificial societies a conceptfor basic research on the societal impacts of informationtechnologyrdquo RAND Report 149 RAND Corp 1991
[7] F-Y Wang ldquoComputational theory and method on complexsystemrdquo China Basic Science vol 6 no 5 pp 3ndash10 2004
[8] F-Y Wang ldquoToward a paradigm shift in social computing theACP approachrdquo IEEE Intelligent Systems vol 22 no 5 pp 65ndash67 2007
[9] F-Y Wang and S Tang ldquoArtificial societies for integratedand sustainable development of metropolitan systemsrdquo IEEEIntelligent Systems vol 19 no 4 pp 82ndash87 2004
[10] D Wen Y Yuan and X-R Li ldquoArtificial societies computa-tional experiments and parallel systems an investigation ona computational theory for complex socioeconomic systemsrdquoIEEE Transactions on Services Computing vol 6 no 2 pp 177ndash185 2013
[11] F Wang F Zeng and Y Yuan ldquoAn ACP-based approach forcomplexity analysis of E-commerce systemrdquo Complex Systemsand Complexity Science vol 5 no 3 pp 1ndash8 2008 (Chinese)
[12] F-Y Wang and P K Wong ldquoResearch commentary intelligentsystems and technology for integrative and predictivemedicinean ACP approachrdquoACMTransactions on Intelligent Systems andTechnology vol 4 no 2 pp 1ndash6 2013
[13] J Sifeng X Gang F Dong and H Chunpeng ldquoStudy on theemergency rescue decision support system of petrochemicalplant based on ACP theoryrdquo in Proceedings of the 6th Manage-ment Annual Meeting (MAM rsquo11) 2011 (Chinese)
[14] X Dong D Fan G Xiong F Zhu Z Zhang and Y Yao ldquoPar-allel Bus Rapid Transit (BRT) operation management systembased on ACP approachrdquo in Proceedings of the 6th ManagementAnnual Meeting (MAM rsquo11) Beijing China 2011
[15] B Ning F-Y Wang H-R Dong R-M Li D Wen and L LildquoParallel systems for Urban rail transportation based on ACP
approachrdquo Journal of Transportation Systems Engineering andInformation Technology vol 10 no 6 pp 23ndash28 2010 (Chinese)
[16] F-Y Wang J Zhao and S-X Lun ldquoArtificial power systemsfor the operation and management of complex power gridsrdquoSouthern Power System Technology vol 2 no 3 pp 1ndash11 2008(Chinese)
[17] M J North T R Howe N T Collier and J R Vos ldquoRepastsimphony development environmentrdquo in Proceedings of theAgent 2005 Conference on Generative Social Processes Modelsand Mechanisms 2005
[18] M J North T RHowe N T Collier and J R Vos ldquoRepast Sim-phony runtime systemrdquo in Proceedings of the Agent Conferenceon Generative Social Processes Models and Mechanisms 2005
[19] SWARM Development Group Swarm 211 Reference Guide2000 httpwwwswarmorg
[20] S Tisue and U Wilensky ldquoNetLogo a simple environment formodeling complexityrdquo in International Conference on ComplexSystems pp 16ndash21 2004
[21] S Luke C Cioffi-Revilla L Panait K Sullivan and G BalanldquoMASON a multiagent simulation environmentrdquo Simulationvol 81 no 7 pp 517ndash527 2005
[22] M Ichikawa H Tanuma Y Koyama andH Deguchi ldquoSOARSintroduction as a social microscope for simulations of socialinteractions and gamingrdquo in Organizing and Learning throughGaming and Simulation Proceedings of the 38th Conference ofthe International Simulation and Gaming Association (ISAGArsquo07) pp 149ndash158 2007
[23] D S Burke J M Epstein D A T Cummings et al ldquoIndividual-based computational modeling of smallpox epidemic controlstrategiesrdquo Academic Emergency Medicine vol 13 no 11 pp1142ndash1149 2006
[24] F-Y Wang ldquoParallel control and management for intelligenttransportation systems concepts architectures and applica-tionsrdquo IEEE Transactions on Intelligent Transportation Systemsvol 11 no 3 pp 630ndash638 2010
[25] Z Molnar D Balasubramanian and A Ledeczi ldquoAn introduc-tion to the genericmodeling environmentrdquo in Proceedings of theTOOLS Europe 2007 Workshop on Model-Driven DevelopmentTool Implementers Forum Zurich Switzerland 2007
[26] S Kelly and J-P Tolvanen Domain Framework in Domain-Specific Modeling Enabling Full Code Generation JohnWiley ampSons Hoboken NJ USA 2007
[27] B P Zeigler H Praehofer and T G Kim Theory of Modelingand Simulation Integrating Discrete Event and ContinuousComplex Dynamic Systems Academic Press San Diego CalifUSA 2000
[28] G Booch J Rumbaugh and I JacobsonThe Unified ModelingLanguage User Guide Addison-Wesley Reading Mass USA1999
[29] F Wagner R Schmuki T Wagne and P Wolstenholme Mod-eling Software with Finite State Machines A Practical ApproachCRC amp Taylor amp Francis Boca Raton Fla USA 2006
[30] T Espiner ldquoChina builds worldrsquos fastest supercomputerrdquo ZDNetUK October 2010
[31] B Chen and G Guo ldquoA two-tier parallel architecture for arti-ficial society simulationrdquo in Proceedings of the ACMIEEESCS26th Workshop on Principles of Advanced and DistributedSimulation (PADS rsquo12) pp 184ndash186 July 2012
[32] G Guo B Chen X G Qiu and Z Li ldquoParallel simulation oflarge-scale artificial society on CPUGPU mixed architecturerdquo
Mathematical Problems in Engineering 27
in Proceedings of the ACMIEEESCS 26th Workshop on Prin-ciples of Advanced and Distributed Simulation (PADS rsquo12) pp174ndash177 July 2012
[33] S-X Lun ldquoResearch on the classification of parallel executionmodes of ACP theoryrdquo Acta Automatica Sinica vol 38 no 10pp 1602ndash1608 2012
[34] F-YWang ldquoA framework for social signal processing and anal-ysis from social sensing networks to computational dialecticalanalyticsrdquo Science China vol 43 no 12 pp 1598ndash1611 2013
[35] G GuoOne Model Userrsquos Manual version 12 National Univer-sity of Defense University 2011
[36] Y Ge R Meng Z Cao X Qiu and K Huang ldquoVirtual citymdashanindividual-based digital environment for human mobility andinteractive behaviorrdquo Simulation Transactions of the Society forthe Modeling and Simulation International 2014
[37] M Kretzschmar and R T Mikolajczyk ldquoContact profiles ineight European countries and implications for modelling thespread of airborne infectious diseasesrdquo PLoS ONE vol 4 no6 Article ID e5931 2009
[38] W Duan and X Qiu ldquoFostering artificial societies using sociallearning and social control in parallel emergency managementsystemsrdquo Frontiers in Computer Science vol 6 no 5 pp 604ndash610 2012
[39] W J Edmunds C J OrsquoCallaghan and D J Nokes ldquoWhomixes with whom Amethod to determine the contact patternsof adults that may lead to the spread of airborne infectionsrdquoProceedings of the Royal Society B Biological Sciences vol 264no 1384 pp 949ndash957 1997
[40] J Banks J S Carson B L Neison and D M Nicol Discrete-Event System Simulation Prentice Hall Englewood Cliffs NJUSA 4th edition 2007
[41] H Yongxia ldquoInvestigation and analysis of college studentsrsquo workand restrdquo Value Engineering vol 31 no 3 2012 (Chinese)
[42] W Duan Z Cao Y Wang et al ldquoAn ACP approach to publichealth emergency management using a campus outbreak ofh1n1 influenza as a case studyrdquo IEEE Transactions on SystemsMan and Cybernetics Part A Systems and Humans vol 43 no5 pp 1028ndash1041 2013
[43] HWHethcote ldquoThemathematics of infectious diseasesrdquo SIAMReview vol 42 no 4 pp 599ndash653 2000
[44] G Chowell E Shim F Brauer and et al ldquoModelling thetransmission dynamics of acute haemorrhagic conjunctivitisapplication to the 2003 outbreak in Mexicordquo Statistics inMedicine vol 25 no 11 pp 1840ndash1857 2006
[45] Y Ge L Liu B Chen X Qiu and K Huang ldquoAgent-basedmodeling for InfluenzaH1N1 in an artificial classroomrdquo SystemsEngineering Procedia vol 2 pp 94ndash104 2011
[46] A R Tuite A L Greer MWhelan et al ldquoEstimated epidemio-logic parameters andmorbidity associatedwith pandemicH1N1influenzardquo CMAJ vol 182 no 2 pp 131ndash136 2010
[47] L Brouwers B Cakici M Camitz A Tegnell and MBoman ldquoEconomic consequences to society of pandemic H1N1influenza 2009mdashpreliminary results for Swedenrdquo Euro Surveil-lance vol 14 no 37 pp 1ndash7 2009
[48] Y Ge L Liu X Qiu H Song Y Wang and K Huang ldquoAframework of multilayer social networks for communicationbehavior with agent-based modelingrdquo Simulation Transactionsof the Society for the Modeling and Simulation International vol89 no 7 pp 810ndash828 2013
[49] W Duan X Qiu Z Cao X Zheng and K Cui ldquoHeterogeneousand stochastic agent-based models for analyzing infectious
diseasesrsquo super spreadersrdquo IEEE Intelligent Systems vol 28 no4 pp 18ndash25 2013
[50] The Preparation and Emergency Response Plans to PandemicPublished by Ministry of Health Ministry of Health 2005(Chinese)
[51] R M Fujimoto Parallel and Distributed Simulation SystemsWiley Interscience 2000
[52] F Cicirelli A Giordano and L Nigro ldquoDistributed simulationof situated multi-agent systemsrdquo in Proceedings of the 15thIEEEACM International Symposium on Distributed Simulationand Real Time Applications (DS-RT rsquo11) pp 28ndash35 September2011
[53] B Chen and G Guo ldquoA two-tier parallel architecturefor artificial society simulationrdquo in Proceedings of the 26thACMIEEESCS Workshop on Principles of Advanced and Dis-tributed Simulation (PADS rsquo12) pp 184ndash186 July 2012
[54] G Guo B Chen and X Qiu ldquoParallel simulation of large-scale artificial society with GPU as coprocessorrdquo InternationalJournal of Modeling Simulation and Scientific Computing vol4 no 2 Article ID 1350005 2013
[55] B Chen L-B Zhang X-C Liu and H Vangheluwe ldquoActivity-based simulation using DEVS increasing performance by anactivity model in parallel DEVS simulationrdquo Journal of ZhejiangUniversity Science C vol 15 no 1 pp 13ndash30 2014
[56] X-L Wang P Yang Z-D Cao et al ldquoQuantitative evaluationon the effectiveness of prevention and control measures againstpandemic influenza A (H1N1) in Beijing 2009rdquoChinese Journalof Epidemiology vol 31 no 12 pp 1374ndash1378 2010 (Chinese)
[57] S Mei A D van de Vijver L Xuan Y Zhu and P M ASloot ldquoQuantitatively evaluating interventions in the influenzaA (H1N1) epidemic on China campus grounded on individual-based simulationsrdquo Procedia Computer Science vol 1 no 1 pp1675ndash1682 2010
[58] X Qiu Z Cao and Z Li Mid-Term Reports of China NationalScience Foundation 91024030 2014
Submit your manuscripts athttpwwwhindawicom
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
MathematicsJournal of
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Mathematical Problems in Engineering
Hindawi Publishing Corporationhttpwwwhindawicom
Differential EquationsInternational Journal of
Volume 2014
Applied MathematicsJournal of
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Probability and StatisticsHindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Journal of
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Mathematical PhysicsAdvances in
Complex AnalysisJournal of
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
OptimizationJournal of
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
CombinatoricsHindawi Publishing Corporationhttpwwwhindawicom Volume 2014
International Journal of
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Operations ResearchAdvances in
Journal of
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Function Spaces
Abstract and Applied AnalysisHindawi Publishing Corporationhttpwwwhindawicom Volume 2014
International Journal of Mathematics and Mathematical Sciences
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
The Scientific World JournalHindawi Publishing Corporation httpwwwhindawicom Volume 2014
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Algebra
Discrete Dynamics in Nature and Society
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Decision SciencesAdvances in
Discrete MathematicsJournal of
Hindawi Publishing Corporationhttpwwwhindawicom
Volume 2014 Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Stochastic AnalysisInternational Journal of
14 Mathematical Problems in Engineering
Table 3 The contact frequency and duration in different locations and activities by social relationships in case of student and teacher
Social relationship Activity Location 119879contact(119860 119894) (minute) 119865contact(119860 119894)
Student-student
Sleep Dormitory 0 0
Breakfast Restaurant 119873(52) 119873(32)Breakfast Dormitory 119873(52) 119873(21)Sports Playground 119873(21) 119873(53)Class Classroom 119873(21) 119873(21)Study Library 119873(21) 119873(21)Lunch Restaurant 119873(158) 119873(32)
Shopping Convenience Store 119873(203) 119873(21)Dinner Restaurant 119873(108) 119873(21)Dinner Dormitory 119873(158) 119873(32)Rest Dormitory 119873(63) 119873(52)
Teacher-student Class Classroom 119873(21) 119873(21)
Subway network of Beijing city (red lines) Road network of Beijing city (green lines)
Figure 13 The implementation of transportation for artificial Beijing
and subway networks With the help of Google map thekey points of road are sampled first and the road networksare generated by the links among point Similarly subwaynetworks are generated by the links between subway stationsmapped from the real stations in Google map Figure 13shows the road and subway networks in artificial Beijingwith the road networks in green and the subway networks inred Most of the districts in Beijing city are covered by thesetwo networks The algorithm of hierarchical route planningis proposed as follows
(i) Find the next position where agent will be located innext action
(ii) Obtain the travelingmodels from the statistical trans-portation data
(iii) Search the path from the starting position to thenearest road entry or subway stations
(iv) Search the optimal path from the starting station orroad entry to the nearest station or road exit of targetposition
(v) Search the path from the target subway station or roadexit to the target position
(vi) Attach the subway train number or the bus numberincluding the transfer information to the path accord-ing to the timetable of subway or bus
(vii) Connect all the paths obtained before Generate thepath from the starting position to the target position
Based on the algorithm for each agent in artificial Beijingthe transportation is simulated during the computationalexperiments It is worth noting that the activities insidethe buses or train cars are considered because the agentcommunication during travelling is also necessary to besimulated
Mathematical Problems in Engineering 15
(a) The implementation of agent in MDT (b) The code framework of agent
Figure 14 The agent model implementation in MDT [35]
415 The Implementation of Models by MDT All the modelsare developed in the MDT Figure 14 shows the GUI of thetool MDT is used to implement the state space and statetransitions of FSM models The code framework is shownin Figure 14(b) Using the C++ inheritance features and thetechniques of dynamic link library (dll) the models areencapsulated in the dll componentsThe flexible compositionmechanism is designed to support the models evolution incomputational experiments
416 The Description of Artificial Beijing by ASE As men-tioned in Section 34 ASE describes the artificial Beijingin macroview The statistical information of Beijing city iscollected by ASE Table 4 lists part of the description ofartificial Beijing generated by ASE
42 How to Obtain the Data of Beijing City
421 Population and Geospatial Environment Database ofBeijing City Based on the models developed for ArtificialBeijing it is necessary to generate the initial data to do thecomputational experiments Because the individual level dataare not available it is necessary to construct an individual-based population database for both accurate computationalexperiment and determining optimal decisions
According to the state spaces of models discussed inSection 412 the geospatial and population database of Arti-ficial Beijing is designed as shown in Figure 15 The kernelpart of database consists of the tables such as agent list tablegeospatial environment list table household list table andagent distribution table Agent list table is an individual leveltable used to store the data to initialize the state space of agentmodel such as id and gender age Geospatial environmentlist table is also an individual level table which stores thedata of environment attributes such as id street id andtype Household list table and agent distribution table storethe statistical data for the data generation for agent andenvironment models With the help of the geospatial andpopulation database the data generating of artificial Beijingis proposed in the next section
422 Generating Geospatial and Social Environment Datafor Artificial Beijing by PGET The algorithm of generatinggeospatial and social environment data for artificial Beijingis used to quantify the spatial distribution of population andformalize geospatial behavior of each agent The algorithmshown in Figure 16 is capable of generating a syntheticartificial society which allowsmultiresolution statistical datasocial interactive behavior and multilayer social networksto be integrated together The synthetic population canrepresent individual agents in the form of households andhousehold members and the synthetic population is statis-tically equivalent to a real population For each householdcharacteristics such as address household size family typesand relationships are generated Each person is described bycharacteristics such as age gender social role and correlatedlocations The algorithm provides an effective methodologyto reconstruct the computing environment in high resolutionby using statistical data in low resolution leading to betterprediction and management of emergencies The details ofalgorithm are illustrated in [36]
As shown in Figure 17 the implementation of the algo-rithm is embedded inside the GEPTThe collective input datasuch as the data files of population distribution is collected inthe tool first then the algorithm is activated to generate thedatabase The generation lasts almost twelve hours one timewith the specified parameters settings When the generationis finished the data of artificial Beijing is obtained in thedatabase with 19610000 agents and 16000 environments
5 H1N1 Experiments in ArtificialBeijing with KD-ACP
With the support of the artificial Beijing modeled before thecomputational experiments and parallel execution are alsoimplemented by KD-ACP The experiments are designed tofind the most optimal emergency decision response plansEpidemic in city is a typical scenario in emergency manage-ment H1N1 epidemic in Beijing in 2009 is used to be the usedcase to test KD-ACP According to Section 34 it is necessarytomodel emergencymanagement first in order to support thecomputational experiments
16 Mathematical Problems in Engineering
Table 4 The description of artificial Beijing generated by ASE
Statistical features ofartificial Beijing Descriptions of statistical features Data sourcedefault values
Agent roles Baby child primary student middle school student college student worker retired
Demographic data file ofBeijing
Environment types Super market store restaurant park stadium hospital school cliniccommercial building dormitory building residential building
Demographic data file ofBeijing
Demographic data Age distribution of population sex distribution of population householddistribution of population retired age distribution children distribution
Demographic data file ofBeijing
Environment data Distribution of all types of environments Environment distributiondata file of Beijing
Relationships Classmate relative college friendship family Demographic data file ofBeijing
Varying parameters
Age difference of couples119873(120583120590) 119873(03) isin [0 12]Minimummarriage age (male female) (22 20)Age difference of children119873(120583120590) 119873(13) isin [1 10]Enrollment rate of kindergarten 120572 08Enrollment rate of primary school 120573 1Enrollment rate of middle school 120574 1Enrollment rate of college 120575 08Entrance age of kindergarten 1205721015840 3Entrance age of primary school 1205731015840 7Entrance age of middle school 1205741015840 13Entrance age of middle school 1205751015840 19
Age distributioninformation table
Agent settable
Geospatialenvironment set table
Household settable
Figure 15 The geospatial and population database of Beijing city
51 Modelling Emergency Management The modelling ofemergency management consists of two parts the modellingof emergent events and the modelling of intervention mea-sures H1N1 and the intervention measures for H1N1 aremodeled in our case
511 Modelling H1N1 by MDT H1N1 Model in agent simu-lates the states transitions of health status and the relevantactions Referring to SIR (SusceptibleInfectiveRecovered)[43 44] models agent has three health statuses susceptibleinfected and healthy with immunity Only susceptible agentcan be infected by the contact with infected agents After
the three infected phases (incubation being symptomaticand recovery) of H1N1 model agent is healthy again It isworthy mentioning that the recovered agents can be infectediteratively when the immunity disappears gradually by timeThe state transitions are illustrated in Figure 18
According to [45] the latent period of H1N1 was in aWeibull distribution within people from one to seven days[46 47] The distribution was usually centered in the rangebetween one and three days AWeibull random variable [40]is used to model the latent period as described in
119879Lat = 120572 times [minus ln (1 minus 120574)]1120573+ 120592 (119879Lat ge 120592) (5)
Mathematical Problems in Engineering 17
Censusdata
Generate GUID
Assign individualhousehold
Generate familystructure
Assign family roleto every member
Endow individualwith age gender
social role
Distribute environmentlocations on the map
School Workplace Entertainmentlocations
Consumptionlocations
Assign school orworkplace to
individual
Correlatedindividual withother locations
Generate friendshipin school
Generate coworkerrelations inworkplace
Assign individualadministrative
region
Generate housebuildings
Locate householdto house building
Generate familyrelation
Generateneighborhood
relation
PopulationEnvironment
Environment listSocial network
Figure 16 The algorithm of generating geospatial and social environment data [36]
Edit
Edit
Edit
Edit
Edit
Edit
BrowseStatistical characteristic of population
Geospatial population data
Population data of street
Population types
Geospatial and social environment generation tool
Workspaces
Parameters distribution settings
Household types
Roles in household
Household structure
Social roles
BrowseBrowseBrowseBrowseBrowseBrowse
Population distribution data
File path
Retired distributionPop distributionHousehold numHousehold distributionSex distributionAge distributionData type
Figure 17 The GUI of Population and Geospatial Environment generation Tool (PGET)
18 Mathematical Problems in Engineering
Infected
SusceptibleHealth
withimmunity
Infected
by contac
t
with in
fected
agents
Recovery
Increase before beingsymptomatic symptomatic
Decrease after being
Decrease by time
Infectivity
Immunity
Figure 18 The transitions of agent health status
119879Lat denotes the duration of latent period 120592 120572 and 120573 arethe location parameter scale parameter and shape parameterof Weibull distribution respectively 120574 is a uniform randomnumber in the range [0 1] According to statistics [47] 120592 120572and 120573 are set as 0 18 and 121 respectively Then the mean(standard deviation) of latent period is calculated as 159 days(058 day2) Furthermore infectious period is set as 7 days
A susceptible studentrsquos probability of getting infecteddepended on the infectivity of infectious agents his ownimmunity level the duration of the contact action and soon The infectivity of an infected agent evolved with timeduring the infectious period According to the statistics ofinfectivity in the chart with red bars in Figure 19 an infectedstudent had the highest level of infectivity in the second dayafter he had the first symptom The infectivity levels in otherdays in comparison with the second day are in Table 5 Theday labeled as ldquominus1rdquo means the day before the starting of thesymptomatic period
Based on the algorithm discussed before theH1N1modelis developed with the help of MDT The settings of influenzamodel will be discussed in Section 521
512 Modelling Intervention Measures by MDT In corre-spondence intervention measures model is used to ceasethe outbreak of H1N1 As discussed in [42 48 49] if theappropriate measures are taken when the infectious diseaseemerges the transmission of the disease could be sloweddown and the damage could be decreased Therefore it isnecessary and important to design emergency interventionmeasures in the artificial Beijing According to the MinistryofHealth inChina the intervention policies [50] are designedto control the spread of influenza such as H1N1 The inter-ventionmeasures of these policies including are interventionsactivated time vaccination rate antibiotic delay to hospitalisolation duration close workspace duration and limitationof activities listed in Table 6 Similarly the models of theseintervention measures are also developed with the help ofMDT Using the reasonable data ranges listed in Table 6intervention models can be initialized to do the computa-tional experiments These experiments are used to show how
Symptomatic1 2 3 4 5 6 7minus1
Recovered
Tim
e (da
y)
Infe
ctiv
ity
Figure 19 The distribution of infectivity rate of an infected agent[37]
Table 5 The proportion of infectivity
Days (119894) minus1 1 2 3 4 5 6 6Proportion of infectivity 03 05 1 08 04 02 01 005
to restrain the H1N1 transmissionThe settings of these inter-vention measures in IMCT are introduced in Section 522
52 Experiment Settings
521 H1N1 Settings by EECT EECT is designed to initializethe emergent events models such as H1N1The settings comefrom the specified emergent event models As shown inFigure 20 in theH1N1 used case the settings are composed ofthe distribution of infection source infection rate infectionperiod and so on The infection rate changes with envi-ronments and infection periods The settings are initializedfrom both the research of medicine on H1N1 and the data ofH1N1 outbreak in Beijing in 2009The process of influenza inartificial Beijing is simulated to repeat the possible scenariosof H1N1 break During the computational experiments ofartificial Beijing theH1N1 settings such as the period of beingsymptomatic are updated by the dynamic data registrationsfrom OsdRT The updating corrects the H1N1 model duringthe experiments With the help of OsdRT artificial Beijing isable to approach actual society It is also an implementationof parallel execution
522 Intervention Measures Settings by IMCT In the H1N1case the settings of intervention measures are shown inFigure 21 in IMCT These measures are designed under theinstruction of Emergency Decision Organizations like ChinaDisease Center (CDC) The measures such as vaccinationrate the rate of closedworkspace and the starting time can beinitialized in the tool Emergency response plans are imple-mented in these settings As illustrated in Figure 7 all the pos-sible compositions within value settings in reasonable inter-vals are tested to find the most optimal emergency responseplan For example different compositions of measures suchas vaccination rates and rates of workspace closed are usedto design the plans of experiments By the analysis of theresults the best composition of interventionmeasures can beobtainedThe analysis will be discussed in case study in detail
Mathematical Problems in Engineering 19
Table 6 The intervention measures of influenza
Intervention measures ValuesInterventions activated Time 119879Intervention isin [0 80th day]Vaccination rate 119877Vaccination isin [0 1]Antibiotic rate 119877Antibiotic isin [0 1]Delay to hospital 119867ospital(119860 119894) 119879infected lt 119905 lt 119879recovered + 119879HDelay 119879HDelay isin (0 3th day]Isolation duration 119868solation(119860 119894) 119879Intervention lt 119905 lt 119879Intervention + 119879IDuration 119879IDuration isin (0 14th day]Close workspace duration 119862lose(119864119894) 119879Intervention lt 119905 lt 119879Intervention + 119879CWDuration 119864119894 isinWorkspace 119879CWDuration isin (0 7th day]Limitation of agent activity duration 119871 imitation(119860 119894) 119879Intervention lt 119905 lt 119879Intervention + 119879LAADuration 119879LAADuration isin (0 7th day]119877Vaccination means the vaccination rate of artificial Beijing while 119877Antibiotic means the antibiotic rate used for infected agents in hospital119867ospital (119860119894) means that the infected agent 119860119894 is sent to hospital he or she will receive the retreatment immediately119868solation (119860119894) means that the infected agent 119860119894 is isolated at home he or she can only contact with the other agents until isolation duration is over119862lose (119864119894) means that the environment 119864119894 typed of workspace is closed until the close workspace duration is over119871 imitation (119860119894) means that the activities of agent 119860119894 are limited only the workspace and home are allowed until limitation duration is over
Incubation Weibulldistribution
Symbolic LogNormaldistribution
Recovering LogNormaldistribution
H1N1influenza
Settings of H1N1 influenza model
Disease model
Transmissionmodel
Figure 20 The settings of H1N1 in EECT
It is worth notifying that the settings of interventionmeasures are also influenced by OsdRT For example theH1N1 outbreak areas are notified by OsdRTThe interventionmeasures such as close workspace are used only in theoutbreak areas Therefore the unnecessary computation ofintervention measures is avoided So the data injection byOsdRT not only makes artificial society approach actualsociety but also increases the performance of computationalexperiments
53 Experiments by EMT When the initialization of arti-ficial Beijing is ready EMT is used to do the computa-tional experiments EMT is composed of two parts thecentral controller in the central node and the residen-tial service deployed in the computational nodes Centralcontroller loads the experiment plans and controls the whole
experiments process the GUI of controller is shown inFigure 22 EMT manages the computational experimentsin four steps cluster configuration load experiment plansmodels deployment and running experiments
531 Cluster Configuration Before the configuration clusterinformation is collected from the residential services insidethe cluster The information includes nodes amount nodenames node IPs and CPU occupations It is used to quantifythe computing power of the clusterThe cluster configurationis based on this information The nodes are selected forexperiments Then the LP (local process) [51] number foreach node is set according to the number of the CPU kernelsAfter the configuration the experiment plans are loaded tocustomize the models and data
20 Mathematical Problems in Engineering
Streets of close workspaces
Limitations ofactivities
IsolationsRespirator
Closeworkspace
Antibiotic
VaccinationHospital
Surveillance
Figure 21 The settings of intervention measures in IMCT
Information ofexperiment plan
List of experiment plans
Start | Pause | Reset | Stop | Speedup | Slowdown | Back
NodeName | CPU occupation | Memory occupation |Network | Models | Status
Figure 22 The deployment and running of experiments in EMT
532 Load Experiment Plan As discussed in Section 35 thedescription of experiment plans is composed of the infor-mation in three aspects models and the initial data for theexperiments the deployment mapping tables from the mod-els to the nodes and the plan of experiment executionWhenthe experiment plan is loaded by the EMT the file namesand model descriptions of models and data are listed Withthe support of cluster configuration models can be mappedto the nodes in the GUI according to the computationrequirements EMT also provides the setting of experimentsitself including the running times start time and end time
533 Models and Data Deployment Models are stored inthe agent population and services repository According tothe settings in experiment plan the customized models aredownloaded from the repository first then the models suchas agent model environment model emergent events modeland intervention measures model are integrated to be thecompositionmodels of artificial BeijingThemodels and datalocations in the nodes are configured in the ldquomodel pathrdquoand ldquodata pathrdquo in the GUI of central controller Finallythe models are uploaded to the nodes in cluster with theconsideration of partition The partition of artificial society
Mathematical Problems in Engineering 21
models is different from the multiagent system in [52]Models in artificial society are divided into three levels agentsand environments grids and city Agents and environmentsare encapsulated in grids which compose the whole cityThus the partition of artificial society becomes the partitionof grids The parallelism degrees are found to support thepartition of models agent independence and grid indepen-dence According to the parallelisms agents are partitioned toLPs according to the population distribution in grids Agentmovements among grids are simulated by the movementevents among grids in different LPs As a result a two-tierparallel architecture is proposed to support the simulation ofartificial society The architecture is detailed in [53]
534 Running of Experiments When the preparation isready the experiments can be done by the central controllerof EMT The running and control commands of the experi-ments are grouped into three aspects start pause or stop theexperiments get the running information of the nodes suchas the snapshot of the desktop of node With the help of thesecommands users can run the computational experimentsto do the research on the emergency management It isworth notifying that the simulation engine is optimized inorder to support the ten million agent simulation UsingGPU as coprocessor a two-tier parallel simulation engine isdesigned with support of MPI and OpenCL through phasedsynchronization [54] One-sided communication is used forreflection of remote simulation objects and message passingbetween processes A general kernel function prototype iselaborately designed and conditionally compiled for execu-tion on both CPU and GPU Moreover the optimizationoperations like load balance is developed under the instruc-tion of activity-based simulation [55] The densely populatedgrids are given more CPU and memory resource accordingto the activity predictions
54 Results of H1N1 Experiments with KD-ACP On thebasis of artificial society models and the intervention mea-sures mentioned in Section 512 a series of computationalexperiments are performed to study the H1N1 influenza inBeijing In our case of artificial Beijing 19610000 agentsare simulated in the cluster within 48 CPU cores and 128Gmemories the nodes are connected by kilomega networks Ittakes 18 hours to simulate a 250-day disease spreading
In order to illustrate how the optimal plans are obtainedbyKD-ACP the experiments are divided into four groups theexperiments ofmodel validation the experiments of sensitiv-ity analysis of vaccination rate the experiments of sensitivityanalysis of isolation and the experiments of the combinationof interventionmeasuresThese four groups show the generalprocess of the research on emergency management in H1N1influenza Firstly the models of H1N1 are validated withthe support of historical data Secondly the traditionalmedical interventions of pandemic such as vaccination areanalyzed to find the optimal plan Thirdly the traditionalnonmedical interventions of pandemic such as isolation areanalyzed to find the optimal plan Finally the combination ofinterventions is tested to show the combined effect
It is worth mentioning that all the experiments in ourwork are performed 100 times These experiments are ini-tialized with different settings of interventions in IMCT Theresults shown in the figures are the mean values
541 The Experiments of Model Validation In our artifi-cial Beijing the models of H1N1 influenza are built basedon the existing researches Wang and her colleagues hadstudied the H1N1 influenza by SEIR (SusceptibleExposedInfectiveRecovered) models [56] With the support of theirwork we build the models of artificial Beijing in the mannerof multiagents
In Figure 23 the control group shown in red line isdrawn based on the realistic statistical data collected in H1N1influenza in Beijing in 2009 The influenza lasts more thansix months and more than 170000 people are infectedComparing with the control group the simulated resultswithout interventions are drawn in blue It is obvious that thesimulated case fits the control group well before 70 days Butthe ascending of control group is slowed after 70 days andthe peak value (17883) is much less than the simulated results(25135) It is because the vaccination is activated at day 67 incontrol group according to the intervention measures usedin 2009 The restrain by vaccination becomes effective eightdays later The change of infected number after vaccinationis shown by red spotted line The trend of control group isdifferent from the simulation data after 75 days
However the simulated results (blue line) in compu-tational experiments fit the control group (as shown inFigure 23) in the first 70 days when no intervention mea-sures are executed The key features such as effective basicreproduction number (119877
0) are also almost the same in both
cases Therefore it can be concluded that the models ofartificial society built in microview are validated by therealistic statistical data in influenza in Beijing in 2009
In this group of experiments no interventions are exe-cuted in order to validate the models Therefore the simu-lated results of this extreme case which cannot happen todaycould be obtained As shown by blue line the number ofinfected agents without interventions reaches peak value atday 91 with 25135 consisting of 8052 in symptomatic phaseand 28950 in recovery phase It is obvious that the totalnumber of infected agents grows slowly at the beginningtime Then the number increases quickly when the influenzabreaks and it reaches its maximum value with 383870 in theend This phenomenon can be explained as in the followingAt the beginning of influenza due to the limited numberof infected agents the propagation of the disease remainsslow but many agents become susceptible When the numberof infected agents increases rapidly the spreading becomessignificant However the number of infected agents decreasesafter day 91 It is because the disease spreading relies on thesocial networks and spatial contact networks [49] It meansthat the infected agents can only infect agents either in thesocial relationship networks or in the spatially contacted
542 The Experiments of Sensitivity Analysis of VaccinationRate According to Section 512 vaccination is a typical
22 Mathematical Problems in Engineering
0
Infe
cted
agen
ts
0 50 100 150 200 250
Statistical dataSimulation data
25000
20000
15000
10000
5000
Day d
Figure 23 The results with no intervention executed and the com-parison with realistic statistical data in Beijing in 2009
medical intervention in influenza Traditionally the effective-ness of vaccination can only be evaluated after the influenzaand the key settings such as vaccination rate are determinedby the experience It is always difficult for emergency man-agement organizations to find the most optimal vaccinationrate in advanceTherefore the sensitivity of vaccination rate isanalyzed in this group of experiments The settings are listedin Table 7 average results are shown in Figure 24
Vaccination rates are set linearly (10 30 50 and70) to test the changes of influenza The peak value ofinfected agents decreases from 25135 22392 15787 6907to 638 respectively It shows that the more the agents arevaccinated the smaller the peak value of the infected agentswill be In the case of 70 vaccination the total infected agentis only 13799 It means that the influenza in this case does nothappen due to the high vaccination rate
The change of peak values also shows an interestingphenomenon Vaccination rates increase linearly the peakvalues decrease slower than the ascending of vaccination ratesat beginning but the descending becomes more and morefaster with the linear increase of vaccination rate It meansthat the small scale of vaccination (less than 30) does notmake senseWith the increase of vaccination rate the numberof susceptible agents decreases significantly In sequencethe possibility of infections is lowered greatly Additionallyit could be found that the decrease of peak values comeslater than the increase of the vaccination rates This is quitereasonable according to Mei et alrsquos work in [57]
As a result it is concluded that high vaccination rate iseffective in the control of H1N1 influenza though costly
543 The Experiments of Sensitivity Analysis of IsolationAccording to Section 512 isolation is a typical nonmedicalintervention It is the most used intervention measure in
Table 7 The initial settings of experiments of vaccination
Parameters ValueAgent count 19610000Initial infected count 40Interventions activated time(119879Intervention)
0
Vaccination rate (119877Vaccination) 0 10 30 50 70
0
Infe
cted
agen
ts
Loading intervention of vaccination
No vaccination10 vaccination30 vaccination
50 vaccination70 vaccination
0 50 100 150 200 250
25000
20000
15000
10000
5000
Day d
Figure 24 The results of loading intervention of vaccination
the emergency management in influenza As mentionedbefore a problem also exists that the measure cannot bevalidated before the intervention Therefore the sensitivityof isolation is analyzed in this group of experiments It isworth mentioning that the isolation is usually combinedwith the hospital policy It means the infected individual issent to hospital meanwhile the people who contact withthe infected are all isolated As a result parameters delay tohospital and isolation duration are designed together in thesettings of experiment Delay to hospital stands for the periodfrom infection time to hospital time while Isolation durationrepresents the period when an agent cannot contact others ifhe has contact with an infected agent The settings of theseparameters are listed in Table 8 average results are shown inFigure 25
Firstly the parameter delay to hospital is simulated aloneto test the sensitivity The simulated results are shown insolid line in Figure 25 The peak value decreases from 6708to 2218 when the delay to hospital decreases from 15 daysto 1 day The half day change brings the 67 reductionMoreover the peak value almost approximates to zero inthe case of 05 days The influenza does not happen in thiscase It can be concluded that the earlier the infected agents
Mathematical Problems in Engineering 23
Table 8 The initial settings of experiments of isolation
Parameters ValueAgent count 19610000Initial infected count 40Interventions activated time (119879Intervention) 80Delay to hospital (119879HDelay) isolation duration (119879IDuration) (15 0) (1 0) (05 0) (15 3) (1 3) (05 3)
0 50 100 150 200 2500
1000
2000
3000
4000
5000
6000
7000
Infe
cted
agen
ts
Loading intervention of isolation
Delay to hospital = 15
Delay to hospital = 10
Delay to hospital = 05
Delay to hospital = 15 isolation duration = 3
Delay to hospital = 10 isolation duration = 3
Delay to hospital = 05 isolation duration = 3
Day d
Figure 25 The results of loading intervention of isolation
are sent to hospital the less the agents will be infectedSo the construction of effective mechanism to find infectedindividuals is important in the emergency management ofinfluenza
Secondly the parameter isolation duration is added tosimulate the isolation of interventionObviously the isolationmeasure decreases the infected agents greatly Compared tothe parameter delay to hospital of 15-day case the peak valuedecreases from 6708 to 362when the isolation is added in theinterventionThe reduction of 6346 agents comes from the 3-day isolation of agents who had contacted with the infectedagents However the decrease of parameter delay to hospitalcannot bring the similar reduction in the 3-day isolationcases According to the figure the peak value decreases from362 to 51 from 15-day case to 05-day case only 311 agents aresaved from infection As a result the sensitivity of simulatedresults does not change obviously with delay to hospital inthe case of isolation It can be concluded that isolation is amore effective measure Though it is not easy for emergencymanagement organizations to decrease the delay to hospitaldue to the current monitoring mechanism the influenza canstill be controlled by isolation
544 The Experiments of Combination of Intervention Mea-sures Isolation is not the only nonmedical intervention ininfluenza close workspace is another common measureWhen close workspace is activated the workspaces are allclosed agents can only stay home at the closed time Luckilyisolation and close workspace are independent from oneanother It is possible for emergency management organi-zations to use these two measures together So this groupof experiments is designed to obtain the effectiveness of thecombination of intervention measures The parameters areinitialized with different settings listed in Table 9
Parameters delay to hospital and isolation duration is setin the middle values (1 day 3 days) according to the lastsection According to the experience close duration is set asseven days It is known that close workspace is a really costintervention measure the activated time is a key parameterWith the fixed parameters mentioned before the changeof interventions activated time is analyzed in this group ofcomputational experimentsThe average results are shown inFigure 26
Compared with the simulated results of experimentsdiscussed above Figure 26 strongly illustrates that influenzais ceased when the intervention measures are combinedtogether The peak value of infected agents is 141 in the ldquo80thdays of close-workspace and isolation with 1 day delay tohospitalrdquo case while 25135 in ldquono interventionrdquo case 2218 inldquo1 day delay to hospitalrdquo case 203 in ldquo1 day delay to hospitaland isolationrdquo case The interventions bring a great reductionof infection the peak values decrease from ten thousandslevel to hundred level
On the other hand the peak value even decreases to 108when the execution time of close workspace is 10 days earlierThe date of maximum value of infected agents is advancedfrom the 135th day to the 108th day The whole circle of H1N1influenza is reducedThe results of this group of experimentsshow that intervention measures should be activated as soonas possible and the compositions of interventions are moreeffective in emergency management of H1N1 influenza
In summary based on the general process lots of exper-iments could be done by KD-ACP to study emergency man-agement in H1N1 influenza Many conclusions are obtainedbased on the analysis of the experiments in this sectionAccording to these conclusions themost optimal response inour case study is the case of ldquo70th days of close-workspace andisolation with 1 day delay to hospitalrdquo It is because this casehas the fewest infected agents relatively But the modelling ofH1N1 influenza is not sufficient Economic and social factorsof interventions such as the money cost by vaccination andisolation the influence of the close-workspace in society
24 Mathematical Problems in Engineering
Table 9 The initial settings of experiments of isolation and closeworkspace
Parameters ValueAgent count 19610000Initial infected count 40Interventions activated time (119879Intervention) 70th day 80th dayDelay to hospital (119879HDelay) 1 dayIsolation duration (119879IDuration) 3 daysClose workspace duration (119879CWDuration) 7 days
50 100 150 200 2500
50
100
150
200
Infe
cted
agen
ts
Loading intervention of isolation and close workspace
Delay to hospital = 1 isolation duration = 3Start close time = 70 close duration time = 7Delay to hospital = 1 isolation duration = 3Start close time = 80 close duration time = 7
Day d
Figure 26The results of loading interventions of isolation and closeworkspace
and even opinion transmission of H1N1 in internet are notconsidered in our case study Without the analysis of thesefactors it does not make sense to talk about the most optimalintervention measures However KD-ACP had provided asoftware framework and general process of the research onemergency management Supported by KD-ACP the expertsof emergency management are able to study emergent eventssuch as H1N1 influenza as detailed as possible Based on theexperiments and analysis again and again the most optimalintervention measures will be found ultimately
55 Parallel Execution by ASST and OsdRT As discussed inSection 35 ASST andOsdRT are used to support the parallelexecution of artificial Beijing The statistical data is collectedand stored in artificial society runtime database at runtimeASST downloads the infection data from the database anddisplays it in the map As shown in Figure 27 the infectedagents are distributed in the districts in BeijingThe snapshotgives the geospatial distribution of the H1N1 epidemic Thelegends of the agents in different status are listed in the left
right corner in Figure 27 Five statuses are shown in the figurehealth without vaccination health with vaccination incuba-tion being symptomatic and convalescence ASST providesa very direct feeling for the emergency experts Partly withthe situation from ASST Emergency Decision Organizationmakes the emergency response plans to the response of theemergency scenario With the help of plans the epidemicin actual society can be restrained In the meantime theinfluenza information is reflected in the Internet Accordingto Figure 8 the data acquisition data extraction and datastandardization of H1N1 influenza are processed in OsdRTOsdRT is implemented by Dr Caorsquos team from Institute ofAutomation from Chinese Academy of Sciences [58] Thetool provides not only the reports of H1N1 SARS hand-foot-mouth disease and intestinal diseases are all monitored Theweb page is shown in Figure 28 H1N1 reports from Internetare collected first Based on the data extraction of the reportsH1N1 case of distribution is obtained and shown in theincidencemap As a result H1N1 incidence data is regarded asthe knowledge which is sent to EECT and IMCTThe settingsof emergent events and intervention measures are modifiedin order to make artificial society approach actual society innext turn of computational experiments With the help ofOsdRT the loop of parallel execution process integrates ACPapproach
6 Conclusion
This work provides an integrated software framework forsocial Computing in emergencymanagement From a systemperspective KD-ACP provides a reliable flexible low costand effective platform for the scientists to do the research onemergency response problems The analysis and predictionof these problems with inherent complexity can be solvedby the repetition of possible alternative experiments onthe software framework Therefore KD-ACP which is anattempt to implement ACP approach seemed as an effectivecomputational framework to support the decision making inemergency response
Currently KD-ACP is used to study the H1N1 epidemicin Beijing in 2009 The most optimal compositions of inter-vention measures are testified to support the response tonext influenza However it is still a long way ahead beforethe automation systematization and practicalization of KD-ACP A lot of work will be carried out along several directionsas in the following
(i) The parallel execution process in KD-ACP is actuallynot strongly connected with the emergency responseorganization Traditional experience and theory stillplay a dominant role in the decision of emergencymanagement It is necessary to propose a mechanismof hall for workshop to facilitate the parallel executionloop in actual society
(ii) Section 33 introduces the principle and process ofmodelling artificial society Metamodels and modelsof artificial Beijing are built by GME Model transfor-mation and code generation are used to implementcodes from FSM based models But many details
Mathematical Problems in Engineering 25
Fangshan district Daxing district Fengtai district Xicheng district
Mentougou district Haidian districtShijingshan district Dongcheng district
Incubation
Symptomatic
Convalescence
Health withoutvaccination
Health withvaccinationLegend of agent
health status
Figure 27 The situation of artificial Beijing by ASST
H1N1
Districts of Beijing
H1N1 case ofdistribution
Incidence map
Distribution ofH1N1 reportsfrom internet
Incidence of H1N1
Fangshan districtLiangxiang town
Population 19000Incidence 135789
(1100000)
Figure 28 The implementation of OsdRT by Institute of Automation from Chinese Academy of Sciences
are still not illustrated due to the topic of thispaper The key techniques such as template of codeframework and model transformations from specificdomains to FSM will be focused and introduced indetail in our next paper
(iii) The performance of computational experiments inKD-ACP is still slower than expectation The opti-mization of simulation engine and models structureis necessary to improve the performance So it ispossible to run artificial society with the real worldin parallel
In summary KD-ACP paves a new way for scientists inboth emergencymanagement and computation simulation to
collaborate with each other in solving the emergency man-agement problems by social computing
Research Highlights
(i) KD-ACP is implemented based on ACP approach
(ii) Artificial Beijing is built with the help of KD-ACP
(iii) H1N1 is modeled to simulate the influenza in Beijingin 2009
(iv) Computational experiments testify the effectivenessof intervention measures
26 Mathematical Problems in Engineering
Conflict of Interests
The authors declare that there is no conflict of interestsregarding the publication of this paper
Acknowledgment
This work was supported in part by the National ScienceFoundation of China (nos 91024030 71303252 61403402and 91324013)
References
[1] W Mao and F-Y Wang Advances in Intelligence and SecurityInformatics Academic Press Oxford UK 2012
[2] E Bonabeau ldquoAgent-based modeling methods and techniquesfor simulating human systemsrdquo Proceedings of the NationalAcademy of Sciences of the United States of America vol 99 no3 pp 7280ndash7287 2002
[3] K M Carley D B Fridsma E Casman et al ldquoBioWar scalableagent-basedmodel of bioattacksrdquo IEEETransactions on SystemsMan and Cybernetics A Systems and Humans vol 36 no 2 pp252ndash265 2006
[4] J M Epstein ldquoModelling to contain pandemicsrdquo Nature vol460 no 7256 p 687 2009
[5] Los Alamos National Laboratory Reports about EPISIMS LosAlamos National Laboratory 2003
[6] C H Builder and S C Bankes ldquoArtificial societies a conceptfor basic research on the societal impacts of informationtechnologyrdquo RAND Report 149 RAND Corp 1991
[7] F-Y Wang ldquoComputational theory and method on complexsystemrdquo China Basic Science vol 6 no 5 pp 3ndash10 2004
[8] F-Y Wang ldquoToward a paradigm shift in social computing theACP approachrdquo IEEE Intelligent Systems vol 22 no 5 pp 65ndash67 2007
[9] F-Y Wang and S Tang ldquoArtificial societies for integratedand sustainable development of metropolitan systemsrdquo IEEEIntelligent Systems vol 19 no 4 pp 82ndash87 2004
[10] D Wen Y Yuan and X-R Li ldquoArtificial societies computa-tional experiments and parallel systems an investigation ona computational theory for complex socioeconomic systemsrdquoIEEE Transactions on Services Computing vol 6 no 2 pp 177ndash185 2013
[11] F Wang F Zeng and Y Yuan ldquoAn ACP-based approach forcomplexity analysis of E-commerce systemrdquo Complex Systemsand Complexity Science vol 5 no 3 pp 1ndash8 2008 (Chinese)
[12] F-Y Wang and P K Wong ldquoResearch commentary intelligentsystems and technology for integrative and predictivemedicinean ACP approachrdquoACMTransactions on Intelligent Systems andTechnology vol 4 no 2 pp 1ndash6 2013
[13] J Sifeng X Gang F Dong and H Chunpeng ldquoStudy on theemergency rescue decision support system of petrochemicalplant based on ACP theoryrdquo in Proceedings of the 6th Manage-ment Annual Meeting (MAM rsquo11) 2011 (Chinese)
[14] X Dong D Fan G Xiong F Zhu Z Zhang and Y Yao ldquoPar-allel Bus Rapid Transit (BRT) operation management systembased on ACP approachrdquo in Proceedings of the 6th ManagementAnnual Meeting (MAM rsquo11) Beijing China 2011
[15] B Ning F-Y Wang H-R Dong R-M Li D Wen and L LildquoParallel systems for Urban rail transportation based on ACP
approachrdquo Journal of Transportation Systems Engineering andInformation Technology vol 10 no 6 pp 23ndash28 2010 (Chinese)
[16] F-Y Wang J Zhao and S-X Lun ldquoArtificial power systemsfor the operation and management of complex power gridsrdquoSouthern Power System Technology vol 2 no 3 pp 1ndash11 2008(Chinese)
[17] M J North T R Howe N T Collier and J R Vos ldquoRepastsimphony development environmentrdquo in Proceedings of theAgent 2005 Conference on Generative Social Processes Modelsand Mechanisms 2005
[18] M J North T RHowe N T Collier and J R Vos ldquoRepast Sim-phony runtime systemrdquo in Proceedings of the Agent Conferenceon Generative Social Processes Models and Mechanisms 2005
[19] SWARM Development Group Swarm 211 Reference Guide2000 httpwwwswarmorg
[20] S Tisue and U Wilensky ldquoNetLogo a simple environment formodeling complexityrdquo in International Conference on ComplexSystems pp 16ndash21 2004
[21] S Luke C Cioffi-Revilla L Panait K Sullivan and G BalanldquoMASON a multiagent simulation environmentrdquo Simulationvol 81 no 7 pp 517ndash527 2005
[22] M Ichikawa H Tanuma Y Koyama andH Deguchi ldquoSOARSintroduction as a social microscope for simulations of socialinteractions and gamingrdquo in Organizing and Learning throughGaming and Simulation Proceedings of the 38th Conference ofthe International Simulation and Gaming Association (ISAGArsquo07) pp 149ndash158 2007
[23] D S Burke J M Epstein D A T Cummings et al ldquoIndividual-based computational modeling of smallpox epidemic controlstrategiesrdquo Academic Emergency Medicine vol 13 no 11 pp1142ndash1149 2006
[24] F-Y Wang ldquoParallel control and management for intelligenttransportation systems concepts architectures and applica-tionsrdquo IEEE Transactions on Intelligent Transportation Systemsvol 11 no 3 pp 630ndash638 2010
[25] Z Molnar D Balasubramanian and A Ledeczi ldquoAn introduc-tion to the genericmodeling environmentrdquo in Proceedings of theTOOLS Europe 2007 Workshop on Model-Driven DevelopmentTool Implementers Forum Zurich Switzerland 2007
[26] S Kelly and J-P Tolvanen Domain Framework in Domain-Specific Modeling Enabling Full Code Generation JohnWiley ampSons Hoboken NJ USA 2007
[27] B P Zeigler H Praehofer and T G Kim Theory of Modelingand Simulation Integrating Discrete Event and ContinuousComplex Dynamic Systems Academic Press San Diego CalifUSA 2000
[28] G Booch J Rumbaugh and I JacobsonThe Unified ModelingLanguage User Guide Addison-Wesley Reading Mass USA1999
[29] F Wagner R Schmuki T Wagne and P Wolstenholme Mod-eling Software with Finite State Machines A Practical ApproachCRC amp Taylor amp Francis Boca Raton Fla USA 2006
[30] T Espiner ldquoChina builds worldrsquos fastest supercomputerrdquo ZDNetUK October 2010
[31] B Chen and G Guo ldquoA two-tier parallel architecture for arti-ficial society simulationrdquo in Proceedings of the ACMIEEESCS26th Workshop on Principles of Advanced and DistributedSimulation (PADS rsquo12) pp 184ndash186 July 2012
[32] G Guo B Chen X G Qiu and Z Li ldquoParallel simulation oflarge-scale artificial society on CPUGPU mixed architecturerdquo
Mathematical Problems in Engineering 27
in Proceedings of the ACMIEEESCS 26th Workshop on Prin-ciples of Advanced and Distributed Simulation (PADS rsquo12) pp174ndash177 July 2012
[33] S-X Lun ldquoResearch on the classification of parallel executionmodes of ACP theoryrdquo Acta Automatica Sinica vol 38 no 10pp 1602ndash1608 2012
[34] F-YWang ldquoA framework for social signal processing and anal-ysis from social sensing networks to computational dialecticalanalyticsrdquo Science China vol 43 no 12 pp 1598ndash1611 2013
[35] G GuoOne Model Userrsquos Manual version 12 National Univer-sity of Defense University 2011
[36] Y Ge R Meng Z Cao X Qiu and K Huang ldquoVirtual citymdashanindividual-based digital environment for human mobility andinteractive behaviorrdquo Simulation Transactions of the Society forthe Modeling and Simulation International 2014
[37] M Kretzschmar and R T Mikolajczyk ldquoContact profiles ineight European countries and implications for modelling thespread of airborne infectious diseasesrdquo PLoS ONE vol 4 no6 Article ID e5931 2009
[38] W Duan and X Qiu ldquoFostering artificial societies using sociallearning and social control in parallel emergency managementsystemsrdquo Frontiers in Computer Science vol 6 no 5 pp 604ndash610 2012
[39] W J Edmunds C J OrsquoCallaghan and D J Nokes ldquoWhomixes with whom Amethod to determine the contact patternsof adults that may lead to the spread of airborne infectionsrdquoProceedings of the Royal Society B Biological Sciences vol 264no 1384 pp 949ndash957 1997
[40] J Banks J S Carson B L Neison and D M Nicol Discrete-Event System Simulation Prentice Hall Englewood Cliffs NJUSA 4th edition 2007
[41] H Yongxia ldquoInvestigation and analysis of college studentsrsquo workand restrdquo Value Engineering vol 31 no 3 2012 (Chinese)
[42] W Duan Z Cao Y Wang et al ldquoAn ACP approach to publichealth emergency management using a campus outbreak ofh1n1 influenza as a case studyrdquo IEEE Transactions on SystemsMan and Cybernetics Part A Systems and Humans vol 43 no5 pp 1028ndash1041 2013
[43] HWHethcote ldquoThemathematics of infectious diseasesrdquo SIAMReview vol 42 no 4 pp 599ndash653 2000
[44] G Chowell E Shim F Brauer and et al ldquoModelling thetransmission dynamics of acute haemorrhagic conjunctivitisapplication to the 2003 outbreak in Mexicordquo Statistics inMedicine vol 25 no 11 pp 1840ndash1857 2006
[45] Y Ge L Liu B Chen X Qiu and K Huang ldquoAgent-basedmodeling for InfluenzaH1N1 in an artificial classroomrdquo SystemsEngineering Procedia vol 2 pp 94ndash104 2011
[46] A R Tuite A L Greer MWhelan et al ldquoEstimated epidemio-logic parameters andmorbidity associatedwith pandemicH1N1influenzardquo CMAJ vol 182 no 2 pp 131ndash136 2010
[47] L Brouwers B Cakici M Camitz A Tegnell and MBoman ldquoEconomic consequences to society of pandemic H1N1influenza 2009mdashpreliminary results for Swedenrdquo Euro Surveil-lance vol 14 no 37 pp 1ndash7 2009
[48] Y Ge L Liu X Qiu H Song Y Wang and K Huang ldquoAframework of multilayer social networks for communicationbehavior with agent-based modelingrdquo Simulation Transactionsof the Society for the Modeling and Simulation International vol89 no 7 pp 810ndash828 2013
[49] W Duan X Qiu Z Cao X Zheng and K Cui ldquoHeterogeneousand stochastic agent-based models for analyzing infectious
diseasesrsquo super spreadersrdquo IEEE Intelligent Systems vol 28 no4 pp 18ndash25 2013
[50] The Preparation and Emergency Response Plans to PandemicPublished by Ministry of Health Ministry of Health 2005(Chinese)
[51] R M Fujimoto Parallel and Distributed Simulation SystemsWiley Interscience 2000
[52] F Cicirelli A Giordano and L Nigro ldquoDistributed simulationof situated multi-agent systemsrdquo in Proceedings of the 15thIEEEACM International Symposium on Distributed Simulationand Real Time Applications (DS-RT rsquo11) pp 28ndash35 September2011
[53] B Chen and G Guo ldquoA two-tier parallel architecturefor artificial society simulationrdquo in Proceedings of the 26thACMIEEESCS Workshop on Principles of Advanced and Dis-tributed Simulation (PADS rsquo12) pp 184ndash186 July 2012
[54] G Guo B Chen and X Qiu ldquoParallel simulation of large-scale artificial society with GPU as coprocessorrdquo InternationalJournal of Modeling Simulation and Scientific Computing vol4 no 2 Article ID 1350005 2013
[55] B Chen L-B Zhang X-C Liu and H Vangheluwe ldquoActivity-based simulation using DEVS increasing performance by anactivity model in parallel DEVS simulationrdquo Journal of ZhejiangUniversity Science C vol 15 no 1 pp 13ndash30 2014
[56] X-L Wang P Yang Z-D Cao et al ldquoQuantitative evaluationon the effectiveness of prevention and control measures againstpandemic influenza A (H1N1) in Beijing 2009rdquoChinese Journalof Epidemiology vol 31 no 12 pp 1374ndash1378 2010 (Chinese)
[57] S Mei A D van de Vijver L Xuan Y Zhu and P M ASloot ldquoQuantitatively evaluating interventions in the influenzaA (H1N1) epidemic on China campus grounded on individual-based simulationsrdquo Procedia Computer Science vol 1 no 1 pp1675ndash1682 2010
[58] X Qiu Z Cao and Z Li Mid-Term Reports of China NationalScience Foundation 91024030 2014
Submit your manuscripts athttpwwwhindawicom
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
MathematicsJournal of
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Mathematical Problems in Engineering
Hindawi Publishing Corporationhttpwwwhindawicom
Differential EquationsInternational Journal of
Volume 2014
Applied MathematicsJournal of
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Probability and StatisticsHindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Journal of
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Mathematical PhysicsAdvances in
Complex AnalysisJournal of
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
OptimizationJournal of
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
CombinatoricsHindawi Publishing Corporationhttpwwwhindawicom Volume 2014
International Journal of
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Operations ResearchAdvances in
Journal of
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Function Spaces
Abstract and Applied AnalysisHindawi Publishing Corporationhttpwwwhindawicom Volume 2014
International Journal of Mathematics and Mathematical Sciences
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
The Scientific World JournalHindawi Publishing Corporation httpwwwhindawicom Volume 2014
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Algebra
Discrete Dynamics in Nature and Society
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Decision SciencesAdvances in
Discrete MathematicsJournal of
Hindawi Publishing Corporationhttpwwwhindawicom
Volume 2014 Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Stochastic AnalysisInternational Journal of
Mathematical Problems in Engineering 15
(a) The implementation of agent in MDT (b) The code framework of agent
Figure 14 The agent model implementation in MDT [35]
415 The Implementation of Models by MDT All the modelsare developed in the MDT Figure 14 shows the GUI of thetool MDT is used to implement the state space and statetransitions of FSM models The code framework is shownin Figure 14(b) Using the C++ inheritance features and thetechniques of dynamic link library (dll) the models areencapsulated in the dll componentsThe flexible compositionmechanism is designed to support the models evolution incomputational experiments
416 The Description of Artificial Beijing by ASE As men-tioned in Section 34 ASE describes the artificial Beijingin macroview The statistical information of Beijing city iscollected by ASE Table 4 lists part of the description ofartificial Beijing generated by ASE
42 How to Obtain the Data of Beijing City
421 Population and Geospatial Environment Database ofBeijing City Based on the models developed for ArtificialBeijing it is necessary to generate the initial data to do thecomputational experiments Because the individual level dataare not available it is necessary to construct an individual-based population database for both accurate computationalexperiment and determining optimal decisions
According to the state spaces of models discussed inSection 412 the geospatial and population database of Arti-ficial Beijing is designed as shown in Figure 15 The kernelpart of database consists of the tables such as agent list tablegeospatial environment list table household list table andagent distribution table Agent list table is an individual leveltable used to store the data to initialize the state space of agentmodel such as id and gender age Geospatial environmentlist table is also an individual level table which stores thedata of environment attributes such as id street id andtype Household list table and agent distribution table storethe statistical data for the data generation for agent andenvironment models With the help of the geospatial andpopulation database the data generating of artificial Beijingis proposed in the next section
422 Generating Geospatial and Social Environment Datafor Artificial Beijing by PGET The algorithm of generatinggeospatial and social environment data for artificial Beijingis used to quantify the spatial distribution of population andformalize geospatial behavior of each agent The algorithmshown in Figure 16 is capable of generating a syntheticartificial society which allowsmultiresolution statistical datasocial interactive behavior and multilayer social networksto be integrated together The synthetic population canrepresent individual agents in the form of households andhousehold members and the synthetic population is statis-tically equivalent to a real population For each householdcharacteristics such as address household size family typesand relationships are generated Each person is described bycharacteristics such as age gender social role and correlatedlocations The algorithm provides an effective methodologyto reconstruct the computing environment in high resolutionby using statistical data in low resolution leading to betterprediction and management of emergencies The details ofalgorithm are illustrated in [36]
As shown in Figure 17 the implementation of the algo-rithm is embedded inside the GEPTThe collective input datasuch as the data files of population distribution is collected inthe tool first then the algorithm is activated to generate thedatabase The generation lasts almost twelve hours one timewith the specified parameters settings When the generationis finished the data of artificial Beijing is obtained in thedatabase with 19610000 agents and 16000 environments
5 H1N1 Experiments in ArtificialBeijing with KD-ACP
With the support of the artificial Beijing modeled before thecomputational experiments and parallel execution are alsoimplemented by KD-ACP The experiments are designed tofind the most optimal emergency decision response plansEpidemic in city is a typical scenario in emergency manage-ment H1N1 epidemic in Beijing in 2009 is used to be the usedcase to test KD-ACP According to Section 34 it is necessarytomodel emergencymanagement first in order to support thecomputational experiments
16 Mathematical Problems in Engineering
Table 4 The description of artificial Beijing generated by ASE
Statistical features ofartificial Beijing Descriptions of statistical features Data sourcedefault values
Agent roles Baby child primary student middle school student college student worker retired
Demographic data file ofBeijing
Environment types Super market store restaurant park stadium hospital school cliniccommercial building dormitory building residential building
Demographic data file ofBeijing
Demographic data Age distribution of population sex distribution of population householddistribution of population retired age distribution children distribution
Demographic data file ofBeijing
Environment data Distribution of all types of environments Environment distributiondata file of Beijing
Relationships Classmate relative college friendship family Demographic data file ofBeijing
Varying parameters
Age difference of couples119873(120583120590) 119873(03) isin [0 12]Minimummarriage age (male female) (22 20)Age difference of children119873(120583120590) 119873(13) isin [1 10]Enrollment rate of kindergarten 120572 08Enrollment rate of primary school 120573 1Enrollment rate of middle school 120574 1Enrollment rate of college 120575 08Entrance age of kindergarten 1205721015840 3Entrance age of primary school 1205731015840 7Entrance age of middle school 1205741015840 13Entrance age of middle school 1205751015840 19
Age distributioninformation table
Agent settable
Geospatialenvironment set table
Household settable
Figure 15 The geospatial and population database of Beijing city
51 Modelling Emergency Management The modelling ofemergency management consists of two parts the modellingof emergent events and the modelling of intervention mea-sures H1N1 and the intervention measures for H1N1 aremodeled in our case
511 Modelling H1N1 by MDT H1N1 Model in agent simu-lates the states transitions of health status and the relevantactions Referring to SIR (SusceptibleInfectiveRecovered)[43 44] models agent has three health statuses susceptibleinfected and healthy with immunity Only susceptible agentcan be infected by the contact with infected agents After
the three infected phases (incubation being symptomaticand recovery) of H1N1 model agent is healthy again It isworthy mentioning that the recovered agents can be infectediteratively when the immunity disappears gradually by timeThe state transitions are illustrated in Figure 18
According to [45] the latent period of H1N1 was in aWeibull distribution within people from one to seven days[46 47] The distribution was usually centered in the rangebetween one and three days AWeibull random variable [40]is used to model the latent period as described in
119879Lat = 120572 times [minus ln (1 minus 120574)]1120573+ 120592 (119879Lat ge 120592) (5)
Mathematical Problems in Engineering 17
Censusdata
Generate GUID
Assign individualhousehold
Generate familystructure
Assign family roleto every member
Endow individualwith age gender
social role
Distribute environmentlocations on the map
School Workplace Entertainmentlocations
Consumptionlocations
Assign school orworkplace to
individual
Correlatedindividual withother locations
Generate friendshipin school
Generate coworkerrelations inworkplace
Assign individualadministrative
region
Generate housebuildings
Locate householdto house building
Generate familyrelation
Generateneighborhood
relation
PopulationEnvironment
Environment listSocial network
Figure 16 The algorithm of generating geospatial and social environment data [36]
Edit
Edit
Edit
Edit
Edit
Edit
BrowseStatistical characteristic of population
Geospatial population data
Population data of street
Population types
Geospatial and social environment generation tool
Workspaces
Parameters distribution settings
Household types
Roles in household
Household structure
Social roles
BrowseBrowseBrowseBrowseBrowseBrowse
Population distribution data
File path
Retired distributionPop distributionHousehold numHousehold distributionSex distributionAge distributionData type
Figure 17 The GUI of Population and Geospatial Environment generation Tool (PGET)
18 Mathematical Problems in Engineering
Infected
SusceptibleHealth
withimmunity
Infected
by contac
t
with in
fected
agents
Recovery
Increase before beingsymptomatic symptomatic
Decrease after being
Decrease by time
Infectivity
Immunity
Figure 18 The transitions of agent health status
119879Lat denotes the duration of latent period 120592 120572 and 120573 arethe location parameter scale parameter and shape parameterof Weibull distribution respectively 120574 is a uniform randomnumber in the range [0 1] According to statistics [47] 120592 120572and 120573 are set as 0 18 and 121 respectively Then the mean(standard deviation) of latent period is calculated as 159 days(058 day2) Furthermore infectious period is set as 7 days
A susceptible studentrsquos probability of getting infecteddepended on the infectivity of infectious agents his ownimmunity level the duration of the contact action and soon The infectivity of an infected agent evolved with timeduring the infectious period According to the statistics ofinfectivity in the chart with red bars in Figure 19 an infectedstudent had the highest level of infectivity in the second dayafter he had the first symptom The infectivity levels in otherdays in comparison with the second day are in Table 5 Theday labeled as ldquominus1rdquo means the day before the starting of thesymptomatic period
Based on the algorithm discussed before theH1N1modelis developed with the help of MDT The settings of influenzamodel will be discussed in Section 521
512 Modelling Intervention Measures by MDT In corre-spondence intervention measures model is used to ceasethe outbreak of H1N1 As discussed in [42 48 49] if theappropriate measures are taken when the infectious diseaseemerges the transmission of the disease could be sloweddown and the damage could be decreased Therefore it isnecessary and important to design emergency interventionmeasures in the artificial Beijing According to the MinistryofHealth inChina the intervention policies [50] are designedto control the spread of influenza such as H1N1 The inter-ventionmeasures of these policies including are interventionsactivated time vaccination rate antibiotic delay to hospitalisolation duration close workspace duration and limitationof activities listed in Table 6 Similarly the models of theseintervention measures are also developed with the help ofMDT Using the reasonable data ranges listed in Table 6intervention models can be initialized to do the computa-tional experiments These experiments are used to show how
Symptomatic1 2 3 4 5 6 7minus1
Recovered
Tim
e (da
y)
Infe
ctiv
ity
Figure 19 The distribution of infectivity rate of an infected agent[37]
Table 5 The proportion of infectivity
Days (119894) minus1 1 2 3 4 5 6 6Proportion of infectivity 03 05 1 08 04 02 01 005
to restrain the H1N1 transmissionThe settings of these inter-vention measures in IMCT are introduced in Section 522
52 Experiment Settings
521 H1N1 Settings by EECT EECT is designed to initializethe emergent events models such as H1N1The settings comefrom the specified emergent event models As shown inFigure 20 in theH1N1 used case the settings are composed ofthe distribution of infection source infection rate infectionperiod and so on The infection rate changes with envi-ronments and infection periods The settings are initializedfrom both the research of medicine on H1N1 and the data ofH1N1 outbreak in Beijing in 2009The process of influenza inartificial Beijing is simulated to repeat the possible scenariosof H1N1 break During the computational experiments ofartificial Beijing theH1N1 settings such as the period of beingsymptomatic are updated by the dynamic data registrationsfrom OsdRT The updating corrects the H1N1 model duringthe experiments With the help of OsdRT artificial Beijing isable to approach actual society It is also an implementationof parallel execution
522 Intervention Measures Settings by IMCT In the H1N1case the settings of intervention measures are shown inFigure 21 in IMCT These measures are designed under theinstruction of Emergency Decision Organizations like ChinaDisease Center (CDC) The measures such as vaccinationrate the rate of closedworkspace and the starting time can beinitialized in the tool Emergency response plans are imple-mented in these settings As illustrated in Figure 7 all the pos-sible compositions within value settings in reasonable inter-vals are tested to find the most optimal emergency responseplan For example different compositions of measures suchas vaccination rates and rates of workspace closed are usedto design the plans of experiments By the analysis of theresults the best composition of interventionmeasures can beobtainedThe analysis will be discussed in case study in detail
Mathematical Problems in Engineering 19
Table 6 The intervention measures of influenza
Intervention measures ValuesInterventions activated Time 119879Intervention isin [0 80th day]Vaccination rate 119877Vaccination isin [0 1]Antibiotic rate 119877Antibiotic isin [0 1]Delay to hospital 119867ospital(119860 119894) 119879infected lt 119905 lt 119879recovered + 119879HDelay 119879HDelay isin (0 3th day]Isolation duration 119868solation(119860 119894) 119879Intervention lt 119905 lt 119879Intervention + 119879IDuration 119879IDuration isin (0 14th day]Close workspace duration 119862lose(119864119894) 119879Intervention lt 119905 lt 119879Intervention + 119879CWDuration 119864119894 isinWorkspace 119879CWDuration isin (0 7th day]Limitation of agent activity duration 119871 imitation(119860 119894) 119879Intervention lt 119905 lt 119879Intervention + 119879LAADuration 119879LAADuration isin (0 7th day]119877Vaccination means the vaccination rate of artificial Beijing while 119877Antibiotic means the antibiotic rate used for infected agents in hospital119867ospital (119860119894) means that the infected agent 119860119894 is sent to hospital he or she will receive the retreatment immediately119868solation (119860119894) means that the infected agent 119860119894 is isolated at home he or she can only contact with the other agents until isolation duration is over119862lose (119864119894) means that the environment 119864119894 typed of workspace is closed until the close workspace duration is over119871 imitation (119860119894) means that the activities of agent 119860119894 are limited only the workspace and home are allowed until limitation duration is over
Incubation Weibulldistribution
Symbolic LogNormaldistribution
Recovering LogNormaldistribution
H1N1influenza
Settings of H1N1 influenza model
Disease model
Transmissionmodel
Figure 20 The settings of H1N1 in EECT
It is worth notifying that the settings of interventionmeasures are also influenced by OsdRT For example theH1N1 outbreak areas are notified by OsdRTThe interventionmeasures such as close workspace are used only in theoutbreak areas Therefore the unnecessary computation ofintervention measures is avoided So the data injection byOsdRT not only makes artificial society approach actualsociety but also increases the performance of computationalexperiments
53 Experiments by EMT When the initialization of arti-ficial Beijing is ready EMT is used to do the computa-tional experiments EMT is composed of two parts thecentral controller in the central node and the residen-tial service deployed in the computational nodes Centralcontroller loads the experiment plans and controls the whole
experiments process the GUI of controller is shown inFigure 22 EMT manages the computational experimentsin four steps cluster configuration load experiment plansmodels deployment and running experiments
531 Cluster Configuration Before the configuration clusterinformation is collected from the residential services insidethe cluster The information includes nodes amount nodenames node IPs and CPU occupations It is used to quantifythe computing power of the clusterThe cluster configurationis based on this information The nodes are selected forexperiments Then the LP (local process) [51] number foreach node is set according to the number of the CPU kernelsAfter the configuration the experiment plans are loaded tocustomize the models and data
20 Mathematical Problems in Engineering
Streets of close workspaces
Limitations ofactivities
IsolationsRespirator
Closeworkspace
Antibiotic
VaccinationHospital
Surveillance
Figure 21 The settings of intervention measures in IMCT
Information ofexperiment plan
List of experiment plans
Start | Pause | Reset | Stop | Speedup | Slowdown | Back
NodeName | CPU occupation | Memory occupation |Network | Models | Status
Figure 22 The deployment and running of experiments in EMT
532 Load Experiment Plan As discussed in Section 35 thedescription of experiment plans is composed of the infor-mation in three aspects models and the initial data for theexperiments the deployment mapping tables from the mod-els to the nodes and the plan of experiment executionWhenthe experiment plan is loaded by the EMT the file namesand model descriptions of models and data are listed Withthe support of cluster configuration models can be mappedto the nodes in the GUI according to the computationrequirements EMT also provides the setting of experimentsitself including the running times start time and end time
533 Models and Data Deployment Models are stored inthe agent population and services repository According tothe settings in experiment plan the customized models aredownloaded from the repository first then the models suchas agent model environment model emergent events modeland intervention measures model are integrated to be thecompositionmodels of artificial BeijingThemodels and datalocations in the nodes are configured in the ldquomodel pathrdquoand ldquodata pathrdquo in the GUI of central controller Finallythe models are uploaded to the nodes in cluster with theconsideration of partition The partition of artificial society
Mathematical Problems in Engineering 21
models is different from the multiagent system in [52]Models in artificial society are divided into three levels agentsand environments grids and city Agents and environmentsare encapsulated in grids which compose the whole cityThus the partition of artificial society becomes the partitionof grids The parallelism degrees are found to support thepartition of models agent independence and grid indepen-dence According to the parallelisms agents are partitioned toLPs according to the population distribution in grids Agentmovements among grids are simulated by the movementevents among grids in different LPs As a result a two-tierparallel architecture is proposed to support the simulation ofartificial society The architecture is detailed in [53]
534 Running of Experiments When the preparation isready the experiments can be done by the central controllerof EMT The running and control commands of the experi-ments are grouped into three aspects start pause or stop theexperiments get the running information of the nodes suchas the snapshot of the desktop of node With the help of thesecommands users can run the computational experimentsto do the research on the emergency management It isworth notifying that the simulation engine is optimized inorder to support the ten million agent simulation UsingGPU as coprocessor a two-tier parallel simulation engine isdesigned with support of MPI and OpenCL through phasedsynchronization [54] One-sided communication is used forreflection of remote simulation objects and message passingbetween processes A general kernel function prototype iselaborately designed and conditionally compiled for execu-tion on both CPU and GPU Moreover the optimizationoperations like load balance is developed under the instruc-tion of activity-based simulation [55] The densely populatedgrids are given more CPU and memory resource accordingto the activity predictions
54 Results of H1N1 Experiments with KD-ACP On thebasis of artificial society models and the intervention mea-sures mentioned in Section 512 a series of computationalexperiments are performed to study the H1N1 influenza inBeijing In our case of artificial Beijing 19610000 agentsare simulated in the cluster within 48 CPU cores and 128Gmemories the nodes are connected by kilomega networks Ittakes 18 hours to simulate a 250-day disease spreading
In order to illustrate how the optimal plans are obtainedbyKD-ACP the experiments are divided into four groups theexperiments ofmodel validation the experiments of sensitiv-ity analysis of vaccination rate the experiments of sensitivityanalysis of isolation and the experiments of the combinationof interventionmeasuresThese four groups show the generalprocess of the research on emergency management in H1N1influenza Firstly the models of H1N1 are validated withthe support of historical data Secondly the traditionalmedical interventions of pandemic such as vaccination areanalyzed to find the optimal plan Thirdly the traditionalnonmedical interventions of pandemic such as isolation areanalyzed to find the optimal plan Finally the combination ofinterventions is tested to show the combined effect
It is worth mentioning that all the experiments in ourwork are performed 100 times These experiments are ini-tialized with different settings of interventions in IMCT Theresults shown in the figures are the mean values
541 The Experiments of Model Validation In our artifi-cial Beijing the models of H1N1 influenza are built basedon the existing researches Wang and her colleagues hadstudied the H1N1 influenza by SEIR (SusceptibleExposedInfectiveRecovered) models [56] With the support of theirwork we build the models of artificial Beijing in the mannerof multiagents
In Figure 23 the control group shown in red line isdrawn based on the realistic statistical data collected in H1N1influenza in Beijing in 2009 The influenza lasts more thansix months and more than 170000 people are infectedComparing with the control group the simulated resultswithout interventions are drawn in blue It is obvious that thesimulated case fits the control group well before 70 days Butthe ascending of control group is slowed after 70 days andthe peak value (17883) is much less than the simulated results(25135) It is because the vaccination is activated at day 67 incontrol group according to the intervention measures usedin 2009 The restrain by vaccination becomes effective eightdays later The change of infected number after vaccinationis shown by red spotted line The trend of control group isdifferent from the simulation data after 75 days
However the simulated results (blue line) in compu-tational experiments fit the control group (as shown inFigure 23) in the first 70 days when no intervention mea-sures are executed The key features such as effective basicreproduction number (119877
0) are also almost the same in both
cases Therefore it can be concluded that the models ofartificial society built in microview are validated by therealistic statistical data in influenza in Beijing in 2009
In this group of experiments no interventions are exe-cuted in order to validate the models Therefore the simu-lated results of this extreme case which cannot happen todaycould be obtained As shown by blue line the number ofinfected agents without interventions reaches peak value atday 91 with 25135 consisting of 8052 in symptomatic phaseand 28950 in recovery phase It is obvious that the totalnumber of infected agents grows slowly at the beginningtime Then the number increases quickly when the influenzabreaks and it reaches its maximum value with 383870 in theend This phenomenon can be explained as in the followingAt the beginning of influenza due to the limited numberof infected agents the propagation of the disease remainsslow but many agents become susceptible When the numberof infected agents increases rapidly the spreading becomessignificant However the number of infected agents decreasesafter day 91 It is because the disease spreading relies on thesocial networks and spatial contact networks [49] It meansthat the infected agents can only infect agents either in thesocial relationship networks or in the spatially contacted
542 The Experiments of Sensitivity Analysis of VaccinationRate According to Section 512 vaccination is a typical
22 Mathematical Problems in Engineering
0
Infe
cted
agen
ts
0 50 100 150 200 250
Statistical dataSimulation data
25000
20000
15000
10000
5000
Day d
Figure 23 The results with no intervention executed and the com-parison with realistic statistical data in Beijing in 2009
medical intervention in influenza Traditionally the effective-ness of vaccination can only be evaluated after the influenzaand the key settings such as vaccination rate are determinedby the experience It is always difficult for emergency man-agement organizations to find the most optimal vaccinationrate in advanceTherefore the sensitivity of vaccination rate isanalyzed in this group of experiments The settings are listedin Table 7 average results are shown in Figure 24
Vaccination rates are set linearly (10 30 50 and70) to test the changes of influenza The peak value ofinfected agents decreases from 25135 22392 15787 6907to 638 respectively It shows that the more the agents arevaccinated the smaller the peak value of the infected agentswill be In the case of 70 vaccination the total infected agentis only 13799 It means that the influenza in this case does nothappen due to the high vaccination rate
The change of peak values also shows an interestingphenomenon Vaccination rates increase linearly the peakvalues decrease slower than the ascending of vaccination ratesat beginning but the descending becomes more and morefaster with the linear increase of vaccination rate It meansthat the small scale of vaccination (less than 30) does notmake senseWith the increase of vaccination rate the numberof susceptible agents decreases significantly In sequencethe possibility of infections is lowered greatly Additionallyit could be found that the decrease of peak values comeslater than the increase of the vaccination rates This is quitereasonable according to Mei et alrsquos work in [57]
As a result it is concluded that high vaccination rate iseffective in the control of H1N1 influenza though costly
543 The Experiments of Sensitivity Analysis of IsolationAccording to Section 512 isolation is a typical nonmedicalintervention It is the most used intervention measure in
Table 7 The initial settings of experiments of vaccination
Parameters ValueAgent count 19610000Initial infected count 40Interventions activated time(119879Intervention)
0
Vaccination rate (119877Vaccination) 0 10 30 50 70
0
Infe
cted
agen
ts
Loading intervention of vaccination
No vaccination10 vaccination30 vaccination
50 vaccination70 vaccination
0 50 100 150 200 250
25000
20000
15000
10000
5000
Day d
Figure 24 The results of loading intervention of vaccination
the emergency management in influenza As mentionedbefore a problem also exists that the measure cannot bevalidated before the intervention Therefore the sensitivityof isolation is analyzed in this group of experiments It isworth mentioning that the isolation is usually combinedwith the hospital policy It means the infected individual issent to hospital meanwhile the people who contact withthe infected are all isolated As a result parameters delay tohospital and isolation duration are designed together in thesettings of experiment Delay to hospital stands for the periodfrom infection time to hospital time while Isolation durationrepresents the period when an agent cannot contact others ifhe has contact with an infected agent The settings of theseparameters are listed in Table 8 average results are shown inFigure 25
Firstly the parameter delay to hospital is simulated aloneto test the sensitivity The simulated results are shown insolid line in Figure 25 The peak value decreases from 6708to 2218 when the delay to hospital decreases from 15 daysto 1 day The half day change brings the 67 reductionMoreover the peak value almost approximates to zero inthe case of 05 days The influenza does not happen in thiscase It can be concluded that the earlier the infected agents
Mathematical Problems in Engineering 23
Table 8 The initial settings of experiments of isolation
Parameters ValueAgent count 19610000Initial infected count 40Interventions activated time (119879Intervention) 80Delay to hospital (119879HDelay) isolation duration (119879IDuration) (15 0) (1 0) (05 0) (15 3) (1 3) (05 3)
0 50 100 150 200 2500
1000
2000
3000
4000
5000
6000
7000
Infe
cted
agen
ts
Loading intervention of isolation
Delay to hospital = 15
Delay to hospital = 10
Delay to hospital = 05
Delay to hospital = 15 isolation duration = 3
Delay to hospital = 10 isolation duration = 3
Delay to hospital = 05 isolation duration = 3
Day d
Figure 25 The results of loading intervention of isolation
are sent to hospital the less the agents will be infectedSo the construction of effective mechanism to find infectedindividuals is important in the emergency management ofinfluenza
Secondly the parameter isolation duration is added tosimulate the isolation of interventionObviously the isolationmeasure decreases the infected agents greatly Compared tothe parameter delay to hospital of 15-day case the peak valuedecreases from 6708 to 362when the isolation is added in theinterventionThe reduction of 6346 agents comes from the 3-day isolation of agents who had contacted with the infectedagents However the decrease of parameter delay to hospitalcannot bring the similar reduction in the 3-day isolationcases According to the figure the peak value decreases from362 to 51 from 15-day case to 05-day case only 311 agents aresaved from infection As a result the sensitivity of simulatedresults does not change obviously with delay to hospital inthe case of isolation It can be concluded that isolation is amore effective measure Though it is not easy for emergencymanagement organizations to decrease the delay to hospitaldue to the current monitoring mechanism the influenza canstill be controlled by isolation
544 The Experiments of Combination of Intervention Mea-sures Isolation is not the only nonmedical intervention ininfluenza close workspace is another common measureWhen close workspace is activated the workspaces are allclosed agents can only stay home at the closed time Luckilyisolation and close workspace are independent from oneanother It is possible for emergency management organi-zations to use these two measures together So this groupof experiments is designed to obtain the effectiveness of thecombination of intervention measures The parameters areinitialized with different settings listed in Table 9
Parameters delay to hospital and isolation duration is setin the middle values (1 day 3 days) according to the lastsection According to the experience close duration is set asseven days It is known that close workspace is a really costintervention measure the activated time is a key parameterWith the fixed parameters mentioned before the changeof interventions activated time is analyzed in this group ofcomputational experimentsThe average results are shown inFigure 26
Compared with the simulated results of experimentsdiscussed above Figure 26 strongly illustrates that influenzais ceased when the intervention measures are combinedtogether The peak value of infected agents is 141 in the ldquo80thdays of close-workspace and isolation with 1 day delay tohospitalrdquo case while 25135 in ldquono interventionrdquo case 2218 inldquo1 day delay to hospitalrdquo case 203 in ldquo1 day delay to hospitaland isolationrdquo case The interventions bring a great reductionof infection the peak values decrease from ten thousandslevel to hundred level
On the other hand the peak value even decreases to 108when the execution time of close workspace is 10 days earlierThe date of maximum value of infected agents is advancedfrom the 135th day to the 108th day The whole circle of H1N1influenza is reducedThe results of this group of experimentsshow that intervention measures should be activated as soonas possible and the compositions of interventions are moreeffective in emergency management of H1N1 influenza
In summary based on the general process lots of exper-iments could be done by KD-ACP to study emergency man-agement in H1N1 influenza Many conclusions are obtainedbased on the analysis of the experiments in this sectionAccording to these conclusions themost optimal response inour case study is the case of ldquo70th days of close-workspace andisolation with 1 day delay to hospitalrdquo It is because this casehas the fewest infected agents relatively But the modelling ofH1N1 influenza is not sufficient Economic and social factorsof interventions such as the money cost by vaccination andisolation the influence of the close-workspace in society
24 Mathematical Problems in Engineering
Table 9 The initial settings of experiments of isolation and closeworkspace
Parameters ValueAgent count 19610000Initial infected count 40Interventions activated time (119879Intervention) 70th day 80th dayDelay to hospital (119879HDelay) 1 dayIsolation duration (119879IDuration) 3 daysClose workspace duration (119879CWDuration) 7 days
50 100 150 200 2500
50
100
150
200
Infe
cted
agen
ts
Loading intervention of isolation and close workspace
Delay to hospital = 1 isolation duration = 3Start close time = 70 close duration time = 7Delay to hospital = 1 isolation duration = 3Start close time = 80 close duration time = 7
Day d
Figure 26The results of loading interventions of isolation and closeworkspace
and even opinion transmission of H1N1 in internet are notconsidered in our case study Without the analysis of thesefactors it does not make sense to talk about the most optimalintervention measures However KD-ACP had provided asoftware framework and general process of the research onemergency management Supported by KD-ACP the expertsof emergency management are able to study emergent eventssuch as H1N1 influenza as detailed as possible Based on theexperiments and analysis again and again the most optimalintervention measures will be found ultimately
55 Parallel Execution by ASST and OsdRT As discussed inSection 35 ASST andOsdRT are used to support the parallelexecution of artificial Beijing The statistical data is collectedand stored in artificial society runtime database at runtimeASST downloads the infection data from the database anddisplays it in the map As shown in Figure 27 the infectedagents are distributed in the districts in BeijingThe snapshotgives the geospatial distribution of the H1N1 epidemic Thelegends of the agents in different status are listed in the left
right corner in Figure 27 Five statuses are shown in the figurehealth without vaccination health with vaccination incuba-tion being symptomatic and convalescence ASST providesa very direct feeling for the emergency experts Partly withthe situation from ASST Emergency Decision Organizationmakes the emergency response plans to the response of theemergency scenario With the help of plans the epidemicin actual society can be restrained In the meantime theinfluenza information is reflected in the Internet Accordingto Figure 8 the data acquisition data extraction and datastandardization of H1N1 influenza are processed in OsdRTOsdRT is implemented by Dr Caorsquos team from Institute ofAutomation from Chinese Academy of Sciences [58] Thetool provides not only the reports of H1N1 SARS hand-foot-mouth disease and intestinal diseases are all monitored Theweb page is shown in Figure 28 H1N1 reports from Internetare collected first Based on the data extraction of the reportsH1N1 case of distribution is obtained and shown in theincidencemap As a result H1N1 incidence data is regarded asthe knowledge which is sent to EECT and IMCTThe settingsof emergent events and intervention measures are modifiedin order to make artificial society approach actual society innext turn of computational experiments With the help ofOsdRT the loop of parallel execution process integrates ACPapproach
6 Conclusion
This work provides an integrated software framework forsocial Computing in emergencymanagement From a systemperspective KD-ACP provides a reliable flexible low costand effective platform for the scientists to do the research onemergency response problems The analysis and predictionof these problems with inherent complexity can be solvedby the repetition of possible alternative experiments onthe software framework Therefore KD-ACP which is anattempt to implement ACP approach seemed as an effectivecomputational framework to support the decision making inemergency response
Currently KD-ACP is used to study the H1N1 epidemicin Beijing in 2009 The most optimal compositions of inter-vention measures are testified to support the response tonext influenza However it is still a long way ahead beforethe automation systematization and practicalization of KD-ACP A lot of work will be carried out along several directionsas in the following
(i) The parallel execution process in KD-ACP is actuallynot strongly connected with the emergency responseorganization Traditional experience and theory stillplay a dominant role in the decision of emergencymanagement It is necessary to propose a mechanismof hall for workshop to facilitate the parallel executionloop in actual society
(ii) Section 33 introduces the principle and process ofmodelling artificial society Metamodels and modelsof artificial Beijing are built by GME Model transfor-mation and code generation are used to implementcodes from FSM based models But many details
Mathematical Problems in Engineering 25
Fangshan district Daxing district Fengtai district Xicheng district
Mentougou district Haidian districtShijingshan district Dongcheng district
Incubation
Symptomatic
Convalescence
Health withoutvaccination
Health withvaccinationLegend of agent
health status
Figure 27 The situation of artificial Beijing by ASST
H1N1
Districts of Beijing
H1N1 case ofdistribution
Incidence map
Distribution ofH1N1 reportsfrom internet
Incidence of H1N1
Fangshan districtLiangxiang town
Population 19000Incidence 135789
(1100000)
Figure 28 The implementation of OsdRT by Institute of Automation from Chinese Academy of Sciences
are still not illustrated due to the topic of thispaper The key techniques such as template of codeframework and model transformations from specificdomains to FSM will be focused and introduced indetail in our next paper
(iii) The performance of computational experiments inKD-ACP is still slower than expectation The opti-mization of simulation engine and models structureis necessary to improve the performance So it ispossible to run artificial society with the real worldin parallel
In summary KD-ACP paves a new way for scientists inboth emergencymanagement and computation simulation to
collaborate with each other in solving the emergency man-agement problems by social computing
Research Highlights
(i) KD-ACP is implemented based on ACP approach
(ii) Artificial Beijing is built with the help of KD-ACP
(iii) H1N1 is modeled to simulate the influenza in Beijingin 2009
(iv) Computational experiments testify the effectivenessof intervention measures
26 Mathematical Problems in Engineering
Conflict of Interests
The authors declare that there is no conflict of interestsregarding the publication of this paper
Acknowledgment
This work was supported in part by the National ScienceFoundation of China (nos 91024030 71303252 61403402and 91324013)
References
[1] W Mao and F-Y Wang Advances in Intelligence and SecurityInformatics Academic Press Oxford UK 2012
[2] E Bonabeau ldquoAgent-based modeling methods and techniquesfor simulating human systemsrdquo Proceedings of the NationalAcademy of Sciences of the United States of America vol 99 no3 pp 7280ndash7287 2002
[3] K M Carley D B Fridsma E Casman et al ldquoBioWar scalableagent-basedmodel of bioattacksrdquo IEEETransactions on SystemsMan and Cybernetics A Systems and Humans vol 36 no 2 pp252ndash265 2006
[4] J M Epstein ldquoModelling to contain pandemicsrdquo Nature vol460 no 7256 p 687 2009
[5] Los Alamos National Laboratory Reports about EPISIMS LosAlamos National Laboratory 2003
[6] C H Builder and S C Bankes ldquoArtificial societies a conceptfor basic research on the societal impacts of informationtechnologyrdquo RAND Report 149 RAND Corp 1991
[7] F-Y Wang ldquoComputational theory and method on complexsystemrdquo China Basic Science vol 6 no 5 pp 3ndash10 2004
[8] F-Y Wang ldquoToward a paradigm shift in social computing theACP approachrdquo IEEE Intelligent Systems vol 22 no 5 pp 65ndash67 2007
[9] F-Y Wang and S Tang ldquoArtificial societies for integratedand sustainable development of metropolitan systemsrdquo IEEEIntelligent Systems vol 19 no 4 pp 82ndash87 2004
[10] D Wen Y Yuan and X-R Li ldquoArtificial societies computa-tional experiments and parallel systems an investigation ona computational theory for complex socioeconomic systemsrdquoIEEE Transactions on Services Computing vol 6 no 2 pp 177ndash185 2013
[11] F Wang F Zeng and Y Yuan ldquoAn ACP-based approach forcomplexity analysis of E-commerce systemrdquo Complex Systemsand Complexity Science vol 5 no 3 pp 1ndash8 2008 (Chinese)
[12] F-Y Wang and P K Wong ldquoResearch commentary intelligentsystems and technology for integrative and predictivemedicinean ACP approachrdquoACMTransactions on Intelligent Systems andTechnology vol 4 no 2 pp 1ndash6 2013
[13] J Sifeng X Gang F Dong and H Chunpeng ldquoStudy on theemergency rescue decision support system of petrochemicalplant based on ACP theoryrdquo in Proceedings of the 6th Manage-ment Annual Meeting (MAM rsquo11) 2011 (Chinese)
[14] X Dong D Fan G Xiong F Zhu Z Zhang and Y Yao ldquoPar-allel Bus Rapid Transit (BRT) operation management systembased on ACP approachrdquo in Proceedings of the 6th ManagementAnnual Meeting (MAM rsquo11) Beijing China 2011
[15] B Ning F-Y Wang H-R Dong R-M Li D Wen and L LildquoParallel systems for Urban rail transportation based on ACP
approachrdquo Journal of Transportation Systems Engineering andInformation Technology vol 10 no 6 pp 23ndash28 2010 (Chinese)
[16] F-Y Wang J Zhao and S-X Lun ldquoArtificial power systemsfor the operation and management of complex power gridsrdquoSouthern Power System Technology vol 2 no 3 pp 1ndash11 2008(Chinese)
[17] M J North T R Howe N T Collier and J R Vos ldquoRepastsimphony development environmentrdquo in Proceedings of theAgent 2005 Conference on Generative Social Processes Modelsand Mechanisms 2005
[18] M J North T RHowe N T Collier and J R Vos ldquoRepast Sim-phony runtime systemrdquo in Proceedings of the Agent Conferenceon Generative Social Processes Models and Mechanisms 2005
[19] SWARM Development Group Swarm 211 Reference Guide2000 httpwwwswarmorg
[20] S Tisue and U Wilensky ldquoNetLogo a simple environment formodeling complexityrdquo in International Conference on ComplexSystems pp 16ndash21 2004
[21] S Luke C Cioffi-Revilla L Panait K Sullivan and G BalanldquoMASON a multiagent simulation environmentrdquo Simulationvol 81 no 7 pp 517ndash527 2005
[22] M Ichikawa H Tanuma Y Koyama andH Deguchi ldquoSOARSintroduction as a social microscope for simulations of socialinteractions and gamingrdquo in Organizing and Learning throughGaming and Simulation Proceedings of the 38th Conference ofthe International Simulation and Gaming Association (ISAGArsquo07) pp 149ndash158 2007
[23] D S Burke J M Epstein D A T Cummings et al ldquoIndividual-based computational modeling of smallpox epidemic controlstrategiesrdquo Academic Emergency Medicine vol 13 no 11 pp1142ndash1149 2006
[24] F-Y Wang ldquoParallel control and management for intelligenttransportation systems concepts architectures and applica-tionsrdquo IEEE Transactions on Intelligent Transportation Systemsvol 11 no 3 pp 630ndash638 2010
[25] Z Molnar D Balasubramanian and A Ledeczi ldquoAn introduc-tion to the genericmodeling environmentrdquo in Proceedings of theTOOLS Europe 2007 Workshop on Model-Driven DevelopmentTool Implementers Forum Zurich Switzerland 2007
[26] S Kelly and J-P Tolvanen Domain Framework in Domain-Specific Modeling Enabling Full Code Generation JohnWiley ampSons Hoboken NJ USA 2007
[27] B P Zeigler H Praehofer and T G Kim Theory of Modelingand Simulation Integrating Discrete Event and ContinuousComplex Dynamic Systems Academic Press San Diego CalifUSA 2000
[28] G Booch J Rumbaugh and I JacobsonThe Unified ModelingLanguage User Guide Addison-Wesley Reading Mass USA1999
[29] F Wagner R Schmuki T Wagne and P Wolstenholme Mod-eling Software with Finite State Machines A Practical ApproachCRC amp Taylor amp Francis Boca Raton Fla USA 2006
[30] T Espiner ldquoChina builds worldrsquos fastest supercomputerrdquo ZDNetUK October 2010
[31] B Chen and G Guo ldquoA two-tier parallel architecture for arti-ficial society simulationrdquo in Proceedings of the ACMIEEESCS26th Workshop on Principles of Advanced and DistributedSimulation (PADS rsquo12) pp 184ndash186 July 2012
[32] G Guo B Chen X G Qiu and Z Li ldquoParallel simulation oflarge-scale artificial society on CPUGPU mixed architecturerdquo
Mathematical Problems in Engineering 27
in Proceedings of the ACMIEEESCS 26th Workshop on Prin-ciples of Advanced and Distributed Simulation (PADS rsquo12) pp174ndash177 July 2012
[33] S-X Lun ldquoResearch on the classification of parallel executionmodes of ACP theoryrdquo Acta Automatica Sinica vol 38 no 10pp 1602ndash1608 2012
[34] F-YWang ldquoA framework for social signal processing and anal-ysis from social sensing networks to computational dialecticalanalyticsrdquo Science China vol 43 no 12 pp 1598ndash1611 2013
[35] G GuoOne Model Userrsquos Manual version 12 National Univer-sity of Defense University 2011
[36] Y Ge R Meng Z Cao X Qiu and K Huang ldquoVirtual citymdashanindividual-based digital environment for human mobility andinteractive behaviorrdquo Simulation Transactions of the Society forthe Modeling and Simulation International 2014
[37] M Kretzschmar and R T Mikolajczyk ldquoContact profiles ineight European countries and implications for modelling thespread of airborne infectious diseasesrdquo PLoS ONE vol 4 no6 Article ID e5931 2009
[38] W Duan and X Qiu ldquoFostering artificial societies using sociallearning and social control in parallel emergency managementsystemsrdquo Frontiers in Computer Science vol 6 no 5 pp 604ndash610 2012
[39] W J Edmunds C J OrsquoCallaghan and D J Nokes ldquoWhomixes with whom Amethod to determine the contact patternsof adults that may lead to the spread of airborne infectionsrdquoProceedings of the Royal Society B Biological Sciences vol 264no 1384 pp 949ndash957 1997
[40] J Banks J S Carson B L Neison and D M Nicol Discrete-Event System Simulation Prentice Hall Englewood Cliffs NJUSA 4th edition 2007
[41] H Yongxia ldquoInvestigation and analysis of college studentsrsquo workand restrdquo Value Engineering vol 31 no 3 2012 (Chinese)
[42] W Duan Z Cao Y Wang et al ldquoAn ACP approach to publichealth emergency management using a campus outbreak ofh1n1 influenza as a case studyrdquo IEEE Transactions on SystemsMan and Cybernetics Part A Systems and Humans vol 43 no5 pp 1028ndash1041 2013
[43] HWHethcote ldquoThemathematics of infectious diseasesrdquo SIAMReview vol 42 no 4 pp 599ndash653 2000
[44] G Chowell E Shim F Brauer and et al ldquoModelling thetransmission dynamics of acute haemorrhagic conjunctivitisapplication to the 2003 outbreak in Mexicordquo Statistics inMedicine vol 25 no 11 pp 1840ndash1857 2006
[45] Y Ge L Liu B Chen X Qiu and K Huang ldquoAgent-basedmodeling for InfluenzaH1N1 in an artificial classroomrdquo SystemsEngineering Procedia vol 2 pp 94ndash104 2011
[46] A R Tuite A L Greer MWhelan et al ldquoEstimated epidemio-logic parameters andmorbidity associatedwith pandemicH1N1influenzardquo CMAJ vol 182 no 2 pp 131ndash136 2010
[47] L Brouwers B Cakici M Camitz A Tegnell and MBoman ldquoEconomic consequences to society of pandemic H1N1influenza 2009mdashpreliminary results for Swedenrdquo Euro Surveil-lance vol 14 no 37 pp 1ndash7 2009
[48] Y Ge L Liu X Qiu H Song Y Wang and K Huang ldquoAframework of multilayer social networks for communicationbehavior with agent-based modelingrdquo Simulation Transactionsof the Society for the Modeling and Simulation International vol89 no 7 pp 810ndash828 2013
[49] W Duan X Qiu Z Cao X Zheng and K Cui ldquoHeterogeneousand stochastic agent-based models for analyzing infectious
diseasesrsquo super spreadersrdquo IEEE Intelligent Systems vol 28 no4 pp 18ndash25 2013
[50] The Preparation and Emergency Response Plans to PandemicPublished by Ministry of Health Ministry of Health 2005(Chinese)
[51] R M Fujimoto Parallel and Distributed Simulation SystemsWiley Interscience 2000
[52] F Cicirelli A Giordano and L Nigro ldquoDistributed simulationof situated multi-agent systemsrdquo in Proceedings of the 15thIEEEACM International Symposium on Distributed Simulationand Real Time Applications (DS-RT rsquo11) pp 28ndash35 September2011
[53] B Chen and G Guo ldquoA two-tier parallel architecturefor artificial society simulationrdquo in Proceedings of the 26thACMIEEESCS Workshop on Principles of Advanced and Dis-tributed Simulation (PADS rsquo12) pp 184ndash186 July 2012
[54] G Guo B Chen and X Qiu ldquoParallel simulation of large-scale artificial society with GPU as coprocessorrdquo InternationalJournal of Modeling Simulation and Scientific Computing vol4 no 2 Article ID 1350005 2013
[55] B Chen L-B Zhang X-C Liu and H Vangheluwe ldquoActivity-based simulation using DEVS increasing performance by anactivity model in parallel DEVS simulationrdquo Journal of ZhejiangUniversity Science C vol 15 no 1 pp 13ndash30 2014
[56] X-L Wang P Yang Z-D Cao et al ldquoQuantitative evaluationon the effectiveness of prevention and control measures againstpandemic influenza A (H1N1) in Beijing 2009rdquoChinese Journalof Epidemiology vol 31 no 12 pp 1374ndash1378 2010 (Chinese)
[57] S Mei A D van de Vijver L Xuan Y Zhu and P M ASloot ldquoQuantitatively evaluating interventions in the influenzaA (H1N1) epidemic on China campus grounded on individual-based simulationsrdquo Procedia Computer Science vol 1 no 1 pp1675ndash1682 2010
[58] X Qiu Z Cao and Z Li Mid-Term Reports of China NationalScience Foundation 91024030 2014
Submit your manuscripts athttpwwwhindawicom
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
MathematicsJournal of
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Mathematical Problems in Engineering
Hindawi Publishing Corporationhttpwwwhindawicom
Differential EquationsInternational Journal of
Volume 2014
Applied MathematicsJournal of
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Probability and StatisticsHindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Journal of
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Mathematical PhysicsAdvances in
Complex AnalysisJournal of
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
OptimizationJournal of
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
CombinatoricsHindawi Publishing Corporationhttpwwwhindawicom Volume 2014
International Journal of
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Operations ResearchAdvances in
Journal of
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Function Spaces
Abstract and Applied AnalysisHindawi Publishing Corporationhttpwwwhindawicom Volume 2014
International Journal of Mathematics and Mathematical Sciences
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
The Scientific World JournalHindawi Publishing Corporation httpwwwhindawicom Volume 2014
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Algebra
Discrete Dynamics in Nature and Society
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Decision SciencesAdvances in
Discrete MathematicsJournal of
Hindawi Publishing Corporationhttpwwwhindawicom
Volume 2014 Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Stochastic AnalysisInternational Journal of
16 Mathematical Problems in Engineering
Table 4 The description of artificial Beijing generated by ASE
Statistical features ofartificial Beijing Descriptions of statistical features Data sourcedefault values
Agent roles Baby child primary student middle school student college student worker retired
Demographic data file ofBeijing
Environment types Super market store restaurant park stadium hospital school cliniccommercial building dormitory building residential building
Demographic data file ofBeijing
Demographic data Age distribution of population sex distribution of population householddistribution of population retired age distribution children distribution
Demographic data file ofBeijing
Environment data Distribution of all types of environments Environment distributiondata file of Beijing
Relationships Classmate relative college friendship family Demographic data file ofBeijing
Varying parameters
Age difference of couples119873(120583120590) 119873(03) isin [0 12]Minimummarriage age (male female) (22 20)Age difference of children119873(120583120590) 119873(13) isin [1 10]Enrollment rate of kindergarten 120572 08Enrollment rate of primary school 120573 1Enrollment rate of middle school 120574 1Enrollment rate of college 120575 08Entrance age of kindergarten 1205721015840 3Entrance age of primary school 1205731015840 7Entrance age of middle school 1205741015840 13Entrance age of middle school 1205751015840 19
Age distributioninformation table
Agent settable
Geospatialenvironment set table
Household settable
Figure 15 The geospatial and population database of Beijing city
51 Modelling Emergency Management The modelling ofemergency management consists of two parts the modellingof emergent events and the modelling of intervention mea-sures H1N1 and the intervention measures for H1N1 aremodeled in our case
511 Modelling H1N1 by MDT H1N1 Model in agent simu-lates the states transitions of health status and the relevantactions Referring to SIR (SusceptibleInfectiveRecovered)[43 44] models agent has three health statuses susceptibleinfected and healthy with immunity Only susceptible agentcan be infected by the contact with infected agents After
the three infected phases (incubation being symptomaticand recovery) of H1N1 model agent is healthy again It isworthy mentioning that the recovered agents can be infectediteratively when the immunity disappears gradually by timeThe state transitions are illustrated in Figure 18
According to [45] the latent period of H1N1 was in aWeibull distribution within people from one to seven days[46 47] The distribution was usually centered in the rangebetween one and three days AWeibull random variable [40]is used to model the latent period as described in
119879Lat = 120572 times [minus ln (1 minus 120574)]1120573+ 120592 (119879Lat ge 120592) (5)
Mathematical Problems in Engineering 17
Censusdata
Generate GUID
Assign individualhousehold
Generate familystructure
Assign family roleto every member
Endow individualwith age gender
social role
Distribute environmentlocations on the map
School Workplace Entertainmentlocations
Consumptionlocations
Assign school orworkplace to
individual
Correlatedindividual withother locations
Generate friendshipin school
Generate coworkerrelations inworkplace
Assign individualadministrative
region
Generate housebuildings
Locate householdto house building
Generate familyrelation
Generateneighborhood
relation
PopulationEnvironment
Environment listSocial network
Figure 16 The algorithm of generating geospatial and social environment data [36]
Edit
Edit
Edit
Edit
Edit
Edit
BrowseStatistical characteristic of population
Geospatial population data
Population data of street
Population types
Geospatial and social environment generation tool
Workspaces
Parameters distribution settings
Household types
Roles in household
Household structure
Social roles
BrowseBrowseBrowseBrowseBrowseBrowse
Population distribution data
File path
Retired distributionPop distributionHousehold numHousehold distributionSex distributionAge distributionData type
Figure 17 The GUI of Population and Geospatial Environment generation Tool (PGET)
18 Mathematical Problems in Engineering
Infected
SusceptibleHealth
withimmunity
Infected
by contac
t
with in
fected
agents
Recovery
Increase before beingsymptomatic symptomatic
Decrease after being
Decrease by time
Infectivity
Immunity
Figure 18 The transitions of agent health status
119879Lat denotes the duration of latent period 120592 120572 and 120573 arethe location parameter scale parameter and shape parameterof Weibull distribution respectively 120574 is a uniform randomnumber in the range [0 1] According to statistics [47] 120592 120572and 120573 are set as 0 18 and 121 respectively Then the mean(standard deviation) of latent period is calculated as 159 days(058 day2) Furthermore infectious period is set as 7 days
A susceptible studentrsquos probability of getting infecteddepended on the infectivity of infectious agents his ownimmunity level the duration of the contact action and soon The infectivity of an infected agent evolved with timeduring the infectious period According to the statistics ofinfectivity in the chart with red bars in Figure 19 an infectedstudent had the highest level of infectivity in the second dayafter he had the first symptom The infectivity levels in otherdays in comparison with the second day are in Table 5 Theday labeled as ldquominus1rdquo means the day before the starting of thesymptomatic period
Based on the algorithm discussed before theH1N1modelis developed with the help of MDT The settings of influenzamodel will be discussed in Section 521
512 Modelling Intervention Measures by MDT In corre-spondence intervention measures model is used to ceasethe outbreak of H1N1 As discussed in [42 48 49] if theappropriate measures are taken when the infectious diseaseemerges the transmission of the disease could be sloweddown and the damage could be decreased Therefore it isnecessary and important to design emergency interventionmeasures in the artificial Beijing According to the MinistryofHealth inChina the intervention policies [50] are designedto control the spread of influenza such as H1N1 The inter-ventionmeasures of these policies including are interventionsactivated time vaccination rate antibiotic delay to hospitalisolation duration close workspace duration and limitationof activities listed in Table 6 Similarly the models of theseintervention measures are also developed with the help ofMDT Using the reasonable data ranges listed in Table 6intervention models can be initialized to do the computa-tional experiments These experiments are used to show how
Symptomatic1 2 3 4 5 6 7minus1
Recovered
Tim
e (da
y)
Infe
ctiv
ity
Figure 19 The distribution of infectivity rate of an infected agent[37]
Table 5 The proportion of infectivity
Days (119894) minus1 1 2 3 4 5 6 6Proportion of infectivity 03 05 1 08 04 02 01 005
to restrain the H1N1 transmissionThe settings of these inter-vention measures in IMCT are introduced in Section 522
52 Experiment Settings
521 H1N1 Settings by EECT EECT is designed to initializethe emergent events models such as H1N1The settings comefrom the specified emergent event models As shown inFigure 20 in theH1N1 used case the settings are composed ofthe distribution of infection source infection rate infectionperiod and so on The infection rate changes with envi-ronments and infection periods The settings are initializedfrom both the research of medicine on H1N1 and the data ofH1N1 outbreak in Beijing in 2009The process of influenza inartificial Beijing is simulated to repeat the possible scenariosof H1N1 break During the computational experiments ofartificial Beijing theH1N1 settings such as the period of beingsymptomatic are updated by the dynamic data registrationsfrom OsdRT The updating corrects the H1N1 model duringthe experiments With the help of OsdRT artificial Beijing isable to approach actual society It is also an implementationof parallel execution
522 Intervention Measures Settings by IMCT In the H1N1case the settings of intervention measures are shown inFigure 21 in IMCT These measures are designed under theinstruction of Emergency Decision Organizations like ChinaDisease Center (CDC) The measures such as vaccinationrate the rate of closedworkspace and the starting time can beinitialized in the tool Emergency response plans are imple-mented in these settings As illustrated in Figure 7 all the pos-sible compositions within value settings in reasonable inter-vals are tested to find the most optimal emergency responseplan For example different compositions of measures suchas vaccination rates and rates of workspace closed are usedto design the plans of experiments By the analysis of theresults the best composition of interventionmeasures can beobtainedThe analysis will be discussed in case study in detail
Mathematical Problems in Engineering 19
Table 6 The intervention measures of influenza
Intervention measures ValuesInterventions activated Time 119879Intervention isin [0 80th day]Vaccination rate 119877Vaccination isin [0 1]Antibiotic rate 119877Antibiotic isin [0 1]Delay to hospital 119867ospital(119860 119894) 119879infected lt 119905 lt 119879recovered + 119879HDelay 119879HDelay isin (0 3th day]Isolation duration 119868solation(119860 119894) 119879Intervention lt 119905 lt 119879Intervention + 119879IDuration 119879IDuration isin (0 14th day]Close workspace duration 119862lose(119864119894) 119879Intervention lt 119905 lt 119879Intervention + 119879CWDuration 119864119894 isinWorkspace 119879CWDuration isin (0 7th day]Limitation of agent activity duration 119871 imitation(119860 119894) 119879Intervention lt 119905 lt 119879Intervention + 119879LAADuration 119879LAADuration isin (0 7th day]119877Vaccination means the vaccination rate of artificial Beijing while 119877Antibiotic means the antibiotic rate used for infected agents in hospital119867ospital (119860119894) means that the infected agent 119860119894 is sent to hospital he or she will receive the retreatment immediately119868solation (119860119894) means that the infected agent 119860119894 is isolated at home he or she can only contact with the other agents until isolation duration is over119862lose (119864119894) means that the environment 119864119894 typed of workspace is closed until the close workspace duration is over119871 imitation (119860119894) means that the activities of agent 119860119894 are limited only the workspace and home are allowed until limitation duration is over
Incubation Weibulldistribution
Symbolic LogNormaldistribution
Recovering LogNormaldistribution
H1N1influenza
Settings of H1N1 influenza model
Disease model
Transmissionmodel
Figure 20 The settings of H1N1 in EECT
It is worth notifying that the settings of interventionmeasures are also influenced by OsdRT For example theH1N1 outbreak areas are notified by OsdRTThe interventionmeasures such as close workspace are used only in theoutbreak areas Therefore the unnecessary computation ofintervention measures is avoided So the data injection byOsdRT not only makes artificial society approach actualsociety but also increases the performance of computationalexperiments
53 Experiments by EMT When the initialization of arti-ficial Beijing is ready EMT is used to do the computa-tional experiments EMT is composed of two parts thecentral controller in the central node and the residen-tial service deployed in the computational nodes Centralcontroller loads the experiment plans and controls the whole
experiments process the GUI of controller is shown inFigure 22 EMT manages the computational experimentsin four steps cluster configuration load experiment plansmodels deployment and running experiments
531 Cluster Configuration Before the configuration clusterinformation is collected from the residential services insidethe cluster The information includes nodes amount nodenames node IPs and CPU occupations It is used to quantifythe computing power of the clusterThe cluster configurationis based on this information The nodes are selected forexperiments Then the LP (local process) [51] number foreach node is set according to the number of the CPU kernelsAfter the configuration the experiment plans are loaded tocustomize the models and data
20 Mathematical Problems in Engineering
Streets of close workspaces
Limitations ofactivities
IsolationsRespirator
Closeworkspace
Antibiotic
VaccinationHospital
Surveillance
Figure 21 The settings of intervention measures in IMCT
Information ofexperiment plan
List of experiment plans
Start | Pause | Reset | Stop | Speedup | Slowdown | Back
NodeName | CPU occupation | Memory occupation |Network | Models | Status
Figure 22 The deployment and running of experiments in EMT
532 Load Experiment Plan As discussed in Section 35 thedescription of experiment plans is composed of the infor-mation in three aspects models and the initial data for theexperiments the deployment mapping tables from the mod-els to the nodes and the plan of experiment executionWhenthe experiment plan is loaded by the EMT the file namesand model descriptions of models and data are listed Withthe support of cluster configuration models can be mappedto the nodes in the GUI according to the computationrequirements EMT also provides the setting of experimentsitself including the running times start time and end time
533 Models and Data Deployment Models are stored inthe agent population and services repository According tothe settings in experiment plan the customized models aredownloaded from the repository first then the models suchas agent model environment model emergent events modeland intervention measures model are integrated to be thecompositionmodels of artificial BeijingThemodels and datalocations in the nodes are configured in the ldquomodel pathrdquoand ldquodata pathrdquo in the GUI of central controller Finallythe models are uploaded to the nodes in cluster with theconsideration of partition The partition of artificial society
Mathematical Problems in Engineering 21
models is different from the multiagent system in [52]Models in artificial society are divided into three levels agentsand environments grids and city Agents and environmentsare encapsulated in grids which compose the whole cityThus the partition of artificial society becomes the partitionof grids The parallelism degrees are found to support thepartition of models agent independence and grid indepen-dence According to the parallelisms agents are partitioned toLPs according to the population distribution in grids Agentmovements among grids are simulated by the movementevents among grids in different LPs As a result a two-tierparallel architecture is proposed to support the simulation ofartificial society The architecture is detailed in [53]
534 Running of Experiments When the preparation isready the experiments can be done by the central controllerof EMT The running and control commands of the experi-ments are grouped into three aspects start pause or stop theexperiments get the running information of the nodes suchas the snapshot of the desktop of node With the help of thesecommands users can run the computational experimentsto do the research on the emergency management It isworth notifying that the simulation engine is optimized inorder to support the ten million agent simulation UsingGPU as coprocessor a two-tier parallel simulation engine isdesigned with support of MPI and OpenCL through phasedsynchronization [54] One-sided communication is used forreflection of remote simulation objects and message passingbetween processes A general kernel function prototype iselaborately designed and conditionally compiled for execu-tion on both CPU and GPU Moreover the optimizationoperations like load balance is developed under the instruc-tion of activity-based simulation [55] The densely populatedgrids are given more CPU and memory resource accordingto the activity predictions
54 Results of H1N1 Experiments with KD-ACP On thebasis of artificial society models and the intervention mea-sures mentioned in Section 512 a series of computationalexperiments are performed to study the H1N1 influenza inBeijing In our case of artificial Beijing 19610000 agentsare simulated in the cluster within 48 CPU cores and 128Gmemories the nodes are connected by kilomega networks Ittakes 18 hours to simulate a 250-day disease spreading
In order to illustrate how the optimal plans are obtainedbyKD-ACP the experiments are divided into four groups theexperiments ofmodel validation the experiments of sensitiv-ity analysis of vaccination rate the experiments of sensitivityanalysis of isolation and the experiments of the combinationof interventionmeasuresThese four groups show the generalprocess of the research on emergency management in H1N1influenza Firstly the models of H1N1 are validated withthe support of historical data Secondly the traditionalmedical interventions of pandemic such as vaccination areanalyzed to find the optimal plan Thirdly the traditionalnonmedical interventions of pandemic such as isolation areanalyzed to find the optimal plan Finally the combination ofinterventions is tested to show the combined effect
It is worth mentioning that all the experiments in ourwork are performed 100 times These experiments are ini-tialized with different settings of interventions in IMCT Theresults shown in the figures are the mean values
541 The Experiments of Model Validation In our artifi-cial Beijing the models of H1N1 influenza are built basedon the existing researches Wang and her colleagues hadstudied the H1N1 influenza by SEIR (SusceptibleExposedInfectiveRecovered) models [56] With the support of theirwork we build the models of artificial Beijing in the mannerof multiagents
In Figure 23 the control group shown in red line isdrawn based on the realistic statistical data collected in H1N1influenza in Beijing in 2009 The influenza lasts more thansix months and more than 170000 people are infectedComparing with the control group the simulated resultswithout interventions are drawn in blue It is obvious that thesimulated case fits the control group well before 70 days Butthe ascending of control group is slowed after 70 days andthe peak value (17883) is much less than the simulated results(25135) It is because the vaccination is activated at day 67 incontrol group according to the intervention measures usedin 2009 The restrain by vaccination becomes effective eightdays later The change of infected number after vaccinationis shown by red spotted line The trend of control group isdifferent from the simulation data after 75 days
However the simulated results (blue line) in compu-tational experiments fit the control group (as shown inFigure 23) in the first 70 days when no intervention mea-sures are executed The key features such as effective basicreproduction number (119877
0) are also almost the same in both
cases Therefore it can be concluded that the models ofartificial society built in microview are validated by therealistic statistical data in influenza in Beijing in 2009
In this group of experiments no interventions are exe-cuted in order to validate the models Therefore the simu-lated results of this extreme case which cannot happen todaycould be obtained As shown by blue line the number ofinfected agents without interventions reaches peak value atday 91 with 25135 consisting of 8052 in symptomatic phaseand 28950 in recovery phase It is obvious that the totalnumber of infected agents grows slowly at the beginningtime Then the number increases quickly when the influenzabreaks and it reaches its maximum value with 383870 in theend This phenomenon can be explained as in the followingAt the beginning of influenza due to the limited numberof infected agents the propagation of the disease remainsslow but many agents become susceptible When the numberof infected agents increases rapidly the spreading becomessignificant However the number of infected agents decreasesafter day 91 It is because the disease spreading relies on thesocial networks and spatial contact networks [49] It meansthat the infected agents can only infect agents either in thesocial relationship networks or in the spatially contacted
542 The Experiments of Sensitivity Analysis of VaccinationRate According to Section 512 vaccination is a typical
22 Mathematical Problems in Engineering
0
Infe
cted
agen
ts
0 50 100 150 200 250
Statistical dataSimulation data
25000
20000
15000
10000
5000
Day d
Figure 23 The results with no intervention executed and the com-parison with realistic statistical data in Beijing in 2009
medical intervention in influenza Traditionally the effective-ness of vaccination can only be evaluated after the influenzaand the key settings such as vaccination rate are determinedby the experience It is always difficult for emergency man-agement organizations to find the most optimal vaccinationrate in advanceTherefore the sensitivity of vaccination rate isanalyzed in this group of experiments The settings are listedin Table 7 average results are shown in Figure 24
Vaccination rates are set linearly (10 30 50 and70) to test the changes of influenza The peak value ofinfected agents decreases from 25135 22392 15787 6907to 638 respectively It shows that the more the agents arevaccinated the smaller the peak value of the infected agentswill be In the case of 70 vaccination the total infected agentis only 13799 It means that the influenza in this case does nothappen due to the high vaccination rate
The change of peak values also shows an interestingphenomenon Vaccination rates increase linearly the peakvalues decrease slower than the ascending of vaccination ratesat beginning but the descending becomes more and morefaster with the linear increase of vaccination rate It meansthat the small scale of vaccination (less than 30) does notmake senseWith the increase of vaccination rate the numberof susceptible agents decreases significantly In sequencethe possibility of infections is lowered greatly Additionallyit could be found that the decrease of peak values comeslater than the increase of the vaccination rates This is quitereasonable according to Mei et alrsquos work in [57]
As a result it is concluded that high vaccination rate iseffective in the control of H1N1 influenza though costly
543 The Experiments of Sensitivity Analysis of IsolationAccording to Section 512 isolation is a typical nonmedicalintervention It is the most used intervention measure in
Table 7 The initial settings of experiments of vaccination
Parameters ValueAgent count 19610000Initial infected count 40Interventions activated time(119879Intervention)
0
Vaccination rate (119877Vaccination) 0 10 30 50 70
0
Infe
cted
agen
ts
Loading intervention of vaccination
No vaccination10 vaccination30 vaccination
50 vaccination70 vaccination
0 50 100 150 200 250
25000
20000
15000
10000
5000
Day d
Figure 24 The results of loading intervention of vaccination
the emergency management in influenza As mentionedbefore a problem also exists that the measure cannot bevalidated before the intervention Therefore the sensitivityof isolation is analyzed in this group of experiments It isworth mentioning that the isolation is usually combinedwith the hospital policy It means the infected individual issent to hospital meanwhile the people who contact withthe infected are all isolated As a result parameters delay tohospital and isolation duration are designed together in thesettings of experiment Delay to hospital stands for the periodfrom infection time to hospital time while Isolation durationrepresents the period when an agent cannot contact others ifhe has contact with an infected agent The settings of theseparameters are listed in Table 8 average results are shown inFigure 25
Firstly the parameter delay to hospital is simulated aloneto test the sensitivity The simulated results are shown insolid line in Figure 25 The peak value decreases from 6708to 2218 when the delay to hospital decreases from 15 daysto 1 day The half day change brings the 67 reductionMoreover the peak value almost approximates to zero inthe case of 05 days The influenza does not happen in thiscase It can be concluded that the earlier the infected agents
Mathematical Problems in Engineering 23
Table 8 The initial settings of experiments of isolation
Parameters ValueAgent count 19610000Initial infected count 40Interventions activated time (119879Intervention) 80Delay to hospital (119879HDelay) isolation duration (119879IDuration) (15 0) (1 0) (05 0) (15 3) (1 3) (05 3)
0 50 100 150 200 2500
1000
2000
3000
4000
5000
6000
7000
Infe
cted
agen
ts
Loading intervention of isolation
Delay to hospital = 15
Delay to hospital = 10
Delay to hospital = 05
Delay to hospital = 15 isolation duration = 3
Delay to hospital = 10 isolation duration = 3
Delay to hospital = 05 isolation duration = 3
Day d
Figure 25 The results of loading intervention of isolation
are sent to hospital the less the agents will be infectedSo the construction of effective mechanism to find infectedindividuals is important in the emergency management ofinfluenza
Secondly the parameter isolation duration is added tosimulate the isolation of interventionObviously the isolationmeasure decreases the infected agents greatly Compared tothe parameter delay to hospital of 15-day case the peak valuedecreases from 6708 to 362when the isolation is added in theinterventionThe reduction of 6346 agents comes from the 3-day isolation of agents who had contacted with the infectedagents However the decrease of parameter delay to hospitalcannot bring the similar reduction in the 3-day isolationcases According to the figure the peak value decreases from362 to 51 from 15-day case to 05-day case only 311 agents aresaved from infection As a result the sensitivity of simulatedresults does not change obviously with delay to hospital inthe case of isolation It can be concluded that isolation is amore effective measure Though it is not easy for emergencymanagement organizations to decrease the delay to hospitaldue to the current monitoring mechanism the influenza canstill be controlled by isolation
544 The Experiments of Combination of Intervention Mea-sures Isolation is not the only nonmedical intervention ininfluenza close workspace is another common measureWhen close workspace is activated the workspaces are allclosed agents can only stay home at the closed time Luckilyisolation and close workspace are independent from oneanother It is possible for emergency management organi-zations to use these two measures together So this groupof experiments is designed to obtain the effectiveness of thecombination of intervention measures The parameters areinitialized with different settings listed in Table 9
Parameters delay to hospital and isolation duration is setin the middle values (1 day 3 days) according to the lastsection According to the experience close duration is set asseven days It is known that close workspace is a really costintervention measure the activated time is a key parameterWith the fixed parameters mentioned before the changeof interventions activated time is analyzed in this group ofcomputational experimentsThe average results are shown inFigure 26
Compared with the simulated results of experimentsdiscussed above Figure 26 strongly illustrates that influenzais ceased when the intervention measures are combinedtogether The peak value of infected agents is 141 in the ldquo80thdays of close-workspace and isolation with 1 day delay tohospitalrdquo case while 25135 in ldquono interventionrdquo case 2218 inldquo1 day delay to hospitalrdquo case 203 in ldquo1 day delay to hospitaland isolationrdquo case The interventions bring a great reductionof infection the peak values decrease from ten thousandslevel to hundred level
On the other hand the peak value even decreases to 108when the execution time of close workspace is 10 days earlierThe date of maximum value of infected agents is advancedfrom the 135th day to the 108th day The whole circle of H1N1influenza is reducedThe results of this group of experimentsshow that intervention measures should be activated as soonas possible and the compositions of interventions are moreeffective in emergency management of H1N1 influenza
In summary based on the general process lots of exper-iments could be done by KD-ACP to study emergency man-agement in H1N1 influenza Many conclusions are obtainedbased on the analysis of the experiments in this sectionAccording to these conclusions themost optimal response inour case study is the case of ldquo70th days of close-workspace andisolation with 1 day delay to hospitalrdquo It is because this casehas the fewest infected agents relatively But the modelling ofH1N1 influenza is not sufficient Economic and social factorsof interventions such as the money cost by vaccination andisolation the influence of the close-workspace in society
24 Mathematical Problems in Engineering
Table 9 The initial settings of experiments of isolation and closeworkspace
Parameters ValueAgent count 19610000Initial infected count 40Interventions activated time (119879Intervention) 70th day 80th dayDelay to hospital (119879HDelay) 1 dayIsolation duration (119879IDuration) 3 daysClose workspace duration (119879CWDuration) 7 days
50 100 150 200 2500
50
100
150
200
Infe
cted
agen
ts
Loading intervention of isolation and close workspace
Delay to hospital = 1 isolation duration = 3Start close time = 70 close duration time = 7Delay to hospital = 1 isolation duration = 3Start close time = 80 close duration time = 7
Day d
Figure 26The results of loading interventions of isolation and closeworkspace
and even opinion transmission of H1N1 in internet are notconsidered in our case study Without the analysis of thesefactors it does not make sense to talk about the most optimalintervention measures However KD-ACP had provided asoftware framework and general process of the research onemergency management Supported by KD-ACP the expertsof emergency management are able to study emergent eventssuch as H1N1 influenza as detailed as possible Based on theexperiments and analysis again and again the most optimalintervention measures will be found ultimately
55 Parallel Execution by ASST and OsdRT As discussed inSection 35 ASST andOsdRT are used to support the parallelexecution of artificial Beijing The statistical data is collectedand stored in artificial society runtime database at runtimeASST downloads the infection data from the database anddisplays it in the map As shown in Figure 27 the infectedagents are distributed in the districts in BeijingThe snapshotgives the geospatial distribution of the H1N1 epidemic Thelegends of the agents in different status are listed in the left
right corner in Figure 27 Five statuses are shown in the figurehealth without vaccination health with vaccination incuba-tion being symptomatic and convalescence ASST providesa very direct feeling for the emergency experts Partly withthe situation from ASST Emergency Decision Organizationmakes the emergency response plans to the response of theemergency scenario With the help of plans the epidemicin actual society can be restrained In the meantime theinfluenza information is reflected in the Internet Accordingto Figure 8 the data acquisition data extraction and datastandardization of H1N1 influenza are processed in OsdRTOsdRT is implemented by Dr Caorsquos team from Institute ofAutomation from Chinese Academy of Sciences [58] Thetool provides not only the reports of H1N1 SARS hand-foot-mouth disease and intestinal diseases are all monitored Theweb page is shown in Figure 28 H1N1 reports from Internetare collected first Based on the data extraction of the reportsH1N1 case of distribution is obtained and shown in theincidencemap As a result H1N1 incidence data is regarded asthe knowledge which is sent to EECT and IMCTThe settingsof emergent events and intervention measures are modifiedin order to make artificial society approach actual society innext turn of computational experiments With the help ofOsdRT the loop of parallel execution process integrates ACPapproach
6 Conclusion
This work provides an integrated software framework forsocial Computing in emergencymanagement From a systemperspective KD-ACP provides a reliable flexible low costand effective platform for the scientists to do the research onemergency response problems The analysis and predictionof these problems with inherent complexity can be solvedby the repetition of possible alternative experiments onthe software framework Therefore KD-ACP which is anattempt to implement ACP approach seemed as an effectivecomputational framework to support the decision making inemergency response
Currently KD-ACP is used to study the H1N1 epidemicin Beijing in 2009 The most optimal compositions of inter-vention measures are testified to support the response tonext influenza However it is still a long way ahead beforethe automation systematization and practicalization of KD-ACP A lot of work will be carried out along several directionsas in the following
(i) The parallel execution process in KD-ACP is actuallynot strongly connected with the emergency responseorganization Traditional experience and theory stillplay a dominant role in the decision of emergencymanagement It is necessary to propose a mechanismof hall for workshop to facilitate the parallel executionloop in actual society
(ii) Section 33 introduces the principle and process ofmodelling artificial society Metamodels and modelsof artificial Beijing are built by GME Model transfor-mation and code generation are used to implementcodes from FSM based models But many details
Mathematical Problems in Engineering 25
Fangshan district Daxing district Fengtai district Xicheng district
Mentougou district Haidian districtShijingshan district Dongcheng district
Incubation
Symptomatic
Convalescence
Health withoutvaccination
Health withvaccinationLegend of agent
health status
Figure 27 The situation of artificial Beijing by ASST
H1N1
Districts of Beijing
H1N1 case ofdistribution
Incidence map
Distribution ofH1N1 reportsfrom internet
Incidence of H1N1
Fangshan districtLiangxiang town
Population 19000Incidence 135789
(1100000)
Figure 28 The implementation of OsdRT by Institute of Automation from Chinese Academy of Sciences
are still not illustrated due to the topic of thispaper The key techniques such as template of codeframework and model transformations from specificdomains to FSM will be focused and introduced indetail in our next paper
(iii) The performance of computational experiments inKD-ACP is still slower than expectation The opti-mization of simulation engine and models structureis necessary to improve the performance So it ispossible to run artificial society with the real worldin parallel
In summary KD-ACP paves a new way for scientists inboth emergencymanagement and computation simulation to
collaborate with each other in solving the emergency man-agement problems by social computing
Research Highlights
(i) KD-ACP is implemented based on ACP approach
(ii) Artificial Beijing is built with the help of KD-ACP
(iii) H1N1 is modeled to simulate the influenza in Beijingin 2009
(iv) Computational experiments testify the effectivenessof intervention measures
26 Mathematical Problems in Engineering
Conflict of Interests
The authors declare that there is no conflict of interestsregarding the publication of this paper
Acknowledgment
This work was supported in part by the National ScienceFoundation of China (nos 91024030 71303252 61403402and 91324013)
References
[1] W Mao and F-Y Wang Advances in Intelligence and SecurityInformatics Academic Press Oxford UK 2012
[2] E Bonabeau ldquoAgent-based modeling methods and techniquesfor simulating human systemsrdquo Proceedings of the NationalAcademy of Sciences of the United States of America vol 99 no3 pp 7280ndash7287 2002
[3] K M Carley D B Fridsma E Casman et al ldquoBioWar scalableagent-basedmodel of bioattacksrdquo IEEETransactions on SystemsMan and Cybernetics A Systems and Humans vol 36 no 2 pp252ndash265 2006
[4] J M Epstein ldquoModelling to contain pandemicsrdquo Nature vol460 no 7256 p 687 2009
[5] Los Alamos National Laboratory Reports about EPISIMS LosAlamos National Laboratory 2003
[6] C H Builder and S C Bankes ldquoArtificial societies a conceptfor basic research on the societal impacts of informationtechnologyrdquo RAND Report 149 RAND Corp 1991
[7] F-Y Wang ldquoComputational theory and method on complexsystemrdquo China Basic Science vol 6 no 5 pp 3ndash10 2004
[8] F-Y Wang ldquoToward a paradigm shift in social computing theACP approachrdquo IEEE Intelligent Systems vol 22 no 5 pp 65ndash67 2007
[9] F-Y Wang and S Tang ldquoArtificial societies for integratedand sustainable development of metropolitan systemsrdquo IEEEIntelligent Systems vol 19 no 4 pp 82ndash87 2004
[10] D Wen Y Yuan and X-R Li ldquoArtificial societies computa-tional experiments and parallel systems an investigation ona computational theory for complex socioeconomic systemsrdquoIEEE Transactions on Services Computing vol 6 no 2 pp 177ndash185 2013
[11] F Wang F Zeng and Y Yuan ldquoAn ACP-based approach forcomplexity analysis of E-commerce systemrdquo Complex Systemsand Complexity Science vol 5 no 3 pp 1ndash8 2008 (Chinese)
[12] F-Y Wang and P K Wong ldquoResearch commentary intelligentsystems and technology for integrative and predictivemedicinean ACP approachrdquoACMTransactions on Intelligent Systems andTechnology vol 4 no 2 pp 1ndash6 2013
[13] J Sifeng X Gang F Dong and H Chunpeng ldquoStudy on theemergency rescue decision support system of petrochemicalplant based on ACP theoryrdquo in Proceedings of the 6th Manage-ment Annual Meeting (MAM rsquo11) 2011 (Chinese)
[14] X Dong D Fan G Xiong F Zhu Z Zhang and Y Yao ldquoPar-allel Bus Rapid Transit (BRT) operation management systembased on ACP approachrdquo in Proceedings of the 6th ManagementAnnual Meeting (MAM rsquo11) Beijing China 2011
[15] B Ning F-Y Wang H-R Dong R-M Li D Wen and L LildquoParallel systems for Urban rail transportation based on ACP
approachrdquo Journal of Transportation Systems Engineering andInformation Technology vol 10 no 6 pp 23ndash28 2010 (Chinese)
[16] F-Y Wang J Zhao and S-X Lun ldquoArtificial power systemsfor the operation and management of complex power gridsrdquoSouthern Power System Technology vol 2 no 3 pp 1ndash11 2008(Chinese)
[17] M J North T R Howe N T Collier and J R Vos ldquoRepastsimphony development environmentrdquo in Proceedings of theAgent 2005 Conference on Generative Social Processes Modelsand Mechanisms 2005
[18] M J North T RHowe N T Collier and J R Vos ldquoRepast Sim-phony runtime systemrdquo in Proceedings of the Agent Conferenceon Generative Social Processes Models and Mechanisms 2005
[19] SWARM Development Group Swarm 211 Reference Guide2000 httpwwwswarmorg
[20] S Tisue and U Wilensky ldquoNetLogo a simple environment formodeling complexityrdquo in International Conference on ComplexSystems pp 16ndash21 2004
[21] S Luke C Cioffi-Revilla L Panait K Sullivan and G BalanldquoMASON a multiagent simulation environmentrdquo Simulationvol 81 no 7 pp 517ndash527 2005
[22] M Ichikawa H Tanuma Y Koyama andH Deguchi ldquoSOARSintroduction as a social microscope for simulations of socialinteractions and gamingrdquo in Organizing and Learning throughGaming and Simulation Proceedings of the 38th Conference ofthe International Simulation and Gaming Association (ISAGArsquo07) pp 149ndash158 2007
[23] D S Burke J M Epstein D A T Cummings et al ldquoIndividual-based computational modeling of smallpox epidemic controlstrategiesrdquo Academic Emergency Medicine vol 13 no 11 pp1142ndash1149 2006
[24] F-Y Wang ldquoParallel control and management for intelligenttransportation systems concepts architectures and applica-tionsrdquo IEEE Transactions on Intelligent Transportation Systemsvol 11 no 3 pp 630ndash638 2010
[25] Z Molnar D Balasubramanian and A Ledeczi ldquoAn introduc-tion to the genericmodeling environmentrdquo in Proceedings of theTOOLS Europe 2007 Workshop on Model-Driven DevelopmentTool Implementers Forum Zurich Switzerland 2007
[26] S Kelly and J-P Tolvanen Domain Framework in Domain-Specific Modeling Enabling Full Code Generation JohnWiley ampSons Hoboken NJ USA 2007
[27] B P Zeigler H Praehofer and T G Kim Theory of Modelingand Simulation Integrating Discrete Event and ContinuousComplex Dynamic Systems Academic Press San Diego CalifUSA 2000
[28] G Booch J Rumbaugh and I JacobsonThe Unified ModelingLanguage User Guide Addison-Wesley Reading Mass USA1999
[29] F Wagner R Schmuki T Wagne and P Wolstenholme Mod-eling Software with Finite State Machines A Practical ApproachCRC amp Taylor amp Francis Boca Raton Fla USA 2006
[30] T Espiner ldquoChina builds worldrsquos fastest supercomputerrdquo ZDNetUK October 2010
[31] B Chen and G Guo ldquoA two-tier parallel architecture for arti-ficial society simulationrdquo in Proceedings of the ACMIEEESCS26th Workshop on Principles of Advanced and DistributedSimulation (PADS rsquo12) pp 184ndash186 July 2012
[32] G Guo B Chen X G Qiu and Z Li ldquoParallel simulation oflarge-scale artificial society on CPUGPU mixed architecturerdquo
Mathematical Problems in Engineering 27
in Proceedings of the ACMIEEESCS 26th Workshop on Prin-ciples of Advanced and Distributed Simulation (PADS rsquo12) pp174ndash177 July 2012
[33] S-X Lun ldquoResearch on the classification of parallel executionmodes of ACP theoryrdquo Acta Automatica Sinica vol 38 no 10pp 1602ndash1608 2012
[34] F-YWang ldquoA framework for social signal processing and anal-ysis from social sensing networks to computational dialecticalanalyticsrdquo Science China vol 43 no 12 pp 1598ndash1611 2013
[35] G GuoOne Model Userrsquos Manual version 12 National Univer-sity of Defense University 2011
[36] Y Ge R Meng Z Cao X Qiu and K Huang ldquoVirtual citymdashanindividual-based digital environment for human mobility andinteractive behaviorrdquo Simulation Transactions of the Society forthe Modeling and Simulation International 2014
[37] M Kretzschmar and R T Mikolajczyk ldquoContact profiles ineight European countries and implications for modelling thespread of airborne infectious diseasesrdquo PLoS ONE vol 4 no6 Article ID e5931 2009
[38] W Duan and X Qiu ldquoFostering artificial societies using sociallearning and social control in parallel emergency managementsystemsrdquo Frontiers in Computer Science vol 6 no 5 pp 604ndash610 2012
[39] W J Edmunds C J OrsquoCallaghan and D J Nokes ldquoWhomixes with whom Amethod to determine the contact patternsof adults that may lead to the spread of airborne infectionsrdquoProceedings of the Royal Society B Biological Sciences vol 264no 1384 pp 949ndash957 1997
[40] J Banks J S Carson B L Neison and D M Nicol Discrete-Event System Simulation Prentice Hall Englewood Cliffs NJUSA 4th edition 2007
[41] H Yongxia ldquoInvestigation and analysis of college studentsrsquo workand restrdquo Value Engineering vol 31 no 3 2012 (Chinese)
[42] W Duan Z Cao Y Wang et al ldquoAn ACP approach to publichealth emergency management using a campus outbreak ofh1n1 influenza as a case studyrdquo IEEE Transactions on SystemsMan and Cybernetics Part A Systems and Humans vol 43 no5 pp 1028ndash1041 2013
[43] HWHethcote ldquoThemathematics of infectious diseasesrdquo SIAMReview vol 42 no 4 pp 599ndash653 2000
[44] G Chowell E Shim F Brauer and et al ldquoModelling thetransmission dynamics of acute haemorrhagic conjunctivitisapplication to the 2003 outbreak in Mexicordquo Statistics inMedicine vol 25 no 11 pp 1840ndash1857 2006
[45] Y Ge L Liu B Chen X Qiu and K Huang ldquoAgent-basedmodeling for InfluenzaH1N1 in an artificial classroomrdquo SystemsEngineering Procedia vol 2 pp 94ndash104 2011
[46] A R Tuite A L Greer MWhelan et al ldquoEstimated epidemio-logic parameters andmorbidity associatedwith pandemicH1N1influenzardquo CMAJ vol 182 no 2 pp 131ndash136 2010
[47] L Brouwers B Cakici M Camitz A Tegnell and MBoman ldquoEconomic consequences to society of pandemic H1N1influenza 2009mdashpreliminary results for Swedenrdquo Euro Surveil-lance vol 14 no 37 pp 1ndash7 2009
[48] Y Ge L Liu X Qiu H Song Y Wang and K Huang ldquoAframework of multilayer social networks for communicationbehavior with agent-based modelingrdquo Simulation Transactionsof the Society for the Modeling and Simulation International vol89 no 7 pp 810ndash828 2013
[49] W Duan X Qiu Z Cao X Zheng and K Cui ldquoHeterogeneousand stochastic agent-based models for analyzing infectious
diseasesrsquo super spreadersrdquo IEEE Intelligent Systems vol 28 no4 pp 18ndash25 2013
[50] The Preparation and Emergency Response Plans to PandemicPublished by Ministry of Health Ministry of Health 2005(Chinese)
[51] R M Fujimoto Parallel and Distributed Simulation SystemsWiley Interscience 2000
[52] F Cicirelli A Giordano and L Nigro ldquoDistributed simulationof situated multi-agent systemsrdquo in Proceedings of the 15thIEEEACM International Symposium on Distributed Simulationand Real Time Applications (DS-RT rsquo11) pp 28ndash35 September2011
[53] B Chen and G Guo ldquoA two-tier parallel architecturefor artificial society simulationrdquo in Proceedings of the 26thACMIEEESCS Workshop on Principles of Advanced and Dis-tributed Simulation (PADS rsquo12) pp 184ndash186 July 2012
[54] G Guo B Chen and X Qiu ldquoParallel simulation of large-scale artificial society with GPU as coprocessorrdquo InternationalJournal of Modeling Simulation and Scientific Computing vol4 no 2 Article ID 1350005 2013
[55] B Chen L-B Zhang X-C Liu and H Vangheluwe ldquoActivity-based simulation using DEVS increasing performance by anactivity model in parallel DEVS simulationrdquo Journal of ZhejiangUniversity Science C vol 15 no 1 pp 13ndash30 2014
[56] X-L Wang P Yang Z-D Cao et al ldquoQuantitative evaluationon the effectiveness of prevention and control measures againstpandemic influenza A (H1N1) in Beijing 2009rdquoChinese Journalof Epidemiology vol 31 no 12 pp 1374ndash1378 2010 (Chinese)
[57] S Mei A D van de Vijver L Xuan Y Zhu and P M ASloot ldquoQuantitatively evaluating interventions in the influenzaA (H1N1) epidemic on China campus grounded on individual-based simulationsrdquo Procedia Computer Science vol 1 no 1 pp1675ndash1682 2010
[58] X Qiu Z Cao and Z Li Mid-Term Reports of China NationalScience Foundation 91024030 2014
Submit your manuscripts athttpwwwhindawicom
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
MathematicsJournal of
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Mathematical Problems in Engineering
Hindawi Publishing Corporationhttpwwwhindawicom
Differential EquationsInternational Journal of
Volume 2014
Applied MathematicsJournal of
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Probability and StatisticsHindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Journal of
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Mathematical PhysicsAdvances in
Complex AnalysisJournal of
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
OptimizationJournal of
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
CombinatoricsHindawi Publishing Corporationhttpwwwhindawicom Volume 2014
International Journal of
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Operations ResearchAdvances in
Journal of
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Function Spaces
Abstract and Applied AnalysisHindawi Publishing Corporationhttpwwwhindawicom Volume 2014
International Journal of Mathematics and Mathematical Sciences
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
The Scientific World JournalHindawi Publishing Corporation httpwwwhindawicom Volume 2014
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Algebra
Discrete Dynamics in Nature and Society
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Decision SciencesAdvances in
Discrete MathematicsJournal of
Hindawi Publishing Corporationhttpwwwhindawicom
Volume 2014 Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Stochastic AnalysisInternational Journal of
Mathematical Problems in Engineering 17
Censusdata
Generate GUID
Assign individualhousehold
Generate familystructure
Assign family roleto every member
Endow individualwith age gender
social role
Distribute environmentlocations on the map
School Workplace Entertainmentlocations
Consumptionlocations
Assign school orworkplace to
individual
Correlatedindividual withother locations
Generate friendshipin school
Generate coworkerrelations inworkplace
Assign individualadministrative
region
Generate housebuildings
Locate householdto house building
Generate familyrelation
Generateneighborhood
relation
PopulationEnvironment
Environment listSocial network
Figure 16 The algorithm of generating geospatial and social environment data [36]
Edit
Edit
Edit
Edit
Edit
Edit
BrowseStatistical characteristic of population
Geospatial population data
Population data of street
Population types
Geospatial and social environment generation tool
Workspaces
Parameters distribution settings
Household types
Roles in household
Household structure
Social roles
BrowseBrowseBrowseBrowseBrowseBrowse
Population distribution data
File path
Retired distributionPop distributionHousehold numHousehold distributionSex distributionAge distributionData type
Figure 17 The GUI of Population and Geospatial Environment generation Tool (PGET)
18 Mathematical Problems in Engineering
Infected
SusceptibleHealth
withimmunity
Infected
by contac
t
with in
fected
agents
Recovery
Increase before beingsymptomatic symptomatic
Decrease after being
Decrease by time
Infectivity
Immunity
Figure 18 The transitions of agent health status
119879Lat denotes the duration of latent period 120592 120572 and 120573 arethe location parameter scale parameter and shape parameterof Weibull distribution respectively 120574 is a uniform randomnumber in the range [0 1] According to statistics [47] 120592 120572and 120573 are set as 0 18 and 121 respectively Then the mean(standard deviation) of latent period is calculated as 159 days(058 day2) Furthermore infectious period is set as 7 days
A susceptible studentrsquos probability of getting infecteddepended on the infectivity of infectious agents his ownimmunity level the duration of the contact action and soon The infectivity of an infected agent evolved with timeduring the infectious period According to the statistics ofinfectivity in the chart with red bars in Figure 19 an infectedstudent had the highest level of infectivity in the second dayafter he had the first symptom The infectivity levels in otherdays in comparison with the second day are in Table 5 Theday labeled as ldquominus1rdquo means the day before the starting of thesymptomatic period
Based on the algorithm discussed before theH1N1modelis developed with the help of MDT The settings of influenzamodel will be discussed in Section 521
512 Modelling Intervention Measures by MDT In corre-spondence intervention measures model is used to ceasethe outbreak of H1N1 As discussed in [42 48 49] if theappropriate measures are taken when the infectious diseaseemerges the transmission of the disease could be sloweddown and the damage could be decreased Therefore it isnecessary and important to design emergency interventionmeasures in the artificial Beijing According to the MinistryofHealth inChina the intervention policies [50] are designedto control the spread of influenza such as H1N1 The inter-ventionmeasures of these policies including are interventionsactivated time vaccination rate antibiotic delay to hospitalisolation duration close workspace duration and limitationof activities listed in Table 6 Similarly the models of theseintervention measures are also developed with the help ofMDT Using the reasonable data ranges listed in Table 6intervention models can be initialized to do the computa-tional experiments These experiments are used to show how
Symptomatic1 2 3 4 5 6 7minus1
Recovered
Tim
e (da
y)
Infe
ctiv
ity
Figure 19 The distribution of infectivity rate of an infected agent[37]
Table 5 The proportion of infectivity
Days (119894) minus1 1 2 3 4 5 6 6Proportion of infectivity 03 05 1 08 04 02 01 005
to restrain the H1N1 transmissionThe settings of these inter-vention measures in IMCT are introduced in Section 522
52 Experiment Settings
521 H1N1 Settings by EECT EECT is designed to initializethe emergent events models such as H1N1The settings comefrom the specified emergent event models As shown inFigure 20 in theH1N1 used case the settings are composed ofthe distribution of infection source infection rate infectionperiod and so on The infection rate changes with envi-ronments and infection periods The settings are initializedfrom both the research of medicine on H1N1 and the data ofH1N1 outbreak in Beijing in 2009The process of influenza inartificial Beijing is simulated to repeat the possible scenariosof H1N1 break During the computational experiments ofartificial Beijing theH1N1 settings such as the period of beingsymptomatic are updated by the dynamic data registrationsfrom OsdRT The updating corrects the H1N1 model duringthe experiments With the help of OsdRT artificial Beijing isable to approach actual society It is also an implementationof parallel execution
522 Intervention Measures Settings by IMCT In the H1N1case the settings of intervention measures are shown inFigure 21 in IMCT These measures are designed under theinstruction of Emergency Decision Organizations like ChinaDisease Center (CDC) The measures such as vaccinationrate the rate of closedworkspace and the starting time can beinitialized in the tool Emergency response plans are imple-mented in these settings As illustrated in Figure 7 all the pos-sible compositions within value settings in reasonable inter-vals are tested to find the most optimal emergency responseplan For example different compositions of measures suchas vaccination rates and rates of workspace closed are usedto design the plans of experiments By the analysis of theresults the best composition of interventionmeasures can beobtainedThe analysis will be discussed in case study in detail
Mathematical Problems in Engineering 19
Table 6 The intervention measures of influenza
Intervention measures ValuesInterventions activated Time 119879Intervention isin [0 80th day]Vaccination rate 119877Vaccination isin [0 1]Antibiotic rate 119877Antibiotic isin [0 1]Delay to hospital 119867ospital(119860 119894) 119879infected lt 119905 lt 119879recovered + 119879HDelay 119879HDelay isin (0 3th day]Isolation duration 119868solation(119860 119894) 119879Intervention lt 119905 lt 119879Intervention + 119879IDuration 119879IDuration isin (0 14th day]Close workspace duration 119862lose(119864119894) 119879Intervention lt 119905 lt 119879Intervention + 119879CWDuration 119864119894 isinWorkspace 119879CWDuration isin (0 7th day]Limitation of agent activity duration 119871 imitation(119860 119894) 119879Intervention lt 119905 lt 119879Intervention + 119879LAADuration 119879LAADuration isin (0 7th day]119877Vaccination means the vaccination rate of artificial Beijing while 119877Antibiotic means the antibiotic rate used for infected agents in hospital119867ospital (119860119894) means that the infected agent 119860119894 is sent to hospital he or she will receive the retreatment immediately119868solation (119860119894) means that the infected agent 119860119894 is isolated at home he or she can only contact with the other agents until isolation duration is over119862lose (119864119894) means that the environment 119864119894 typed of workspace is closed until the close workspace duration is over119871 imitation (119860119894) means that the activities of agent 119860119894 are limited only the workspace and home are allowed until limitation duration is over
Incubation Weibulldistribution
Symbolic LogNormaldistribution
Recovering LogNormaldistribution
H1N1influenza
Settings of H1N1 influenza model
Disease model
Transmissionmodel
Figure 20 The settings of H1N1 in EECT
It is worth notifying that the settings of interventionmeasures are also influenced by OsdRT For example theH1N1 outbreak areas are notified by OsdRTThe interventionmeasures such as close workspace are used only in theoutbreak areas Therefore the unnecessary computation ofintervention measures is avoided So the data injection byOsdRT not only makes artificial society approach actualsociety but also increases the performance of computationalexperiments
53 Experiments by EMT When the initialization of arti-ficial Beijing is ready EMT is used to do the computa-tional experiments EMT is composed of two parts thecentral controller in the central node and the residen-tial service deployed in the computational nodes Centralcontroller loads the experiment plans and controls the whole
experiments process the GUI of controller is shown inFigure 22 EMT manages the computational experimentsin four steps cluster configuration load experiment plansmodels deployment and running experiments
531 Cluster Configuration Before the configuration clusterinformation is collected from the residential services insidethe cluster The information includes nodes amount nodenames node IPs and CPU occupations It is used to quantifythe computing power of the clusterThe cluster configurationis based on this information The nodes are selected forexperiments Then the LP (local process) [51] number foreach node is set according to the number of the CPU kernelsAfter the configuration the experiment plans are loaded tocustomize the models and data
20 Mathematical Problems in Engineering
Streets of close workspaces
Limitations ofactivities
IsolationsRespirator
Closeworkspace
Antibiotic
VaccinationHospital
Surveillance
Figure 21 The settings of intervention measures in IMCT
Information ofexperiment plan
List of experiment plans
Start | Pause | Reset | Stop | Speedup | Slowdown | Back
NodeName | CPU occupation | Memory occupation |Network | Models | Status
Figure 22 The deployment and running of experiments in EMT
532 Load Experiment Plan As discussed in Section 35 thedescription of experiment plans is composed of the infor-mation in three aspects models and the initial data for theexperiments the deployment mapping tables from the mod-els to the nodes and the plan of experiment executionWhenthe experiment plan is loaded by the EMT the file namesand model descriptions of models and data are listed Withthe support of cluster configuration models can be mappedto the nodes in the GUI according to the computationrequirements EMT also provides the setting of experimentsitself including the running times start time and end time
533 Models and Data Deployment Models are stored inthe agent population and services repository According tothe settings in experiment plan the customized models aredownloaded from the repository first then the models suchas agent model environment model emergent events modeland intervention measures model are integrated to be thecompositionmodels of artificial BeijingThemodels and datalocations in the nodes are configured in the ldquomodel pathrdquoand ldquodata pathrdquo in the GUI of central controller Finallythe models are uploaded to the nodes in cluster with theconsideration of partition The partition of artificial society
Mathematical Problems in Engineering 21
models is different from the multiagent system in [52]Models in artificial society are divided into three levels agentsand environments grids and city Agents and environmentsare encapsulated in grids which compose the whole cityThus the partition of artificial society becomes the partitionof grids The parallelism degrees are found to support thepartition of models agent independence and grid indepen-dence According to the parallelisms agents are partitioned toLPs according to the population distribution in grids Agentmovements among grids are simulated by the movementevents among grids in different LPs As a result a two-tierparallel architecture is proposed to support the simulation ofartificial society The architecture is detailed in [53]
534 Running of Experiments When the preparation isready the experiments can be done by the central controllerof EMT The running and control commands of the experi-ments are grouped into three aspects start pause or stop theexperiments get the running information of the nodes suchas the snapshot of the desktop of node With the help of thesecommands users can run the computational experimentsto do the research on the emergency management It isworth notifying that the simulation engine is optimized inorder to support the ten million agent simulation UsingGPU as coprocessor a two-tier parallel simulation engine isdesigned with support of MPI and OpenCL through phasedsynchronization [54] One-sided communication is used forreflection of remote simulation objects and message passingbetween processes A general kernel function prototype iselaborately designed and conditionally compiled for execu-tion on both CPU and GPU Moreover the optimizationoperations like load balance is developed under the instruc-tion of activity-based simulation [55] The densely populatedgrids are given more CPU and memory resource accordingto the activity predictions
54 Results of H1N1 Experiments with KD-ACP On thebasis of artificial society models and the intervention mea-sures mentioned in Section 512 a series of computationalexperiments are performed to study the H1N1 influenza inBeijing In our case of artificial Beijing 19610000 agentsare simulated in the cluster within 48 CPU cores and 128Gmemories the nodes are connected by kilomega networks Ittakes 18 hours to simulate a 250-day disease spreading
In order to illustrate how the optimal plans are obtainedbyKD-ACP the experiments are divided into four groups theexperiments ofmodel validation the experiments of sensitiv-ity analysis of vaccination rate the experiments of sensitivityanalysis of isolation and the experiments of the combinationof interventionmeasuresThese four groups show the generalprocess of the research on emergency management in H1N1influenza Firstly the models of H1N1 are validated withthe support of historical data Secondly the traditionalmedical interventions of pandemic such as vaccination areanalyzed to find the optimal plan Thirdly the traditionalnonmedical interventions of pandemic such as isolation areanalyzed to find the optimal plan Finally the combination ofinterventions is tested to show the combined effect
It is worth mentioning that all the experiments in ourwork are performed 100 times These experiments are ini-tialized with different settings of interventions in IMCT Theresults shown in the figures are the mean values
541 The Experiments of Model Validation In our artifi-cial Beijing the models of H1N1 influenza are built basedon the existing researches Wang and her colleagues hadstudied the H1N1 influenza by SEIR (SusceptibleExposedInfectiveRecovered) models [56] With the support of theirwork we build the models of artificial Beijing in the mannerof multiagents
In Figure 23 the control group shown in red line isdrawn based on the realistic statistical data collected in H1N1influenza in Beijing in 2009 The influenza lasts more thansix months and more than 170000 people are infectedComparing with the control group the simulated resultswithout interventions are drawn in blue It is obvious that thesimulated case fits the control group well before 70 days Butthe ascending of control group is slowed after 70 days andthe peak value (17883) is much less than the simulated results(25135) It is because the vaccination is activated at day 67 incontrol group according to the intervention measures usedin 2009 The restrain by vaccination becomes effective eightdays later The change of infected number after vaccinationis shown by red spotted line The trend of control group isdifferent from the simulation data after 75 days
However the simulated results (blue line) in compu-tational experiments fit the control group (as shown inFigure 23) in the first 70 days when no intervention mea-sures are executed The key features such as effective basicreproduction number (119877
0) are also almost the same in both
cases Therefore it can be concluded that the models ofartificial society built in microview are validated by therealistic statistical data in influenza in Beijing in 2009
In this group of experiments no interventions are exe-cuted in order to validate the models Therefore the simu-lated results of this extreme case which cannot happen todaycould be obtained As shown by blue line the number ofinfected agents without interventions reaches peak value atday 91 with 25135 consisting of 8052 in symptomatic phaseand 28950 in recovery phase It is obvious that the totalnumber of infected agents grows slowly at the beginningtime Then the number increases quickly when the influenzabreaks and it reaches its maximum value with 383870 in theend This phenomenon can be explained as in the followingAt the beginning of influenza due to the limited numberof infected agents the propagation of the disease remainsslow but many agents become susceptible When the numberof infected agents increases rapidly the spreading becomessignificant However the number of infected agents decreasesafter day 91 It is because the disease spreading relies on thesocial networks and spatial contact networks [49] It meansthat the infected agents can only infect agents either in thesocial relationship networks or in the spatially contacted
542 The Experiments of Sensitivity Analysis of VaccinationRate According to Section 512 vaccination is a typical
22 Mathematical Problems in Engineering
0
Infe
cted
agen
ts
0 50 100 150 200 250
Statistical dataSimulation data
25000
20000
15000
10000
5000
Day d
Figure 23 The results with no intervention executed and the com-parison with realistic statistical data in Beijing in 2009
medical intervention in influenza Traditionally the effective-ness of vaccination can only be evaluated after the influenzaand the key settings such as vaccination rate are determinedby the experience It is always difficult for emergency man-agement organizations to find the most optimal vaccinationrate in advanceTherefore the sensitivity of vaccination rate isanalyzed in this group of experiments The settings are listedin Table 7 average results are shown in Figure 24
Vaccination rates are set linearly (10 30 50 and70) to test the changes of influenza The peak value ofinfected agents decreases from 25135 22392 15787 6907to 638 respectively It shows that the more the agents arevaccinated the smaller the peak value of the infected agentswill be In the case of 70 vaccination the total infected agentis only 13799 It means that the influenza in this case does nothappen due to the high vaccination rate
The change of peak values also shows an interestingphenomenon Vaccination rates increase linearly the peakvalues decrease slower than the ascending of vaccination ratesat beginning but the descending becomes more and morefaster with the linear increase of vaccination rate It meansthat the small scale of vaccination (less than 30) does notmake senseWith the increase of vaccination rate the numberof susceptible agents decreases significantly In sequencethe possibility of infections is lowered greatly Additionallyit could be found that the decrease of peak values comeslater than the increase of the vaccination rates This is quitereasonable according to Mei et alrsquos work in [57]
As a result it is concluded that high vaccination rate iseffective in the control of H1N1 influenza though costly
543 The Experiments of Sensitivity Analysis of IsolationAccording to Section 512 isolation is a typical nonmedicalintervention It is the most used intervention measure in
Table 7 The initial settings of experiments of vaccination
Parameters ValueAgent count 19610000Initial infected count 40Interventions activated time(119879Intervention)
0
Vaccination rate (119877Vaccination) 0 10 30 50 70
0
Infe
cted
agen
ts
Loading intervention of vaccination
No vaccination10 vaccination30 vaccination
50 vaccination70 vaccination
0 50 100 150 200 250
25000
20000
15000
10000
5000
Day d
Figure 24 The results of loading intervention of vaccination
the emergency management in influenza As mentionedbefore a problem also exists that the measure cannot bevalidated before the intervention Therefore the sensitivityof isolation is analyzed in this group of experiments It isworth mentioning that the isolation is usually combinedwith the hospital policy It means the infected individual issent to hospital meanwhile the people who contact withthe infected are all isolated As a result parameters delay tohospital and isolation duration are designed together in thesettings of experiment Delay to hospital stands for the periodfrom infection time to hospital time while Isolation durationrepresents the period when an agent cannot contact others ifhe has contact with an infected agent The settings of theseparameters are listed in Table 8 average results are shown inFigure 25
Firstly the parameter delay to hospital is simulated aloneto test the sensitivity The simulated results are shown insolid line in Figure 25 The peak value decreases from 6708to 2218 when the delay to hospital decreases from 15 daysto 1 day The half day change brings the 67 reductionMoreover the peak value almost approximates to zero inthe case of 05 days The influenza does not happen in thiscase It can be concluded that the earlier the infected agents
Mathematical Problems in Engineering 23
Table 8 The initial settings of experiments of isolation
Parameters ValueAgent count 19610000Initial infected count 40Interventions activated time (119879Intervention) 80Delay to hospital (119879HDelay) isolation duration (119879IDuration) (15 0) (1 0) (05 0) (15 3) (1 3) (05 3)
0 50 100 150 200 2500
1000
2000
3000
4000
5000
6000
7000
Infe
cted
agen
ts
Loading intervention of isolation
Delay to hospital = 15
Delay to hospital = 10
Delay to hospital = 05
Delay to hospital = 15 isolation duration = 3
Delay to hospital = 10 isolation duration = 3
Delay to hospital = 05 isolation duration = 3
Day d
Figure 25 The results of loading intervention of isolation
are sent to hospital the less the agents will be infectedSo the construction of effective mechanism to find infectedindividuals is important in the emergency management ofinfluenza
Secondly the parameter isolation duration is added tosimulate the isolation of interventionObviously the isolationmeasure decreases the infected agents greatly Compared tothe parameter delay to hospital of 15-day case the peak valuedecreases from 6708 to 362when the isolation is added in theinterventionThe reduction of 6346 agents comes from the 3-day isolation of agents who had contacted with the infectedagents However the decrease of parameter delay to hospitalcannot bring the similar reduction in the 3-day isolationcases According to the figure the peak value decreases from362 to 51 from 15-day case to 05-day case only 311 agents aresaved from infection As a result the sensitivity of simulatedresults does not change obviously with delay to hospital inthe case of isolation It can be concluded that isolation is amore effective measure Though it is not easy for emergencymanagement organizations to decrease the delay to hospitaldue to the current monitoring mechanism the influenza canstill be controlled by isolation
544 The Experiments of Combination of Intervention Mea-sures Isolation is not the only nonmedical intervention ininfluenza close workspace is another common measureWhen close workspace is activated the workspaces are allclosed agents can only stay home at the closed time Luckilyisolation and close workspace are independent from oneanother It is possible for emergency management organi-zations to use these two measures together So this groupof experiments is designed to obtain the effectiveness of thecombination of intervention measures The parameters areinitialized with different settings listed in Table 9
Parameters delay to hospital and isolation duration is setin the middle values (1 day 3 days) according to the lastsection According to the experience close duration is set asseven days It is known that close workspace is a really costintervention measure the activated time is a key parameterWith the fixed parameters mentioned before the changeof interventions activated time is analyzed in this group ofcomputational experimentsThe average results are shown inFigure 26
Compared with the simulated results of experimentsdiscussed above Figure 26 strongly illustrates that influenzais ceased when the intervention measures are combinedtogether The peak value of infected agents is 141 in the ldquo80thdays of close-workspace and isolation with 1 day delay tohospitalrdquo case while 25135 in ldquono interventionrdquo case 2218 inldquo1 day delay to hospitalrdquo case 203 in ldquo1 day delay to hospitaland isolationrdquo case The interventions bring a great reductionof infection the peak values decrease from ten thousandslevel to hundred level
On the other hand the peak value even decreases to 108when the execution time of close workspace is 10 days earlierThe date of maximum value of infected agents is advancedfrom the 135th day to the 108th day The whole circle of H1N1influenza is reducedThe results of this group of experimentsshow that intervention measures should be activated as soonas possible and the compositions of interventions are moreeffective in emergency management of H1N1 influenza
In summary based on the general process lots of exper-iments could be done by KD-ACP to study emergency man-agement in H1N1 influenza Many conclusions are obtainedbased on the analysis of the experiments in this sectionAccording to these conclusions themost optimal response inour case study is the case of ldquo70th days of close-workspace andisolation with 1 day delay to hospitalrdquo It is because this casehas the fewest infected agents relatively But the modelling ofH1N1 influenza is not sufficient Economic and social factorsof interventions such as the money cost by vaccination andisolation the influence of the close-workspace in society
24 Mathematical Problems in Engineering
Table 9 The initial settings of experiments of isolation and closeworkspace
Parameters ValueAgent count 19610000Initial infected count 40Interventions activated time (119879Intervention) 70th day 80th dayDelay to hospital (119879HDelay) 1 dayIsolation duration (119879IDuration) 3 daysClose workspace duration (119879CWDuration) 7 days
50 100 150 200 2500
50
100
150
200
Infe
cted
agen
ts
Loading intervention of isolation and close workspace
Delay to hospital = 1 isolation duration = 3Start close time = 70 close duration time = 7Delay to hospital = 1 isolation duration = 3Start close time = 80 close duration time = 7
Day d
Figure 26The results of loading interventions of isolation and closeworkspace
and even opinion transmission of H1N1 in internet are notconsidered in our case study Without the analysis of thesefactors it does not make sense to talk about the most optimalintervention measures However KD-ACP had provided asoftware framework and general process of the research onemergency management Supported by KD-ACP the expertsof emergency management are able to study emergent eventssuch as H1N1 influenza as detailed as possible Based on theexperiments and analysis again and again the most optimalintervention measures will be found ultimately
55 Parallel Execution by ASST and OsdRT As discussed inSection 35 ASST andOsdRT are used to support the parallelexecution of artificial Beijing The statistical data is collectedand stored in artificial society runtime database at runtimeASST downloads the infection data from the database anddisplays it in the map As shown in Figure 27 the infectedagents are distributed in the districts in BeijingThe snapshotgives the geospatial distribution of the H1N1 epidemic Thelegends of the agents in different status are listed in the left
right corner in Figure 27 Five statuses are shown in the figurehealth without vaccination health with vaccination incuba-tion being symptomatic and convalescence ASST providesa very direct feeling for the emergency experts Partly withthe situation from ASST Emergency Decision Organizationmakes the emergency response plans to the response of theemergency scenario With the help of plans the epidemicin actual society can be restrained In the meantime theinfluenza information is reflected in the Internet Accordingto Figure 8 the data acquisition data extraction and datastandardization of H1N1 influenza are processed in OsdRTOsdRT is implemented by Dr Caorsquos team from Institute ofAutomation from Chinese Academy of Sciences [58] Thetool provides not only the reports of H1N1 SARS hand-foot-mouth disease and intestinal diseases are all monitored Theweb page is shown in Figure 28 H1N1 reports from Internetare collected first Based on the data extraction of the reportsH1N1 case of distribution is obtained and shown in theincidencemap As a result H1N1 incidence data is regarded asthe knowledge which is sent to EECT and IMCTThe settingsof emergent events and intervention measures are modifiedin order to make artificial society approach actual society innext turn of computational experiments With the help ofOsdRT the loop of parallel execution process integrates ACPapproach
6 Conclusion
This work provides an integrated software framework forsocial Computing in emergencymanagement From a systemperspective KD-ACP provides a reliable flexible low costand effective platform for the scientists to do the research onemergency response problems The analysis and predictionof these problems with inherent complexity can be solvedby the repetition of possible alternative experiments onthe software framework Therefore KD-ACP which is anattempt to implement ACP approach seemed as an effectivecomputational framework to support the decision making inemergency response
Currently KD-ACP is used to study the H1N1 epidemicin Beijing in 2009 The most optimal compositions of inter-vention measures are testified to support the response tonext influenza However it is still a long way ahead beforethe automation systematization and practicalization of KD-ACP A lot of work will be carried out along several directionsas in the following
(i) The parallel execution process in KD-ACP is actuallynot strongly connected with the emergency responseorganization Traditional experience and theory stillplay a dominant role in the decision of emergencymanagement It is necessary to propose a mechanismof hall for workshop to facilitate the parallel executionloop in actual society
(ii) Section 33 introduces the principle and process ofmodelling artificial society Metamodels and modelsof artificial Beijing are built by GME Model transfor-mation and code generation are used to implementcodes from FSM based models But many details
Mathematical Problems in Engineering 25
Fangshan district Daxing district Fengtai district Xicheng district
Mentougou district Haidian districtShijingshan district Dongcheng district
Incubation
Symptomatic
Convalescence
Health withoutvaccination
Health withvaccinationLegend of agent
health status
Figure 27 The situation of artificial Beijing by ASST
H1N1
Districts of Beijing
H1N1 case ofdistribution
Incidence map
Distribution ofH1N1 reportsfrom internet
Incidence of H1N1
Fangshan districtLiangxiang town
Population 19000Incidence 135789
(1100000)
Figure 28 The implementation of OsdRT by Institute of Automation from Chinese Academy of Sciences
are still not illustrated due to the topic of thispaper The key techniques such as template of codeframework and model transformations from specificdomains to FSM will be focused and introduced indetail in our next paper
(iii) The performance of computational experiments inKD-ACP is still slower than expectation The opti-mization of simulation engine and models structureis necessary to improve the performance So it ispossible to run artificial society with the real worldin parallel
In summary KD-ACP paves a new way for scientists inboth emergencymanagement and computation simulation to
collaborate with each other in solving the emergency man-agement problems by social computing
Research Highlights
(i) KD-ACP is implemented based on ACP approach
(ii) Artificial Beijing is built with the help of KD-ACP
(iii) H1N1 is modeled to simulate the influenza in Beijingin 2009
(iv) Computational experiments testify the effectivenessof intervention measures
26 Mathematical Problems in Engineering
Conflict of Interests
The authors declare that there is no conflict of interestsregarding the publication of this paper
Acknowledgment
This work was supported in part by the National ScienceFoundation of China (nos 91024030 71303252 61403402and 91324013)
References
[1] W Mao and F-Y Wang Advances in Intelligence and SecurityInformatics Academic Press Oxford UK 2012
[2] E Bonabeau ldquoAgent-based modeling methods and techniquesfor simulating human systemsrdquo Proceedings of the NationalAcademy of Sciences of the United States of America vol 99 no3 pp 7280ndash7287 2002
[3] K M Carley D B Fridsma E Casman et al ldquoBioWar scalableagent-basedmodel of bioattacksrdquo IEEETransactions on SystemsMan and Cybernetics A Systems and Humans vol 36 no 2 pp252ndash265 2006
[4] J M Epstein ldquoModelling to contain pandemicsrdquo Nature vol460 no 7256 p 687 2009
[5] Los Alamos National Laboratory Reports about EPISIMS LosAlamos National Laboratory 2003
[6] C H Builder and S C Bankes ldquoArtificial societies a conceptfor basic research on the societal impacts of informationtechnologyrdquo RAND Report 149 RAND Corp 1991
[7] F-Y Wang ldquoComputational theory and method on complexsystemrdquo China Basic Science vol 6 no 5 pp 3ndash10 2004
[8] F-Y Wang ldquoToward a paradigm shift in social computing theACP approachrdquo IEEE Intelligent Systems vol 22 no 5 pp 65ndash67 2007
[9] F-Y Wang and S Tang ldquoArtificial societies for integratedand sustainable development of metropolitan systemsrdquo IEEEIntelligent Systems vol 19 no 4 pp 82ndash87 2004
[10] D Wen Y Yuan and X-R Li ldquoArtificial societies computa-tional experiments and parallel systems an investigation ona computational theory for complex socioeconomic systemsrdquoIEEE Transactions on Services Computing vol 6 no 2 pp 177ndash185 2013
[11] F Wang F Zeng and Y Yuan ldquoAn ACP-based approach forcomplexity analysis of E-commerce systemrdquo Complex Systemsand Complexity Science vol 5 no 3 pp 1ndash8 2008 (Chinese)
[12] F-Y Wang and P K Wong ldquoResearch commentary intelligentsystems and technology for integrative and predictivemedicinean ACP approachrdquoACMTransactions on Intelligent Systems andTechnology vol 4 no 2 pp 1ndash6 2013
[13] J Sifeng X Gang F Dong and H Chunpeng ldquoStudy on theemergency rescue decision support system of petrochemicalplant based on ACP theoryrdquo in Proceedings of the 6th Manage-ment Annual Meeting (MAM rsquo11) 2011 (Chinese)
[14] X Dong D Fan G Xiong F Zhu Z Zhang and Y Yao ldquoPar-allel Bus Rapid Transit (BRT) operation management systembased on ACP approachrdquo in Proceedings of the 6th ManagementAnnual Meeting (MAM rsquo11) Beijing China 2011
[15] B Ning F-Y Wang H-R Dong R-M Li D Wen and L LildquoParallel systems for Urban rail transportation based on ACP
approachrdquo Journal of Transportation Systems Engineering andInformation Technology vol 10 no 6 pp 23ndash28 2010 (Chinese)
[16] F-Y Wang J Zhao and S-X Lun ldquoArtificial power systemsfor the operation and management of complex power gridsrdquoSouthern Power System Technology vol 2 no 3 pp 1ndash11 2008(Chinese)
[17] M J North T R Howe N T Collier and J R Vos ldquoRepastsimphony development environmentrdquo in Proceedings of theAgent 2005 Conference on Generative Social Processes Modelsand Mechanisms 2005
[18] M J North T RHowe N T Collier and J R Vos ldquoRepast Sim-phony runtime systemrdquo in Proceedings of the Agent Conferenceon Generative Social Processes Models and Mechanisms 2005
[19] SWARM Development Group Swarm 211 Reference Guide2000 httpwwwswarmorg
[20] S Tisue and U Wilensky ldquoNetLogo a simple environment formodeling complexityrdquo in International Conference on ComplexSystems pp 16ndash21 2004
[21] S Luke C Cioffi-Revilla L Panait K Sullivan and G BalanldquoMASON a multiagent simulation environmentrdquo Simulationvol 81 no 7 pp 517ndash527 2005
[22] M Ichikawa H Tanuma Y Koyama andH Deguchi ldquoSOARSintroduction as a social microscope for simulations of socialinteractions and gamingrdquo in Organizing and Learning throughGaming and Simulation Proceedings of the 38th Conference ofthe International Simulation and Gaming Association (ISAGArsquo07) pp 149ndash158 2007
[23] D S Burke J M Epstein D A T Cummings et al ldquoIndividual-based computational modeling of smallpox epidemic controlstrategiesrdquo Academic Emergency Medicine vol 13 no 11 pp1142ndash1149 2006
[24] F-Y Wang ldquoParallel control and management for intelligenttransportation systems concepts architectures and applica-tionsrdquo IEEE Transactions on Intelligent Transportation Systemsvol 11 no 3 pp 630ndash638 2010
[25] Z Molnar D Balasubramanian and A Ledeczi ldquoAn introduc-tion to the genericmodeling environmentrdquo in Proceedings of theTOOLS Europe 2007 Workshop on Model-Driven DevelopmentTool Implementers Forum Zurich Switzerland 2007
[26] S Kelly and J-P Tolvanen Domain Framework in Domain-Specific Modeling Enabling Full Code Generation JohnWiley ampSons Hoboken NJ USA 2007
[27] B P Zeigler H Praehofer and T G Kim Theory of Modelingand Simulation Integrating Discrete Event and ContinuousComplex Dynamic Systems Academic Press San Diego CalifUSA 2000
[28] G Booch J Rumbaugh and I JacobsonThe Unified ModelingLanguage User Guide Addison-Wesley Reading Mass USA1999
[29] F Wagner R Schmuki T Wagne and P Wolstenholme Mod-eling Software with Finite State Machines A Practical ApproachCRC amp Taylor amp Francis Boca Raton Fla USA 2006
[30] T Espiner ldquoChina builds worldrsquos fastest supercomputerrdquo ZDNetUK October 2010
[31] B Chen and G Guo ldquoA two-tier parallel architecture for arti-ficial society simulationrdquo in Proceedings of the ACMIEEESCS26th Workshop on Principles of Advanced and DistributedSimulation (PADS rsquo12) pp 184ndash186 July 2012
[32] G Guo B Chen X G Qiu and Z Li ldquoParallel simulation oflarge-scale artificial society on CPUGPU mixed architecturerdquo
Mathematical Problems in Engineering 27
in Proceedings of the ACMIEEESCS 26th Workshop on Prin-ciples of Advanced and Distributed Simulation (PADS rsquo12) pp174ndash177 July 2012
[33] S-X Lun ldquoResearch on the classification of parallel executionmodes of ACP theoryrdquo Acta Automatica Sinica vol 38 no 10pp 1602ndash1608 2012
[34] F-YWang ldquoA framework for social signal processing and anal-ysis from social sensing networks to computational dialecticalanalyticsrdquo Science China vol 43 no 12 pp 1598ndash1611 2013
[35] G GuoOne Model Userrsquos Manual version 12 National Univer-sity of Defense University 2011
[36] Y Ge R Meng Z Cao X Qiu and K Huang ldquoVirtual citymdashanindividual-based digital environment for human mobility andinteractive behaviorrdquo Simulation Transactions of the Society forthe Modeling and Simulation International 2014
[37] M Kretzschmar and R T Mikolajczyk ldquoContact profiles ineight European countries and implications for modelling thespread of airborne infectious diseasesrdquo PLoS ONE vol 4 no6 Article ID e5931 2009
[38] W Duan and X Qiu ldquoFostering artificial societies using sociallearning and social control in parallel emergency managementsystemsrdquo Frontiers in Computer Science vol 6 no 5 pp 604ndash610 2012
[39] W J Edmunds C J OrsquoCallaghan and D J Nokes ldquoWhomixes with whom Amethod to determine the contact patternsof adults that may lead to the spread of airborne infectionsrdquoProceedings of the Royal Society B Biological Sciences vol 264no 1384 pp 949ndash957 1997
[40] J Banks J S Carson B L Neison and D M Nicol Discrete-Event System Simulation Prentice Hall Englewood Cliffs NJUSA 4th edition 2007
[41] H Yongxia ldquoInvestigation and analysis of college studentsrsquo workand restrdquo Value Engineering vol 31 no 3 2012 (Chinese)
[42] W Duan Z Cao Y Wang et al ldquoAn ACP approach to publichealth emergency management using a campus outbreak ofh1n1 influenza as a case studyrdquo IEEE Transactions on SystemsMan and Cybernetics Part A Systems and Humans vol 43 no5 pp 1028ndash1041 2013
[43] HWHethcote ldquoThemathematics of infectious diseasesrdquo SIAMReview vol 42 no 4 pp 599ndash653 2000
[44] G Chowell E Shim F Brauer and et al ldquoModelling thetransmission dynamics of acute haemorrhagic conjunctivitisapplication to the 2003 outbreak in Mexicordquo Statistics inMedicine vol 25 no 11 pp 1840ndash1857 2006
[45] Y Ge L Liu B Chen X Qiu and K Huang ldquoAgent-basedmodeling for InfluenzaH1N1 in an artificial classroomrdquo SystemsEngineering Procedia vol 2 pp 94ndash104 2011
[46] A R Tuite A L Greer MWhelan et al ldquoEstimated epidemio-logic parameters andmorbidity associatedwith pandemicH1N1influenzardquo CMAJ vol 182 no 2 pp 131ndash136 2010
[47] L Brouwers B Cakici M Camitz A Tegnell and MBoman ldquoEconomic consequences to society of pandemic H1N1influenza 2009mdashpreliminary results for Swedenrdquo Euro Surveil-lance vol 14 no 37 pp 1ndash7 2009
[48] Y Ge L Liu X Qiu H Song Y Wang and K Huang ldquoAframework of multilayer social networks for communicationbehavior with agent-based modelingrdquo Simulation Transactionsof the Society for the Modeling and Simulation International vol89 no 7 pp 810ndash828 2013
[49] W Duan X Qiu Z Cao X Zheng and K Cui ldquoHeterogeneousand stochastic agent-based models for analyzing infectious
diseasesrsquo super spreadersrdquo IEEE Intelligent Systems vol 28 no4 pp 18ndash25 2013
[50] The Preparation and Emergency Response Plans to PandemicPublished by Ministry of Health Ministry of Health 2005(Chinese)
[51] R M Fujimoto Parallel and Distributed Simulation SystemsWiley Interscience 2000
[52] F Cicirelli A Giordano and L Nigro ldquoDistributed simulationof situated multi-agent systemsrdquo in Proceedings of the 15thIEEEACM International Symposium on Distributed Simulationand Real Time Applications (DS-RT rsquo11) pp 28ndash35 September2011
[53] B Chen and G Guo ldquoA two-tier parallel architecturefor artificial society simulationrdquo in Proceedings of the 26thACMIEEESCS Workshop on Principles of Advanced and Dis-tributed Simulation (PADS rsquo12) pp 184ndash186 July 2012
[54] G Guo B Chen and X Qiu ldquoParallel simulation of large-scale artificial society with GPU as coprocessorrdquo InternationalJournal of Modeling Simulation and Scientific Computing vol4 no 2 Article ID 1350005 2013
[55] B Chen L-B Zhang X-C Liu and H Vangheluwe ldquoActivity-based simulation using DEVS increasing performance by anactivity model in parallel DEVS simulationrdquo Journal of ZhejiangUniversity Science C vol 15 no 1 pp 13ndash30 2014
[56] X-L Wang P Yang Z-D Cao et al ldquoQuantitative evaluationon the effectiveness of prevention and control measures againstpandemic influenza A (H1N1) in Beijing 2009rdquoChinese Journalof Epidemiology vol 31 no 12 pp 1374ndash1378 2010 (Chinese)
[57] S Mei A D van de Vijver L Xuan Y Zhu and P M ASloot ldquoQuantitatively evaluating interventions in the influenzaA (H1N1) epidemic on China campus grounded on individual-based simulationsrdquo Procedia Computer Science vol 1 no 1 pp1675ndash1682 2010
[58] X Qiu Z Cao and Z Li Mid-Term Reports of China NationalScience Foundation 91024030 2014
Submit your manuscripts athttpwwwhindawicom
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
MathematicsJournal of
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Mathematical Problems in Engineering
Hindawi Publishing Corporationhttpwwwhindawicom
Differential EquationsInternational Journal of
Volume 2014
Applied MathematicsJournal of
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Probability and StatisticsHindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Journal of
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Mathematical PhysicsAdvances in
Complex AnalysisJournal of
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
OptimizationJournal of
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
CombinatoricsHindawi Publishing Corporationhttpwwwhindawicom Volume 2014
International Journal of
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Operations ResearchAdvances in
Journal of
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Function Spaces
Abstract and Applied AnalysisHindawi Publishing Corporationhttpwwwhindawicom Volume 2014
International Journal of Mathematics and Mathematical Sciences
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
The Scientific World JournalHindawi Publishing Corporation httpwwwhindawicom Volume 2014
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Algebra
Discrete Dynamics in Nature and Society
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Decision SciencesAdvances in
Discrete MathematicsJournal of
Hindawi Publishing Corporationhttpwwwhindawicom
Volume 2014 Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Stochastic AnalysisInternational Journal of
18 Mathematical Problems in Engineering
Infected
SusceptibleHealth
withimmunity
Infected
by contac
t
with in
fected
agents
Recovery
Increase before beingsymptomatic symptomatic
Decrease after being
Decrease by time
Infectivity
Immunity
Figure 18 The transitions of agent health status
119879Lat denotes the duration of latent period 120592 120572 and 120573 arethe location parameter scale parameter and shape parameterof Weibull distribution respectively 120574 is a uniform randomnumber in the range [0 1] According to statistics [47] 120592 120572and 120573 are set as 0 18 and 121 respectively Then the mean(standard deviation) of latent period is calculated as 159 days(058 day2) Furthermore infectious period is set as 7 days
A susceptible studentrsquos probability of getting infecteddepended on the infectivity of infectious agents his ownimmunity level the duration of the contact action and soon The infectivity of an infected agent evolved with timeduring the infectious period According to the statistics ofinfectivity in the chart with red bars in Figure 19 an infectedstudent had the highest level of infectivity in the second dayafter he had the first symptom The infectivity levels in otherdays in comparison with the second day are in Table 5 Theday labeled as ldquominus1rdquo means the day before the starting of thesymptomatic period
Based on the algorithm discussed before theH1N1modelis developed with the help of MDT The settings of influenzamodel will be discussed in Section 521
512 Modelling Intervention Measures by MDT In corre-spondence intervention measures model is used to ceasethe outbreak of H1N1 As discussed in [42 48 49] if theappropriate measures are taken when the infectious diseaseemerges the transmission of the disease could be sloweddown and the damage could be decreased Therefore it isnecessary and important to design emergency interventionmeasures in the artificial Beijing According to the MinistryofHealth inChina the intervention policies [50] are designedto control the spread of influenza such as H1N1 The inter-ventionmeasures of these policies including are interventionsactivated time vaccination rate antibiotic delay to hospitalisolation duration close workspace duration and limitationof activities listed in Table 6 Similarly the models of theseintervention measures are also developed with the help ofMDT Using the reasonable data ranges listed in Table 6intervention models can be initialized to do the computa-tional experiments These experiments are used to show how
Symptomatic1 2 3 4 5 6 7minus1
Recovered
Tim
e (da
y)
Infe
ctiv
ity
Figure 19 The distribution of infectivity rate of an infected agent[37]
Table 5 The proportion of infectivity
Days (119894) minus1 1 2 3 4 5 6 6Proportion of infectivity 03 05 1 08 04 02 01 005
to restrain the H1N1 transmissionThe settings of these inter-vention measures in IMCT are introduced in Section 522
52 Experiment Settings
521 H1N1 Settings by EECT EECT is designed to initializethe emergent events models such as H1N1The settings comefrom the specified emergent event models As shown inFigure 20 in theH1N1 used case the settings are composed ofthe distribution of infection source infection rate infectionperiod and so on The infection rate changes with envi-ronments and infection periods The settings are initializedfrom both the research of medicine on H1N1 and the data ofH1N1 outbreak in Beijing in 2009The process of influenza inartificial Beijing is simulated to repeat the possible scenariosof H1N1 break During the computational experiments ofartificial Beijing theH1N1 settings such as the period of beingsymptomatic are updated by the dynamic data registrationsfrom OsdRT The updating corrects the H1N1 model duringthe experiments With the help of OsdRT artificial Beijing isable to approach actual society It is also an implementationof parallel execution
522 Intervention Measures Settings by IMCT In the H1N1case the settings of intervention measures are shown inFigure 21 in IMCT These measures are designed under theinstruction of Emergency Decision Organizations like ChinaDisease Center (CDC) The measures such as vaccinationrate the rate of closedworkspace and the starting time can beinitialized in the tool Emergency response plans are imple-mented in these settings As illustrated in Figure 7 all the pos-sible compositions within value settings in reasonable inter-vals are tested to find the most optimal emergency responseplan For example different compositions of measures suchas vaccination rates and rates of workspace closed are usedto design the plans of experiments By the analysis of theresults the best composition of interventionmeasures can beobtainedThe analysis will be discussed in case study in detail
Mathematical Problems in Engineering 19
Table 6 The intervention measures of influenza
Intervention measures ValuesInterventions activated Time 119879Intervention isin [0 80th day]Vaccination rate 119877Vaccination isin [0 1]Antibiotic rate 119877Antibiotic isin [0 1]Delay to hospital 119867ospital(119860 119894) 119879infected lt 119905 lt 119879recovered + 119879HDelay 119879HDelay isin (0 3th day]Isolation duration 119868solation(119860 119894) 119879Intervention lt 119905 lt 119879Intervention + 119879IDuration 119879IDuration isin (0 14th day]Close workspace duration 119862lose(119864119894) 119879Intervention lt 119905 lt 119879Intervention + 119879CWDuration 119864119894 isinWorkspace 119879CWDuration isin (0 7th day]Limitation of agent activity duration 119871 imitation(119860 119894) 119879Intervention lt 119905 lt 119879Intervention + 119879LAADuration 119879LAADuration isin (0 7th day]119877Vaccination means the vaccination rate of artificial Beijing while 119877Antibiotic means the antibiotic rate used for infected agents in hospital119867ospital (119860119894) means that the infected agent 119860119894 is sent to hospital he or she will receive the retreatment immediately119868solation (119860119894) means that the infected agent 119860119894 is isolated at home he or she can only contact with the other agents until isolation duration is over119862lose (119864119894) means that the environment 119864119894 typed of workspace is closed until the close workspace duration is over119871 imitation (119860119894) means that the activities of agent 119860119894 are limited only the workspace and home are allowed until limitation duration is over
Incubation Weibulldistribution
Symbolic LogNormaldistribution
Recovering LogNormaldistribution
H1N1influenza
Settings of H1N1 influenza model
Disease model
Transmissionmodel
Figure 20 The settings of H1N1 in EECT
It is worth notifying that the settings of interventionmeasures are also influenced by OsdRT For example theH1N1 outbreak areas are notified by OsdRTThe interventionmeasures such as close workspace are used only in theoutbreak areas Therefore the unnecessary computation ofintervention measures is avoided So the data injection byOsdRT not only makes artificial society approach actualsociety but also increases the performance of computationalexperiments
53 Experiments by EMT When the initialization of arti-ficial Beijing is ready EMT is used to do the computa-tional experiments EMT is composed of two parts thecentral controller in the central node and the residen-tial service deployed in the computational nodes Centralcontroller loads the experiment plans and controls the whole
experiments process the GUI of controller is shown inFigure 22 EMT manages the computational experimentsin four steps cluster configuration load experiment plansmodels deployment and running experiments
531 Cluster Configuration Before the configuration clusterinformation is collected from the residential services insidethe cluster The information includes nodes amount nodenames node IPs and CPU occupations It is used to quantifythe computing power of the clusterThe cluster configurationis based on this information The nodes are selected forexperiments Then the LP (local process) [51] number foreach node is set according to the number of the CPU kernelsAfter the configuration the experiment plans are loaded tocustomize the models and data
20 Mathematical Problems in Engineering
Streets of close workspaces
Limitations ofactivities
IsolationsRespirator
Closeworkspace
Antibiotic
VaccinationHospital
Surveillance
Figure 21 The settings of intervention measures in IMCT
Information ofexperiment plan
List of experiment plans
Start | Pause | Reset | Stop | Speedup | Slowdown | Back
NodeName | CPU occupation | Memory occupation |Network | Models | Status
Figure 22 The deployment and running of experiments in EMT
532 Load Experiment Plan As discussed in Section 35 thedescription of experiment plans is composed of the infor-mation in three aspects models and the initial data for theexperiments the deployment mapping tables from the mod-els to the nodes and the plan of experiment executionWhenthe experiment plan is loaded by the EMT the file namesand model descriptions of models and data are listed Withthe support of cluster configuration models can be mappedto the nodes in the GUI according to the computationrequirements EMT also provides the setting of experimentsitself including the running times start time and end time
533 Models and Data Deployment Models are stored inthe agent population and services repository According tothe settings in experiment plan the customized models aredownloaded from the repository first then the models suchas agent model environment model emergent events modeland intervention measures model are integrated to be thecompositionmodels of artificial BeijingThemodels and datalocations in the nodes are configured in the ldquomodel pathrdquoand ldquodata pathrdquo in the GUI of central controller Finallythe models are uploaded to the nodes in cluster with theconsideration of partition The partition of artificial society
Mathematical Problems in Engineering 21
models is different from the multiagent system in [52]Models in artificial society are divided into three levels agentsand environments grids and city Agents and environmentsare encapsulated in grids which compose the whole cityThus the partition of artificial society becomes the partitionof grids The parallelism degrees are found to support thepartition of models agent independence and grid indepen-dence According to the parallelisms agents are partitioned toLPs according to the population distribution in grids Agentmovements among grids are simulated by the movementevents among grids in different LPs As a result a two-tierparallel architecture is proposed to support the simulation ofartificial society The architecture is detailed in [53]
534 Running of Experiments When the preparation isready the experiments can be done by the central controllerof EMT The running and control commands of the experi-ments are grouped into three aspects start pause or stop theexperiments get the running information of the nodes suchas the snapshot of the desktop of node With the help of thesecommands users can run the computational experimentsto do the research on the emergency management It isworth notifying that the simulation engine is optimized inorder to support the ten million agent simulation UsingGPU as coprocessor a two-tier parallel simulation engine isdesigned with support of MPI and OpenCL through phasedsynchronization [54] One-sided communication is used forreflection of remote simulation objects and message passingbetween processes A general kernel function prototype iselaborately designed and conditionally compiled for execu-tion on both CPU and GPU Moreover the optimizationoperations like load balance is developed under the instruc-tion of activity-based simulation [55] The densely populatedgrids are given more CPU and memory resource accordingto the activity predictions
54 Results of H1N1 Experiments with KD-ACP On thebasis of artificial society models and the intervention mea-sures mentioned in Section 512 a series of computationalexperiments are performed to study the H1N1 influenza inBeijing In our case of artificial Beijing 19610000 agentsare simulated in the cluster within 48 CPU cores and 128Gmemories the nodes are connected by kilomega networks Ittakes 18 hours to simulate a 250-day disease spreading
In order to illustrate how the optimal plans are obtainedbyKD-ACP the experiments are divided into four groups theexperiments ofmodel validation the experiments of sensitiv-ity analysis of vaccination rate the experiments of sensitivityanalysis of isolation and the experiments of the combinationof interventionmeasuresThese four groups show the generalprocess of the research on emergency management in H1N1influenza Firstly the models of H1N1 are validated withthe support of historical data Secondly the traditionalmedical interventions of pandemic such as vaccination areanalyzed to find the optimal plan Thirdly the traditionalnonmedical interventions of pandemic such as isolation areanalyzed to find the optimal plan Finally the combination ofinterventions is tested to show the combined effect
It is worth mentioning that all the experiments in ourwork are performed 100 times These experiments are ini-tialized with different settings of interventions in IMCT Theresults shown in the figures are the mean values
541 The Experiments of Model Validation In our artifi-cial Beijing the models of H1N1 influenza are built basedon the existing researches Wang and her colleagues hadstudied the H1N1 influenza by SEIR (SusceptibleExposedInfectiveRecovered) models [56] With the support of theirwork we build the models of artificial Beijing in the mannerof multiagents
In Figure 23 the control group shown in red line isdrawn based on the realistic statistical data collected in H1N1influenza in Beijing in 2009 The influenza lasts more thansix months and more than 170000 people are infectedComparing with the control group the simulated resultswithout interventions are drawn in blue It is obvious that thesimulated case fits the control group well before 70 days Butthe ascending of control group is slowed after 70 days andthe peak value (17883) is much less than the simulated results(25135) It is because the vaccination is activated at day 67 incontrol group according to the intervention measures usedin 2009 The restrain by vaccination becomes effective eightdays later The change of infected number after vaccinationis shown by red spotted line The trend of control group isdifferent from the simulation data after 75 days
However the simulated results (blue line) in compu-tational experiments fit the control group (as shown inFigure 23) in the first 70 days when no intervention mea-sures are executed The key features such as effective basicreproduction number (119877
0) are also almost the same in both
cases Therefore it can be concluded that the models ofartificial society built in microview are validated by therealistic statistical data in influenza in Beijing in 2009
In this group of experiments no interventions are exe-cuted in order to validate the models Therefore the simu-lated results of this extreme case which cannot happen todaycould be obtained As shown by blue line the number ofinfected agents without interventions reaches peak value atday 91 with 25135 consisting of 8052 in symptomatic phaseand 28950 in recovery phase It is obvious that the totalnumber of infected agents grows slowly at the beginningtime Then the number increases quickly when the influenzabreaks and it reaches its maximum value with 383870 in theend This phenomenon can be explained as in the followingAt the beginning of influenza due to the limited numberof infected agents the propagation of the disease remainsslow but many agents become susceptible When the numberof infected agents increases rapidly the spreading becomessignificant However the number of infected agents decreasesafter day 91 It is because the disease spreading relies on thesocial networks and spatial contact networks [49] It meansthat the infected agents can only infect agents either in thesocial relationship networks or in the spatially contacted
542 The Experiments of Sensitivity Analysis of VaccinationRate According to Section 512 vaccination is a typical
22 Mathematical Problems in Engineering
0
Infe
cted
agen
ts
0 50 100 150 200 250
Statistical dataSimulation data
25000
20000
15000
10000
5000
Day d
Figure 23 The results with no intervention executed and the com-parison with realistic statistical data in Beijing in 2009
medical intervention in influenza Traditionally the effective-ness of vaccination can only be evaluated after the influenzaand the key settings such as vaccination rate are determinedby the experience It is always difficult for emergency man-agement organizations to find the most optimal vaccinationrate in advanceTherefore the sensitivity of vaccination rate isanalyzed in this group of experiments The settings are listedin Table 7 average results are shown in Figure 24
Vaccination rates are set linearly (10 30 50 and70) to test the changes of influenza The peak value ofinfected agents decreases from 25135 22392 15787 6907to 638 respectively It shows that the more the agents arevaccinated the smaller the peak value of the infected agentswill be In the case of 70 vaccination the total infected agentis only 13799 It means that the influenza in this case does nothappen due to the high vaccination rate
The change of peak values also shows an interestingphenomenon Vaccination rates increase linearly the peakvalues decrease slower than the ascending of vaccination ratesat beginning but the descending becomes more and morefaster with the linear increase of vaccination rate It meansthat the small scale of vaccination (less than 30) does notmake senseWith the increase of vaccination rate the numberof susceptible agents decreases significantly In sequencethe possibility of infections is lowered greatly Additionallyit could be found that the decrease of peak values comeslater than the increase of the vaccination rates This is quitereasonable according to Mei et alrsquos work in [57]
As a result it is concluded that high vaccination rate iseffective in the control of H1N1 influenza though costly
543 The Experiments of Sensitivity Analysis of IsolationAccording to Section 512 isolation is a typical nonmedicalintervention It is the most used intervention measure in
Table 7 The initial settings of experiments of vaccination
Parameters ValueAgent count 19610000Initial infected count 40Interventions activated time(119879Intervention)
0
Vaccination rate (119877Vaccination) 0 10 30 50 70
0
Infe
cted
agen
ts
Loading intervention of vaccination
No vaccination10 vaccination30 vaccination
50 vaccination70 vaccination
0 50 100 150 200 250
25000
20000
15000
10000
5000
Day d
Figure 24 The results of loading intervention of vaccination
the emergency management in influenza As mentionedbefore a problem also exists that the measure cannot bevalidated before the intervention Therefore the sensitivityof isolation is analyzed in this group of experiments It isworth mentioning that the isolation is usually combinedwith the hospital policy It means the infected individual issent to hospital meanwhile the people who contact withthe infected are all isolated As a result parameters delay tohospital and isolation duration are designed together in thesettings of experiment Delay to hospital stands for the periodfrom infection time to hospital time while Isolation durationrepresents the period when an agent cannot contact others ifhe has contact with an infected agent The settings of theseparameters are listed in Table 8 average results are shown inFigure 25
Firstly the parameter delay to hospital is simulated aloneto test the sensitivity The simulated results are shown insolid line in Figure 25 The peak value decreases from 6708to 2218 when the delay to hospital decreases from 15 daysto 1 day The half day change brings the 67 reductionMoreover the peak value almost approximates to zero inthe case of 05 days The influenza does not happen in thiscase It can be concluded that the earlier the infected agents
Mathematical Problems in Engineering 23
Table 8 The initial settings of experiments of isolation
Parameters ValueAgent count 19610000Initial infected count 40Interventions activated time (119879Intervention) 80Delay to hospital (119879HDelay) isolation duration (119879IDuration) (15 0) (1 0) (05 0) (15 3) (1 3) (05 3)
0 50 100 150 200 2500
1000
2000
3000
4000
5000
6000
7000
Infe
cted
agen
ts
Loading intervention of isolation
Delay to hospital = 15
Delay to hospital = 10
Delay to hospital = 05
Delay to hospital = 15 isolation duration = 3
Delay to hospital = 10 isolation duration = 3
Delay to hospital = 05 isolation duration = 3
Day d
Figure 25 The results of loading intervention of isolation
are sent to hospital the less the agents will be infectedSo the construction of effective mechanism to find infectedindividuals is important in the emergency management ofinfluenza
Secondly the parameter isolation duration is added tosimulate the isolation of interventionObviously the isolationmeasure decreases the infected agents greatly Compared tothe parameter delay to hospital of 15-day case the peak valuedecreases from 6708 to 362when the isolation is added in theinterventionThe reduction of 6346 agents comes from the 3-day isolation of agents who had contacted with the infectedagents However the decrease of parameter delay to hospitalcannot bring the similar reduction in the 3-day isolationcases According to the figure the peak value decreases from362 to 51 from 15-day case to 05-day case only 311 agents aresaved from infection As a result the sensitivity of simulatedresults does not change obviously with delay to hospital inthe case of isolation It can be concluded that isolation is amore effective measure Though it is not easy for emergencymanagement organizations to decrease the delay to hospitaldue to the current monitoring mechanism the influenza canstill be controlled by isolation
544 The Experiments of Combination of Intervention Mea-sures Isolation is not the only nonmedical intervention ininfluenza close workspace is another common measureWhen close workspace is activated the workspaces are allclosed agents can only stay home at the closed time Luckilyisolation and close workspace are independent from oneanother It is possible for emergency management organi-zations to use these two measures together So this groupof experiments is designed to obtain the effectiveness of thecombination of intervention measures The parameters areinitialized with different settings listed in Table 9
Parameters delay to hospital and isolation duration is setin the middle values (1 day 3 days) according to the lastsection According to the experience close duration is set asseven days It is known that close workspace is a really costintervention measure the activated time is a key parameterWith the fixed parameters mentioned before the changeof interventions activated time is analyzed in this group ofcomputational experimentsThe average results are shown inFigure 26
Compared with the simulated results of experimentsdiscussed above Figure 26 strongly illustrates that influenzais ceased when the intervention measures are combinedtogether The peak value of infected agents is 141 in the ldquo80thdays of close-workspace and isolation with 1 day delay tohospitalrdquo case while 25135 in ldquono interventionrdquo case 2218 inldquo1 day delay to hospitalrdquo case 203 in ldquo1 day delay to hospitaland isolationrdquo case The interventions bring a great reductionof infection the peak values decrease from ten thousandslevel to hundred level
On the other hand the peak value even decreases to 108when the execution time of close workspace is 10 days earlierThe date of maximum value of infected agents is advancedfrom the 135th day to the 108th day The whole circle of H1N1influenza is reducedThe results of this group of experimentsshow that intervention measures should be activated as soonas possible and the compositions of interventions are moreeffective in emergency management of H1N1 influenza
In summary based on the general process lots of exper-iments could be done by KD-ACP to study emergency man-agement in H1N1 influenza Many conclusions are obtainedbased on the analysis of the experiments in this sectionAccording to these conclusions themost optimal response inour case study is the case of ldquo70th days of close-workspace andisolation with 1 day delay to hospitalrdquo It is because this casehas the fewest infected agents relatively But the modelling ofH1N1 influenza is not sufficient Economic and social factorsof interventions such as the money cost by vaccination andisolation the influence of the close-workspace in society
24 Mathematical Problems in Engineering
Table 9 The initial settings of experiments of isolation and closeworkspace
Parameters ValueAgent count 19610000Initial infected count 40Interventions activated time (119879Intervention) 70th day 80th dayDelay to hospital (119879HDelay) 1 dayIsolation duration (119879IDuration) 3 daysClose workspace duration (119879CWDuration) 7 days
50 100 150 200 2500
50
100
150
200
Infe
cted
agen
ts
Loading intervention of isolation and close workspace
Delay to hospital = 1 isolation duration = 3Start close time = 70 close duration time = 7Delay to hospital = 1 isolation duration = 3Start close time = 80 close duration time = 7
Day d
Figure 26The results of loading interventions of isolation and closeworkspace
and even opinion transmission of H1N1 in internet are notconsidered in our case study Without the analysis of thesefactors it does not make sense to talk about the most optimalintervention measures However KD-ACP had provided asoftware framework and general process of the research onemergency management Supported by KD-ACP the expertsof emergency management are able to study emergent eventssuch as H1N1 influenza as detailed as possible Based on theexperiments and analysis again and again the most optimalintervention measures will be found ultimately
55 Parallel Execution by ASST and OsdRT As discussed inSection 35 ASST andOsdRT are used to support the parallelexecution of artificial Beijing The statistical data is collectedand stored in artificial society runtime database at runtimeASST downloads the infection data from the database anddisplays it in the map As shown in Figure 27 the infectedagents are distributed in the districts in BeijingThe snapshotgives the geospatial distribution of the H1N1 epidemic Thelegends of the agents in different status are listed in the left
right corner in Figure 27 Five statuses are shown in the figurehealth without vaccination health with vaccination incuba-tion being symptomatic and convalescence ASST providesa very direct feeling for the emergency experts Partly withthe situation from ASST Emergency Decision Organizationmakes the emergency response plans to the response of theemergency scenario With the help of plans the epidemicin actual society can be restrained In the meantime theinfluenza information is reflected in the Internet Accordingto Figure 8 the data acquisition data extraction and datastandardization of H1N1 influenza are processed in OsdRTOsdRT is implemented by Dr Caorsquos team from Institute ofAutomation from Chinese Academy of Sciences [58] Thetool provides not only the reports of H1N1 SARS hand-foot-mouth disease and intestinal diseases are all monitored Theweb page is shown in Figure 28 H1N1 reports from Internetare collected first Based on the data extraction of the reportsH1N1 case of distribution is obtained and shown in theincidencemap As a result H1N1 incidence data is regarded asthe knowledge which is sent to EECT and IMCTThe settingsof emergent events and intervention measures are modifiedin order to make artificial society approach actual society innext turn of computational experiments With the help ofOsdRT the loop of parallel execution process integrates ACPapproach
6 Conclusion
This work provides an integrated software framework forsocial Computing in emergencymanagement From a systemperspective KD-ACP provides a reliable flexible low costand effective platform for the scientists to do the research onemergency response problems The analysis and predictionof these problems with inherent complexity can be solvedby the repetition of possible alternative experiments onthe software framework Therefore KD-ACP which is anattempt to implement ACP approach seemed as an effectivecomputational framework to support the decision making inemergency response
Currently KD-ACP is used to study the H1N1 epidemicin Beijing in 2009 The most optimal compositions of inter-vention measures are testified to support the response tonext influenza However it is still a long way ahead beforethe automation systematization and practicalization of KD-ACP A lot of work will be carried out along several directionsas in the following
(i) The parallel execution process in KD-ACP is actuallynot strongly connected with the emergency responseorganization Traditional experience and theory stillplay a dominant role in the decision of emergencymanagement It is necessary to propose a mechanismof hall for workshop to facilitate the parallel executionloop in actual society
(ii) Section 33 introduces the principle and process ofmodelling artificial society Metamodels and modelsof artificial Beijing are built by GME Model transfor-mation and code generation are used to implementcodes from FSM based models But many details
Mathematical Problems in Engineering 25
Fangshan district Daxing district Fengtai district Xicheng district
Mentougou district Haidian districtShijingshan district Dongcheng district
Incubation
Symptomatic
Convalescence
Health withoutvaccination
Health withvaccinationLegend of agent
health status
Figure 27 The situation of artificial Beijing by ASST
H1N1
Districts of Beijing
H1N1 case ofdistribution
Incidence map
Distribution ofH1N1 reportsfrom internet
Incidence of H1N1
Fangshan districtLiangxiang town
Population 19000Incidence 135789
(1100000)
Figure 28 The implementation of OsdRT by Institute of Automation from Chinese Academy of Sciences
are still not illustrated due to the topic of thispaper The key techniques such as template of codeframework and model transformations from specificdomains to FSM will be focused and introduced indetail in our next paper
(iii) The performance of computational experiments inKD-ACP is still slower than expectation The opti-mization of simulation engine and models structureis necessary to improve the performance So it ispossible to run artificial society with the real worldin parallel
In summary KD-ACP paves a new way for scientists inboth emergencymanagement and computation simulation to
collaborate with each other in solving the emergency man-agement problems by social computing
Research Highlights
(i) KD-ACP is implemented based on ACP approach
(ii) Artificial Beijing is built with the help of KD-ACP
(iii) H1N1 is modeled to simulate the influenza in Beijingin 2009
(iv) Computational experiments testify the effectivenessof intervention measures
26 Mathematical Problems in Engineering
Conflict of Interests
The authors declare that there is no conflict of interestsregarding the publication of this paper
Acknowledgment
This work was supported in part by the National ScienceFoundation of China (nos 91024030 71303252 61403402and 91324013)
References
[1] W Mao and F-Y Wang Advances in Intelligence and SecurityInformatics Academic Press Oxford UK 2012
[2] E Bonabeau ldquoAgent-based modeling methods and techniquesfor simulating human systemsrdquo Proceedings of the NationalAcademy of Sciences of the United States of America vol 99 no3 pp 7280ndash7287 2002
[3] K M Carley D B Fridsma E Casman et al ldquoBioWar scalableagent-basedmodel of bioattacksrdquo IEEETransactions on SystemsMan and Cybernetics A Systems and Humans vol 36 no 2 pp252ndash265 2006
[4] J M Epstein ldquoModelling to contain pandemicsrdquo Nature vol460 no 7256 p 687 2009
[5] Los Alamos National Laboratory Reports about EPISIMS LosAlamos National Laboratory 2003
[6] C H Builder and S C Bankes ldquoArtificial societies a conceptfor basic research on the societal impacts of informationtechnologyrdquo RAND Report 149 RAND Corp 1991
[7] F-Y Wang ldquoComputational theory and method on complexsystemrdquo China Basic Science vol 6 no 5 pp 3ndash10 2004
[8] F-Y Wang ldquoToward a paradigm shift in social computing theACP approachrdquo IEEE Intelligent Systems vol 22 no 5 pp 65ndash67 2007
[9] F-Y Wang and S Tang ldquoArtificial societies for integratedand sustainable development of metropolitan systemsrdquo IEEEIntelligent Systems vol 19 no 4 pp 82ndash87 2004
[10] D Wen Y Yuan and X-R Li ldquoArtificial societies computa-tional experiments and parallel systems an investigation ona computational theory for complex socioeconomic systemsrdquoIEEE Transactions on Services Computing vol 6 no 2 pp 177ndash185 2013
[11] F Wang F Zeng and Y Yuan ldquoAn ACP-based approach forcomplexity analysis of E-commerce systemrdquo Complex Systemsand Complexity Science vol 5 no 3 pp 1ndash8 2008 (Chinese)
[12] F-Y Wang and P K Wong ldquoResearch commentary intelligentsystems and technology for integrative and predictivemedicinean ACP approachrdquoACMTransactions on Intelligent Systems andTechnology vol 4 no 2 pp 1ndash6 2013
[13] J Sifeng X Gang F Dong and H Chunpeng ldquoStudy on theemergency rescue decision support system of petrochemicalplant based on ACP theoryrdquo in Proceedings of the 6th Manage-ment Annual Meeting (MAM rsquo11) 2011 (Chinese)
[14] X Dong D Fan G Xiong F Zhu Z Zhang and Y Yao ldquoPar-allel Bus Rapid Transit (BRT) operation management systembased on ACP approachrdquo in Proceedings of the 6th ManagementAnnual Meeting (MAM rsquo11) Beijing China 2011
[15] B Ning F-Y Wang H-R Dong R-M Li D Wen and L LildquoParallel systems for Urban rail transportation based on ACP
approachrdquo Journal of Transportation Systems Engineering andInformation Technology vol 10 no 6 pp 23ndash28 2010 (Chinese)
[16] F-Y Wang J Zhao and S-X Lun ldquoArtificial power systemsfor the operation and management of complex power gridsrdquoSouthern Power System Technology vol 2 no 3 pp 1ndash11 2008(Chinese)
[17] M J North T R Howe N T Collier and J R Vos ldquoRepastsimphony development environmentrdquo in Proceedings of theAgent 2005 Conference on Generative Social Processes Modelsand Mechanisms 2005
[18] M J North T RHowe N T Collier and J R Vos ldquoRepast Sim-phony runtime systemrdquo in Proceedings of the Agent Conferenceon Generative Social Processes Models and Mechanisms 2005
[19] SWARM Development Group Swarm 211 Reference Guide2000 httpwwwswarmorg
[20] S Tisue and U Wilensky ldquoNetLogo a simple environment formodeling complexityrdquo in International Conference on ComplexSystems pp 16ndash21 2004
[21] S Luke C Cioffi-Revilla L Panait K Sullivan and G BalanldquoMASON a multiagent simulation environmentrdquo Simulationvol 81 no 7 pp 517ndash527 2005
[22] M Ichikawa H Tanuma Y Koyama andH Deguchi ldquoSOARSintroduction as a social microscope for simulations of socialinteractions and gamingrdquo in Organizing and Learning throughGaming and Simulation Proceedings of the 38th Conference ofthe International Simulation and Gaming Association (ISAGArsquo07) pp 149ndash158 2007
[23] D S Burke J M Epstein D A T Cummings et al ldquoIndividual-based computational modeling of smallpox epidemic controlstrategiesrdquo Academic Emergency Medicine vol 13 no 11 pp1142ndash1149 2006
[24] F-Y Wang ldquoParallel control and management for intelligenttransportation systems concepts architectures and applica-tionsrdquo IEEE Transactions on Intelligent Transportation Systemsvol 11 no 3 pp 630ndash638 2010
[25] Z Molnar D Balasubramanian and A Ledeczi ldquoAn introduc-tion to the genericmodeling environmentrdquo in Proceedings of theTOOLS Europe 2007 Workshop on Model-Driven DevelopmentTool Implementers Forum Zurich Switzerland 2007
[26] S Kelly and J-P Tolvanen Domain Framework in Domain-Specific Modeling Enabling Full Code Generation JohnWiley ampSons Hoboken NJ USA 2007
[27] B P Zeigler H Praehofer and T G Kim Theory of Modelingand Simulation Integrating Discrete Event and ContinuousComplex Dynamic Systems Academic Press San Diego CalifUSA 2000
[28] G Booch J Rumbaugh and I JacobsonThe Unified ModelingLanguage User Guide Addison-Wesley Reading Mass USA1999
[29] F Wagner R Schmuki T Wagne and P Wolstenholme Mod-eling Software with Finite State Machines A Practical ApproachCRC amp Taylor amp Francis Boca Raton Fla USA 2006
[30] T Espiner ldquoChina builds worldrsquos fastest supercomputerrdquo ZDNetUK October 2010
[31] B Chen and G Guo ldquoA two-tier parallel architecture for arti-ficial society simulationrdquo in Proceedings of the ACMIEEESCS26th Workshop on Principles of Advanced and DistributedSimulation (PADS rsquo12) pp 184ndash186 July 2012
[32] G Guo B Chen X G Qiu and Z Li ldquoParallel simulation oflarge-scale artificial society on CPUGPU mixed architecturerdquo
Mathematical Problems in Engineering 27
in Proceedings of the ACMIEEESCS 26th Workshop on Prin-ciples of Advanced and Distributed Simulation (PADS rsquo12) pp174ndash177 July 2012
[33] S-X Lun ldquoResearch on the classification of parallel executionmodes of ACP theoryrdquo Acta Automatica Sinica vol 38 no 10pp 1602ndash1608 2012
[34] F-YWang ldquoA framework for social signal processing and anal-ysis from social sensing networks to computational dialecticalanalyticsrdquo Science China vol 43 no 12 pp 1598ndash1611 2013
[35] G GuoOne Model Userrsquos Manual version 12 National Univer-sity of Defense University 2011
[36] Y Ge R Meng Z Cao X Qiu and K Huang ldquoVirtual citymdashanindividual-based digital environment for human mobility andinteractive behaviorrdquo Simulation Transactions of the Society forthe Modeling and Simulation International 2014
[37] M Kretzschmar and R T Mikolajczyk ldquoContact profiles ineight European countries and implications for modelling thespread of airborne infectious diseasesrdquo PLoS ONE vol 4 no6 Article ID e5931 2009
[38] W Duan and X Qiu ldquoFostering artificial societies using sociallearning and social control in parallel emergency managementsystemsrdquo Frontiers in Computer Science vol 6 no 5 pp 604ndash610 2012
[39] W J Edmunds C J OrsquoCallaghan and D J Nokes ldquoWhomixes with whom Amethod to determine the contact patternsof adults that may lead to the spread of airborne infectionsrdquoProceedings of the Royal Society B Biological Sciences vol 264no 1384 pp 949ndash957 1997
[40] J Banks J S Carson B L Neison and D M Nicol Discrete-Event System Simulation Prentice Hall Englewood Cliffs NJUSA 4th edition 2007
[41] H Yongxia ldquoInvestigation and analysis of college studentsrsquo workand restrdquo Value Engineering vol 31 no 3 2012 (Chinese)
[42] W Duan Z Cao Y Wang et al ldquoAn ACP approach to publichealth emergency management using a campus outbreak ofh1n1 influenza as a case studyrdquo IEEE Transactions on SystemsMan and Cybernetics Part A Systems and Humans vol 43 no5 pp 1028ndash1041 2013
[43] HWHethcote ldquoThemathematics of infectious diseasesrdquo SIAMReview vol 42 no 4 pp 599ndash653 2000
[44] G Chowell E Shim F Brauer and et al ldquoModelling thetransmission dynamics of acute haemorrhagic conjunctivitisapplication to the 2003 outbreak in Mexicordquo Statistics inMedicine vol 25 no 11 pp 1840ndash1857 2006
[45] Y Ge L Liu B Chen X Qiu and K Huang ldquoAgent-basedmodeling for InfluenzaH1N1 in an artificial classroomrdquo SystemsEngineering Procedia vol 2 pp 94ndash104 2011
[46] A R Tuite A L Greer MWhelan et al ldquoEstimated epidemio-logic parameters andmorbidity associatedwith pandemicH1N1influenzardquo CMAJ vol 182 no 2 pp 131ndash136 2010
[47] L Brouwers B Cakici M Camitz A Tegnell and MBoman ldquoEconomic consequences to society of pandemic H1N1influenza 2009mdashpreliminary results for Swedenrdquo Euro Surveil-lance vol 14 no 37 pp 1ndash7 2009
[48] Y Ge L Liu X Qiu H Song Y Wang and K Huang ldquoAframework of multilayer social networks for communicationbehavior with agent-based modelingrdquo Simulation Transactionsof the Society for the Modeling and Simulation International vol89 no 7 pp 810ndash828 2013
[49] W Duan X Qiu Z Cao X Zheng and K Cui ldquoHeterogeneousand stochastic agent-based models for analyzing infectious
diseasesrsquo super spreadersrdquo IEEE Intelligent Systems vol 28 no4 pp 18ndash25 2013
[50] The Preparation and Emergency Response Plans to PandemicPublished by Ministry of Health Ministry of Health 2005(Chinese)
[51] R M Fujimoto Parallel and Distributed Simulation SystemsWiley Interscience 2000
[52] F Cicirelli A Giordano and L Nigro ldquoDistributed simulationof situated multi-agent systemsrdquo in Proceedings of the 15thIEEEACM International Symposium on Distributed Simulationand Real Time Applications (DS-RT rsquo11) pp 28ndash35 September2011
[53] B Chen and G Guo ldquoA two-tier parallel architecturefor artificial society simulationrdquo in Proceedings of the 26thACMIEEESCS Workshop on Principles of Advanced and Dis-tributed Simulation (PADS rsquo12) pp 184ndash186 July 2012
[54] G Guo B Chen and X Qiu ldquoParallel simulation of large-scale artificial society with GPU as coprocessorrdquo InternationalJournal of Modeling Simulation and Scientific Computing vol4 no 2 Article ID 1350005 2013
[55] B Chen L-B Zhang X-C Liu and H Vangheluwe ldquoActivity-based simulation using DEVS increasing performance by anactivity model in parallel DEVS simulationrdquo Journal of ZhejiangUniversity Science C vol 15 no 1 pp 13ndash30 2014
[56] X-L Wang P Yang Z-D Cao et al ldquoQuantitative evaluationon the effectiveness of prevention and control measures againstpandemic influenza A (H1N1) in Beijing 2009rdquoChinese Journalof Epidemiology vol 31 no 12 pp 1374ndash1378 2010 (Chinese)
[57] S Mei A D van de Vijver L Xuan Y Zhu and P M ASloot ldquoQuantitatively evaluating interventions in the influenzaA (H1N1) epidemic on China campus grounded on individual-based simulationsrdquo Procedia Computer Science vol 1 no 1 pp1675ndash1682 2010
[58] X Qiu Z Cao and Z Li Mid-Term Reports of China NationalScience Foundation 91024030 2014
Submit your manuscripts athttpwwwhindawicom
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
MathematicsJournal of
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Mathematical Problems in Engineering
Hindawi Publishing Corporationhttpwwwhindawicom
Differential EquationsInternational Journal of
Volume 2014
Applied MathematicsJournal of
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Probability and StatisticsHindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Journal of
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Mathematical PhysicsAdvances in
Complex AnalysisJournal of
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
OptimizationJournal of
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
CombinatoricsHindawi Publishing Corporationhttpwwwhindawicom Volume 2014
International Journal of
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Operations ResearchAdvances in
Journal of
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Function Spaces
Abstract and Applied AnalysisHindawi Publishing Corporationhttpwwwhindawicom Volume 2014
International Journal of Mathematics and Mathematical Sciences
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
The Scientific World JournalHindawi Publishing Corporation httpwwwhindawicom Volume 2014
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Algebra
Discrete Dynamics in Nature and Society
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Decision SciencesAdvances in
Discrete MathematicsJournal of
Hindawi Publishing Corporationhttpwwwhindawicom
Volume 2014 Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Stochastic AnalysisInternational Journal of
Mathematical Problems in Engineering 19
Table 6 The intervention measures of influenza
Intervention measures ValuesInterventions activated Time 119879Intervention isin [0 80th day]Vaccination rate 119877Vaccination isin [0 1]Antibiotic rate 119877Antibiotic isin [0 1]Delay to hospital 119867ospital(119860 119894) 119879infected lt 119905 lt 119879recovered + 119879HDelay 119879HDelay isin (0 3th day]Isolation duration 119868solation(119860 119894) 119879Intervention lt 119905 lt 119879Intervention + 119879IDuration 119879IDuration isin (0 14th day]Close workspace duration 119862lose(119864119894) 119879Intervention lt 119905 lt 119879Intervention + 119879CWDuration 119864119894 isinWorkspace 119879CWDuration isin (0 7th day]Limitation of agent activity duration 119871 imitation(119860 119894) 119879Intervention lt 119905 lt 119879Intervention + 119879LAADuration 119879LAADuration isin (0 7th day]119877Vaccination means the vaccination rate of artificial Beijing while 119877Antibiotic means the antibiotic rate used for infected agents in hospital119867ospital (119860119894) means that the infected agent 119860119894 is sent to hospital he or she will receive the retreatment immediately119868solation (119860119894) means that the infected agent 119860119894 is isolated at home he or she can only contact with the other agents until isolation duration is over119862lose (119864119894) means that the environment 119864119894 typed of workspace is closed until the close workspace duration is over119871 imitation (119860119894) means that the activities of agent 119860119894 are limited only the workspace and home are allowed until limitation duration is over
Incubation Weibulldistribution
Symbolic LogNormaldistribution
Recovering LogNormaldistribution
H1N1influenza
Settings of H1N1 influenza model
Disease model
Transmissionmodel
Figure 20 The settings of H1N1 in EECT
It is worth notifying that the settings of interventionmeasures are also influenced by OsdRT For example theH1N1 outbreak areas are notified by OsdRTThe interventionmeasures such as close workspace are used only in theoutbreak areas Therefore the unnecessary computation ofintervention measures is avoided So the data injection byOsdRT not only makes artificial society approach actualsociety but also increases the performance of computationalexperiments
53 Experiments by EMT When the initialization of arti-ficial Beijing is ready EMT is used to do the computa-tional experiments EMT is composed of two parts thecentral controller in the central node and the residen-tial service deployed in the computational nodes Centralcontroller loads the experiment plans and controls the whole
experiments process the GUI of controller is shown inFigure 22 EMT manages the computational experimentsin four steps cluster configuration load experiment plansmodels deployment and running experiments
531 Cluster Configuration Before the configuration clusterinformation is collected from the residential services insidethe cluster The information includes nodes amount nodenames node IPs and CPU occupations It is used to quantifythe computing power of the clusterThe cluster configurationis based on this information The nodes are selected forexperiments Then the LP (local process) [51] number foreach node is set according to the number of the CPU kernelsAfter the configuration the experiment plans are loaded tocustomize the models and data
20 Mathematical Problems in Engineering
Streets of close workspaces
Limitations ofactivities
IsolationsRespirator
Closeworkspace
Antibiotic
VaccinationHospital
Surveillance
Figure 21 The settings of intervention measures in IMCT
Information ofexperiment plan
List of experiment plans
Start | Pause | Reset | Stop | Speedup | Slowdown | Back
NodeName | CPU occupation | Memory occupation |Network | Models | Status
Figure 22 The deployment and running of experiments in EMT
532 Load Experiment Plan As discussed in Section 35 thedescription of experiment plans is composed of the infor-mation in three aspects models and the initial data for theexperiments the deployment mapping tables from the mod-els to the nodes and the plan of experiment executionWhenthe experiment plan is loaded by the EMT the file namesand model descriptions of models and data are listed Withthe support of cluster configuration models can be mappedto the nodes in the GUI according to the computationrequirements EMT also provides the setting of experimentsitself including the running times start time and end time
533 Models and Data Deployment Models are stored inthe agent population and services repository According tothe settings in experiment plan the customized models aredownloaded from the repository first then the models suchas agent model environment model emergent events modeland intervention measures model are integrated to be thecompositionmodels of artificial BeijingThemodels and datalocations in the nodes are configured in the ldquomodel pathrdquoand ldquodata pathrdquo in the GUI of central controller Finallythe models are uploaded to the nodes in cluster with theconsideration of partition The partition of artificial society
Mathematical Problems in Engineering 21
models is different from the multiagent system in [52]Models in artificial society are divided into three levels agentsand environments grids and city Agents and environmentsare encapsulated in grids which compose the whole cityThus the partition of artificial society becomes the partitionof grids The parallelism degrees are found to support thepartition of models agent independence and grid indepen-dence According to the parallelisms agents are partitioned toLPs according to the population distribution in grids Agentmovements among grids are simulated by the movementevents among grids in different LPs As a result a two-tierparallel architecture is proposed to support the simulation ofartificial society The architecture is detailed in [53]
534 Running of Experiments When the preparation isready the experiments can be done by the central controllerof EMT The running and control commands of the experi-ments are grouped into three aspects start pause or stop theexperiments get the running information of the nodes suchas the snapshot of the desktop of node With the help of thesecommands users can run the computational experimentsto do the research on the emergency management It isworth notifying that the simulation engine is optimized inorder to support the ten million agent simulation UsingGPU as coprocessor a two-tier parallel simulation engine isdesigned with support of MPI and OpenCL through phasedsynchronization [54] One-sided communication is used forreflection of remote simulation objects and message passingbetween processes A general kernel function prototype iselaborately designed and conditionally compiled for execu-tion on both CPU and GPU Moreover the optimizationoperations like load balance is developed under the instruc-tion of activity-based simulation [55] The densely populatedgrids are given more CPU and memory resource accordingto the activity predictions
54 Results of H1N1 Experiments with KD-ACP On thebasis of artificial society models and the intervention mea-sures mentioned in Section 512 a series of computationalexperiments are performed to study the H1N1 influenza inBeijing In our case of artificial Beijing 19610000 agentsare simulated in the cluster within 48 CPU cores and 128Gmemories the nodes are connected by kilomega networks Ittakes 18 hours to simulate a 250-day disease spreading
In order to illustrate how the optimal plans are obtainedbyKD-ACP the experiments are divided into four groups theexperiments ofmodel validation the experiments of sensitiv-ity analysis of vaccination rate the experiments of sensitivityanalysis of isolation and the experiments of the combinationof interventionmeasuresThese four groups show the generalprocess of the research on emergency management in H1N1influenza Firstly the models of H1N1 are validated withthe support of historical data Secondly the traditionalmedical interventions of pandemic such as vaccination areanalyzed to find the optimal plan Thirdly the traditionalnonmedical interventions of pandemic such as isolation areanalyzed to find the optimal plan Finally the combination ofinterventions is tested to show the combined effect
It is worth mentioning that all the experiments in ourwork are performed 100 times These experiments are ini-tialized with different settings of interventions in IMCT Theresults shown in the figures are the mean values
541 The Experiments of Model Validation In our artifi-cial Beijing the models of H1N1 influenza are built basedon the existing researches Wang and her colleagues hadstudied the H1N1 influenza by SEIR (SusceptibleExposedInfectiveRecovered) models [56] With the support of theirwork we build the models of artificial Beijing in the mannerof multiagents
In Figure 23 the control group shown in red line isdrawn based on the realistic statistical data collected in H1N1influenza in Beijing in 2009 The influenza lasts more thansix months and more than 170000 people are infectedComparing with the control group the simulated resultswithout interventions are drawn in blue It is obvious that thesimulated case fits the control group well before 70 days Butthe ascending of control group is slowed after 70 days andthe peak value (17883) is much less than the simulated results(25135) It is because the vaccination is activated at day 67 incontrol group according to the intervention measures usedin 2009 The restrain by vaccination becomes effective eightdays later The change of infected number after vaccinationis shown by red spotted line The trend of control group isdifferent from the simulation data after 75 days
However the simulated results (blue line) in compu-tational experiments fit the control group (as shown inFigure 23) in the first 70 days when no intervention mea-sures are executed The key features such as effective basicreproduction number (119877
0) are also almost the same in both
cases Therefore it can be concluded that the models ofartificial society built in microview are validated by therealistic statistical data in influenza in Beijing in 2009
In this group of experiments no interventions are exe-cuted in order to validate the models Therefore the simu-lated results of this extreme case which cannot happen todaycould be obtained As shown by blue line the number ofinfected agents without interventions reaches peak value atday 91 with 25135 consisting of 8052 in symptomatic phaseand 28950 in recovery phase It is obvious that the totalnumber of infected agents grows slowly at the beginningtime Then the number increases quickly when the influenzabreaks and it reaches its maximum value with 383870 in theend This phenomenon can be explained as in the followingAt the beginning of influenza due to the limited numberof infected agents the propagation of the disease remainsslow but many agents become susceptible When the numberof infected agents increases rapidly the spreading becomessignificant However the number of infected agents decreasesafter day 91 It is because the disease spreading relies on thesocial networks and spatial contact networks [49] It meansthat the infected agents can only infect agents either in thesocial relationship networks or in the spatially contacted
542 The Experiments of Sensitivity Analysis of VaccinationRate According to Section 512 vaccination is a typical
22 Mathematical Problems in Engineering
0
Infe
cted
agen
ts
0 50 100 150 200 250
Statistical dataSimulation data
25000
20000
15000
10000
5000
Day d
Figure 23 The results with no intervention executed and the com-parison with realistic statistical data in Beijing in 2009
medical intervention in influenza Traditionally the effective-ness of vaccination can only be evaluated after the influenzaand the key settings such as vaccination rate are determinedby the experience It is always difficult for emergency man-agement organizations to find the most optimal vaccinationrate in advanceTherefore the sensitivity of vaccination rate isanalyzed in this group of experiments The settings are listedin Table 7 average results are shown in Figure 24
Vaccination rates are set linearly (10 30 50 and70) to test the changes of influenza The peak value ofinfected agents decreases from 25135 22392 15787 6907to 638 respectively It shows that the more the agents arevaccinated the smaller the peak value of the infected agentswill be In the case of 70 vaccination the total infected agentis only 13799 It means that the influenza in this case does nothappen due to the high vaccination rate
The change of peak values also shows an interestingphenomenon Vaccination rates increase linearly the peakvalues decrease slower than the ascending of vaccination ratesat beginning but the descending becomes more and morefaster with the linear increase of vaccination rate It meansthat the small scale of vaccination (less than 30) does notmake senseWith the increase of vaccination rate the numberof susceptible agents decreases significantly In sequencethe possibility of infections is lowered greatly Additionallyit could be found that the decrease of peak values comeslater than the increase of the vaccination rates This is quitereasonable according to Mei et alrsquos work in [57]
As a result it is concluded that high vaccination rate iseffective in the control of H1N1 influenza though costly
543 The Experiments of Sensitivity Analysis of IsolationAccording to Section 512 isolation is a typical nonmedicalintervention It is the most used intervention measure in
Table 7 The initial settings of experiments of vaccination
Parameters ValueAgent count 19610000Initial infected count 40Interventions activated time(119879Intervention)
0
Vaccination rate (119877Vaccination) 0 10 30 50 70
0
Infe
cted
agen
ts
Loading intervention of vaccination
No vaccination10 vaccination30 vaccination
50 vaccination70 vaccination
0 50 100 150 200 250
25000
20000
15000
10000
5000
Day d
Figure 24 The results of loading intervention of vaccination
the emergency management in influenza As mentionedbefore a problem also exists that the measure cannot bevalidated before the intervention Therefore the sensitivityof isolation is analyzed in this group of experiments It isworth mentioning that the isolation is usually combinedwith the hospital policy It means the infected individual issent to hospital meanwhile the people who contact withthe infected are all isolated As a result parameters delay tohospital and isolation duration are designed together in thesettings of experiment Delay to hospital stands for the periodfrom infection time to hospital time while Isolation durationrepresents the period when an agent cannot contact others ifhe has contact with an infected agent The settings of theseparameters are listed in Table 8 average results are shown inFigure 25
Firstly the parameter delay to hospital is simulated aloneto test the sensitivity The simulated results are shown insolid line in Figure 25 The peak value decreases from 6708to 2218 when the delay to hospital decreases from 15 daysto 1 day The half day change brings the 67 reductionMoreover the peak value almost approximates to zero inthe case of 05 days The influenza does not happen in thiscase It can be concluded that the earlier the infected agents
Mathematical Problems in Engineering 23
Table 8 The initial settings of experiments of isolation
Parameters ValueAgent count 19610000Initial infected count 40Interventions activated time (119879Intervention) 80Delay to hospital (119879HDelay) isolation duration (119879IDuration) (15 0) (1 0) (05 0) (15 3) (1 3) (05 3)
0 50 100 150 200 2500
1000
2000
3000
4000
5000
6000
7000
Infe
cted
agen
ts
Loading intervention of isolation
Delay to hospital = 15
Delay to hospital = 10
Delay to hospital = 05
Delay to hospital = 15 isolation duration = 3
Delay to hospital = 10 isolation duration = 3
Delay to hospital = 05 isolation duration = 3
Day d
Figure 25 The results of loading intervention of isolation
are sent to hospital the less the agents will be infectedSo the construction of effective mechanism to find infectedindividuals is important in the emergency management ofinfluenza
Secondly the parameter isolation duration is added tosimulate the isolation of interventionObviously the isolationmeasure decreases the infected agents greatly Compared tothe parameter delay to hospital of 15-day case the peak valuedecreases from 6708 to 362when the isolation is added in theinterventionThe reduction of 6346 agents comes from the 3-day isolation of agents who had contacted with the infectedagents However the decrease of parameter delay to hospitalcannot bring the similar reduction in the 3-day isolationcases According to the figure the peak value decreases from362 to 51 from 15-day case to 05-day case only 311 agents aresaved from infection As a result the sensitivity of simulatedresults does not change obviously with delay to hospital inthe case of isolation It can be concluded that isolation is amore effective measure Though it is not easy for emergencymanagement organizations to decrease the delay to hospitaldue to the current monitoring mechanism the influenza canstill be controlled by isolation
544 The Experiments of Combination of Intervention Mea-sures Isolation is not the only nonmedical intervention ininfluenza close workspace is another common measureWhen close workspace is activated the workspaces are allclosed agents can only stay home at the closed time Luckilyisolation and close workspace are independent from oneanother It is possible for emergency management organi-zations to use these two measures together So this groupof experiments is designed to obtain the effectiveness of thecombination of intervention measures The parameters areinitialized with different settings listed in Table 9
Parameters delay to hospital and isolation duration is setin the middle values (1 day 3 days) according to the lastsection According to the experience close duration is set asseven days It is known that close workspace is a really costintervention measure the activated time is a key parameterWith the fixed parameters mentioned before the changeof interventions activated time is analyzed in this group ofcomputational experimentsThe average results are shown inFigure 26
Compared with the simulated results of experimentsdiscussed above Figure 26 strongly illustrates that influenzais ceased when the intervention measures are combinedtogether The peak value of infected agents is 141 in the ldquo80thdays of close-workspace and isolation with 1 day delay tohospitalrdquo case while 25135 in ldquono interventionrdquo case 2218 inldquo1 day delay to hospitalrdquo case 203 in ldquo1 day delay to hospitaland isolationrdquo case The interventions bring a great reductionof infection the peak values decrease from ten thousandslevel to hundred level
On the other hand the peak value even decreases to 108when the execution time of close workspace is 10 days earlierThe date of maximum value of infected agents is advancedfrom the 135th day to the 108th day The whole circle of H1N1influenza is reducedThe results of this group of experimentsshow that intervention measures should be activated as soonas possible and the compositions of interventions are moreeffective in emergency management of H1N1 influenza
In summary based on the general process lots of exper-iments could be done by KD-ACP to study emergency man-agement in H1N1 influenza Many conclusions are obtainedbased on the analysis of the experiments in this sectionAccording to these conclusions themost optimal response inour case study is the case of ldquo70th days of close-workspace andisolation with 1 day delay to hospitalrdquo It is because this casehas the fewest infected agents relatively But the modelling ofH1N1 influenza is not sufficient Economic and social factorsof interventions such as the money cost by vaccination andisolation the influence of the close-workspace in society
24 Mathematical Problems in Engineering
Table 9 The initial settings of experiments of isolation and closeworkspace
Parameters ValueAgent count 19610000Initial infected count 40Interventions activated time (119879Intervention) 70th day 80th dayDelay to hospital (119879HDelay) 1 dayIsolation duration (119879IDuration) 3 daysClose workspace duration (119879CWDuration) 7 days
50 100 150 200 2500
50
100
150
200
Infe
cted
agen
ts
Loading intervention of isolation and close workspace
Delay to hospital = 1 isolation duration = 3Start close time = 70 close duration time = 7Delay to hospital = 1 isolation duration = 3Start close time = 80 close duration time = 7
Day d
Figure 26The results of loading interventions of isolation and closeworkspace
and even opinion transmission of H1N1 in internet are notconsidered in our case study Without the analysis of thesefactors it does not make sense to talk about the most optimalintervention measures However KD-ACP had provided asoftware framework and general process of the research onemergency management Supported by KD-ACP the expertsof emergency management are able to study emergent eventssuch as H1N1 influenza as detailed as possible Based on theexperiments and analysis again and again the most optimalintervention measures will be found ultimately
55 Parallel Execution by ASST and OsdRT As discussed inSection 35 ASST andOsdRT are used to support the parallelexecution of artificial Beijing The statistical data is collectedand stored in artificial society runtime database at runtimeASST downloads the infection data from the database anddisplays it in the map As shown in Figure 27 the infectedagents are distributed in the districts in BeijingThe snapshotgives the geospatial distribution of the H1N1 epidemic Thelegends of the agents in different status are listed in the left
right corner in Figure 27 Five statuses are shown in the figurehealth without vaccination health with vaccination incuba-tion being symptomatic and convalescence ASST providesa very direct feeling for the emergency experts Partly withthe situation from ASST Emergency Decision Organizationmakes the emergency response plans to the response of theemergency scenario With the help of plans the epidemicin actual society can be restrained In the meantime theinfluenza information is reflected in the Internet Accordingto Figure 8 the data acquisition data extraction and datastandardization of H1N1 influenza are processed in OsdRTOsdRT is implemented by Dr Caorsquos team from Institute ofAutomation from Chinese Academy of Sciences [58] Thetool provides not only the reports of H1N1 SARS hand-foot-mouth disease and intestinal diseases are all monitored Theweb page is shown in Figure 28 H1N1 reports from Internetare collected first Based on the data extraction of the reportsH1N1 case of distribution is obtained and shown in theincidencemap As a result H1N1 incidence data is regarded asthe knowledge which is sent to EECT and IMCTThe settingsof emergent events and intervention measures are modifiedin order to make artificial society approach actual society innext turn of computational experiments With the help ofOsdRT the loop of parallel execution process integrates ACPapproach
6 Conclusion
This work provides an integrated software framework forsocial Computing in emergencymanagement From a systemperspective KD-ACP provides a reliable flexible low costand effective platform for the scientists to do the research onemergency response problems The analysis and predictionof these problems with inherent complexity can be solvedby the repetition of possible alternative experiments onthe software framework Therefore KD-ACP which is anattempt to implement ACP approach seemed as an effectivecomputational framework to support the decision making inemergency response
Currently KD-ACP is used to study the H1N1 epidemicin Beijing in 2009 The most optimal compositions of inter-vention measures are testified to support the response tonext influenza However it is still a long way ahead beforethe automation systematization and practicalization of KD-ACP A lot of work will be carried out along several directionsas in the following
(i) The parallel execution process in KD-ACP is actuallynot strongly connected with the emergency responseorganization Traditional experience and theory stillplay a dominant role in the decision of emergencymanagement It is necessary to propose a mechanismof hall for workshop to facilitate the parallel executionloop in actual society
(ii) Section 33 introduces the principle and process ofmodelling artificial society Metamodels and modelsof artificial Beijing are built by GME Model transfor-mation and code generation are used to implementcodes from FSM based models But many details
Mathematical Problems in Engineering 25
Fangshan district Daxing district Fengtai district Xicheng district
Mentougou district Haidian districtShijingshan district Dongcheng district
Incubation
Symptomatic
Convalescence
Health withoutvaccination
Health withvaccinationLegend of agent
health status
Figure 27 The situation of artificial Beijing by ASST
H1N1
Districts of Beijing
H1N1 case ofdistribution
Incidence map
Distribution ofH1N1 reportsfrom internet
Incidence of H1N1
Fangshan districtLiangxiang town
Population 19000Incidence 135789
(1100000)
Figure 28 The implementation of OsdRT by Institute of Automation from Chinese Academy of Sciences
are still not illustrated due to the topic of thispaper The key techniques such as template of codeframework and model transformations from specificdomains to FSM will be focused and introduced indetail in our next paper
(iii) The performance of computational experiments inKD-ACP is still slower than expectation The opti-mization of simulation engine and models structureis necessary to improve the performance So it ispossible to run artificial society with the real worldin parallel
In summary KD-ACP paves a new way for scientists inboth emergencymanagement and computation simulation to
collaborate with each other in solving the emergency man-agement problems by social computing
Research Highlights
(i) KD-ACP is implemented based on ACP approach
(ii) Artificial Beijing is built with the help of KD-ACP
(iii) H1N1 is modeled to simulate the influenza in Beijingin 2009
(iv) Computational experiments testify the effectivenessof intervention measures
26 Mathematical Problems in Engineering
Conflict of Interests
The authors declare that there is no conflict of interestsregarding the publication of this paper
Acknowledgment
This work was supported in part by the National ScienceFoundation of China (nos 91024030 71303252 61403402and 91324013)
References
[1] W Mao and F-Y Wang Advances in Intelligence and SecurityInformatics Academic Press Oxford UK 2012
[2] E Bonabeau ldquoAgent-based modeling methods and techniquesfor simulating human systemsrdquo Proceedings of the NationalAcademy of Sciences of the United States of America vol 99 no3 pp 7280ndash7287 2002
[3] K M Carley D B Fridsma E Casman et al ldquoBioWar scalableagent-basedmodel of bioattacksrdquo IEEETransactions on SystemsMan and Cybernetics A Systems and Humans vol 36 no 2 pp252ndash265 2006
[4] J M Epstein ldquoModelling to contain pandemicsrdquo Nature vol460 no 7256 p 687 2009
[5] Los Alamos National Laboratory Reports about EPISIMS LosAlamos National Laboratory 2003
[6] C H Builder and S C Bankes ldquoArtificial societies a conceptfor basic research on the societal impacts of informationtechnologyrdquo RAND Report 149 RAND Corp 1991
[7] F-Y Wang ldquoComputational theory and method on complexsystemrdquo China Basic Science vol 6 no 5 pp 3ndash10 2004
[8] F-Y Wang ldquoToward a paradigm shift in social computing theACP approachrdquo IEEE Intelligent Systems vol 22 no 5 pp 65ndash67 2007
[9] F-Y Wang and S Tang ldquoArtificial societies for integratedand sustainable development of metropolitan systemsrdquo IEEEIntelligent Systems vol 19 no 4 pp 82ndash87 2004
[10] D Wen Y Yuan and X-R Li ldquoArtificial societies computa-tional experiments and parallel systems an investigation ona computational theory for complex socioeconomic systemsrdquoIEEE Transactions on Services Computing vol 6 no 2 pp 177ndash185 2013
[11] F Wang F Zeng and Y Yuan ldquoAn ACP-based approach forcomplexity analysis of E-commerce systemrdquo Complex Systemsand Complexity Science vol 5 no 3 pp 1ndash8 2008 (Chinese)
[12] F-Y Wang and P K Wong ldquoResearch commentary intelligentsystems and technology for integrative and predictivemedicinean ACP approachrdquoACMTransactions on Intelligent Systems andTechnology vol 4 no 2 pp 1ndash6 2013
[13] J Sifeng X Gang F Dong and H Chunpeng ldquoStudy on theemergency rescue decision support system of petrochemicalplant based on ACP theoryrdquo in Proceedings of the 6th Manage-ment Annual Meeting (MAM rsquo11) 2011 (Chinese)
[14] X Dong D Fan G Xiong F Zhu Z Zhang and Y Yao ldquoPar-allel Bus Rapid Transit (BRT) operation management systembased on ACP approachrdquo in Proceedings of the 6th ManagementAnnual Meeting (MAM rsquo11) Beijing China 2011
[15] B Ning F-Y Wang H-R Dong R-M Li D Wen and L LildquoParallel systems for Urban rail transportation based on ACP
approachrdquo Journal of Transportation Systems Engineering andInformation Technology vol 10 no 6 pp 23ndash28 2010 (Chinese)
[16] F-Y Wang J Zhao and S-X Lun ldquoArtificial power systemsfor the operation and management of complex power gridsrdquoSouthern Power System Technology vol 2 no 3 pp 1ndash11 2008(Chinese)
[17] M J North T R Howe N T Collier and J R Vos ldquoRepastsimphony development environmentrdquo in Proceedings of theAgent 2005 Conference on Generative Social Processes Modelsand Mechanisms 2005
[18] M J North T RHowe N T Collier and J R Vos ldquoRepast Sim-phony runtime systemrdquo in Proceedings of the Agent Conferenceon Generative Social Processes Models and Mechanisms 2005
[19] SWARM Development Group Swarm 211 Reference Guide2000 httpwwwswarmorg
[20] S Tisue and U Wilensky ldquoNetLogo a simple environment formodeling complexityrdquo in International Conference on ComplexSystems pp 16ndash21 2004
[21] S Luke C Cioffi-Revilla L Panait K Sullivan and G BalanldquoMASON a multiagent simulation environmentrdquo Simulationvol 81 no 7 pp 517ndash527 2005
[22] M Ichikawa H Tanuma Y Koyama andH Deguchi ldquoSOARSintroduction as a social microscope for simulations of socialinteractions and gamingrdquo in Organizing and Learning throughGaming and Simulation Proceedings of the 38th Conference ofthe International Simulation and Gaming Association (ISAGArsquo07) pp 149ndash158 2007
[23] D S Burke J M Epstein D A T Cummings et al ldquoIndividual-based computational modeling of smallpox epidemic controlstrategiesrdquo Academic Emergency Medicine vol 13 no 11 pp1142ndash1149 2006
[24] F-Y Wang ldquoParallel control and management for intelligenttransportation systems concepts architectures and applica-tionsrdquo IEEE Transactions on Intelligent Transportation Systemsvol 11 no 3 pp 630ndash638 2010
[25] Z Molnar D Balasubramanian and A Ledeczi ldquoAn introduc-tion to the genericmodeling environmentrdquo in Proceedings of theTOOLS Europe 2007 Workshop on Model-Driven DevelopmentTool Implementers Forum Zurich Switzerland 2007
[26] S Kelly and J-P Tolvanen Domain Framework in Domain-Specific Modeling Enabling Full Code Generation JohnWiley ampSons Hoboken NJ USA 2007
[27] B P Zeigler H Praehofer and T G Kim Theory of Modelingand Simulation Integrating Discrete Event and ContinuousComplex Dynamic Systems Academic Press San Diego CalifUSA 2000
[28] G Booch J Rumbaugh and I JacobsonThe Unified ModelingLanguage User Guide Addison-Wesley Reading Mass USA1999
[29] F Wagner R Schmuki T Wagne and P Wolstenholme Mod-eling Software with Finite State Machines A Practical ApproachCRC amp Taylor amp Francis Boca Raton Fla USA 2006
[30] T Espiner ldquoChina builds worldrsquos fastest supercomputerrdquo ZDNetUK October 2010
[31] B Chen and G Guo ldquoA two-tier parallel architecture for arti-ficial society simulationrdquo in Proceedings of the ACMIEEESCS26th Workshop on Principles of Advanced and DistributedSimulation (PADS rsquo12) pp 184ndash186 July 2012
[32] G Guo B Chen X G Qiu and Z Li ldquoParallel simulation oflarge-scale artificial society on CPUGPU mixed architecturerdquo
Mathematical Problems in Engineering 27
in Proceedings of the ACMIEEESCS 26th Workshop on Prin-ciples of Advanced and Distributed Simulation (PADS rsquo12) pp174ndash177 July 2012
[33] S-X Lun ldquoResearch on the classification of parallel executionmodes of ACP theoryrdquo Acta Automatica Sinica vol 38 no 10pp 1602ndash1608 2012
[34] F-YWang ldquoA framework for social signal processing and anal-ysis from social sensing networks to computational dialecticalanalyticsrdquo Science China vol 43 no 12 pp 1598ndash1611 2013
[35] G GuoOne Model Userrsquos Manual version 12 National Univer-sity of Defense University 2011
[36] Y Ge R Meng Z Cao X Qiu and K Huang ldquoVirtual citymdashanindividual-based digital environment for human mobility andinteractive behaviorrdquo Simulation Transactions of the Society forthe Modeling and Simulation International 2014
[37] M Kretzschmar and R T Mikolajczyk ldquoContact profiles ineight European countries and implications for modelling thespread of airborne infectious diseasesrdquo PLoS ONE vol 4 no6 Article ID e5931 2009
[38] W Duan and X Qiu ldquoFostering artificial societies using sociallearning and social control in parallel emergency managementsystemsrdquo Frontiers in Computer Science vol 6 no 5 pp 604ndash610 2012
[39] W J Edmunds C J OrsquoCallaghan and D J Nokes ldquoWhomixes with whom Amethod to determine the contact patternsof adults that may lead to the spread of airborne infectionsrdquoProceedings of the Royal Society B Biological Sciences vol 264no 1384 pp 949ndash957 1997
[40] J Banks J S Carson B L Neison and D M Nicol Discrete-Event System Simulation Prentice Hall Englewood Cliffs NJUSA 4th edition 2007
[41] H Yongxia ldquoInvestigation and analysis of college studentsrsquo workand restrdquo Value Engineering vol 31 no 3 2012 (Chinese)
[42] W Duan Z Cao Y Wang et al ldquoAn ACP approach to publichealth emergency management using a campus outbreak ofh1n1 influenza as a case studyrdquo IEEE Transactions on SystemsMan and Cybernetics Part A Systems and Humans vol 43 no5 pp 1028ndash1041 2013
[43] HWHethcote ldquoThemathematics of infectious diseasesrdquo SIAMReview vol 42 no 4 pp 599ndash653 2000
[44] G Chowell E Shim F Brauer and et al ldquoModelling thetransmission dynamics of acute haemorrhagic conjunctivitisapplication to the 2003 outbreak in Mexicordquo Statistics inMedicine vol 25 no 11 pp 1840ndash1857 2006
[45] Y Ge L Liu B Chen X Qiu and K Huang ldquoAgent-basedmodeling for InfluenzaH1N1 in an artificial classroomrdquo SystemsEngineering Procedia vol 2 pp 94ndash104 2011
[46] A R Tuite A L Greer MWhelan et al ldquoEstimated epidemio-logic parameters andmorbidity associatedwith pandemicH1N1influenzardquo CMAJ vol 182 no 2 pp 131ndash136 2010
[47] L Brouwers B Cakici M Camitz A Tegnell and MBoman ldquoEconomic consequences to society of pandemic H1N1influenza 2009mdashpreliminary results for Swedenrdquo Euro Surveil-lance vol 14 no 37 pp 1ndash7 2009
[48] Y Ge L Liu X Qiu H Song Y Wang and K Huang ldquoAframework of multilayer social networks for communicationbehavior with agent-based modelingrdquo Simulation Transactionsof the Society for the Modeling and Simulation International vol89 no 7 pp 810ndash828 2013
[49] W Duan X Qiu Z Cao X Zheng and K Cui ldquoHeterogeneousand stochastic agent-based models for analyzing infectious
diseasesrsquo super spreadersrdquo IEEE Intelligent Systems vol 28 no4 pp 18ndash25 2013
[50] The Preparation and Emergency Response Plans to PandemicPublished by Ministry of Health Ministry of Health 2005(Chinese)
[51] R M Fujimoto Parallel and Distributed Simulation SystemsWiley Interscience 2000
[52] F Cicirelli A Giordano and L Nigro ldquoDistributed simulationof situated multi-agent systemsrdquo in Proceedings of the 15thIEEEACM International Symposium on Distributed Simulationand Real Time Applications (DS-RT rsquo11) pp 28ndash35 September2011
[53] B Chen and G Guo ldquoA two-tier parallel architecturefor artificial society simulationrdquo in Proceedings of the 26thACMIEEESCS Workshop on Principles of Advanced and Dis-tributed Simulation (PADS rsquo12) pp 184ndash186 July 2012
[54] G Guo B Chen and X Qiu ldquoParallel simulation of large-scale artificial society with GPU as coprocessorrdquo InternationalJournal of Modeling Simulation and Scientific Computing vol4 no 2 Article ID 1350005 2013
[55] B Chen L-B Zhang X-C Liu and H Vangheluwe ldquoActivity-based simulation using DEVS increasing performance by anactivity model in parallel DEVS simulationrdquo Journal of ZhejiangUniversity Science C vol 15 no 1 pp 13ndash30 2014
[56] X-L Wang P Yang Z-D Cao et al ldquoQuantitative evaluationon the effectiveness of prevention and control measures againstpandemic influenza A (H1N1) in Beijing 2009rdquoChinese Journalof Epidemiology vol 31 no 12 pp 1374ndash1378 2010 (Chinese)
[57] S Mei A D van de Vijver L Xuan Y Zhu and P M ASloot ldquoQuantitatively evaluating interventions in the influenzaA (H1N1) epidemic on China campus grounded on individual-based simulationsrdquo Procedia Computer Science vol 1 no 1 pp1675ndash1682 2010
[58] X Qiu Z Cao and Z Li Mid-Term Reports of China NationalScience Foundation 91024030 2014
Submit your manuscripts athttpwwwhindawicom
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
MathematicsJournal of
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Mathematical Problems in Engineering
Hindawi Publishing Corporationhttpwwwhindawicom
Differential EquationsInternational Journal of
Volume 2014
Applied MathematicsJournal of
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Probability and StatisticsHindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Journal of
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Mathematical PhysicsAdvances in
Complex AnalysisJournal of
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
OptimizationJournal of
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
CombinatoricsHindawi Publishing Corporationhttpwwwhindawicom Volume 2014
International Journal of
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Operations ResearchAdvances in
Journal of
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Function Spaces
Abstract and Applied AnalysisHindawi Publishing Corporationhttpwwwhindawicom Volume 2014
International Journal of Mathematics and Mathematical Sciences
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
The Scientific World JournalHindawi Publishing Corporation httpwwwhindawicom Volume 2014
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Algebra
Discrete Dynamics in Nature and Society
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Decision SciencesAdvances in
Discrete MathematicsJournal of
Hindawi Publishing Corporationhttpwwwhindawicom
Volume 2014 Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Stochastic AnalysisInternational Journal of
20 Mathematical Problems in Engineering
Streets of close workspaces
Limitations ofactivities
IsolationsRespirator
Closeworkspace
Antibiotic
VaccinationHospital
Surveillance
Figure 21 The settings of intervention measures in IMCT
Information ofexperiment plan
List of experiment plans
Start | Pause | Reset | Stop | Speedup | Slowdown | Back
NodeName | CPU occupation | Memory occupation |Network | Models | Status
Figure 22 The deployment and running of experiments in EMT
532 Load Experiment Plan As discussed in Section 35 thedescription of experiment plans is composed of the infor-mation in three aspects models and the initial data for theexperiments the deployment mapping tables from the mod-els to the nodes and the plan of experiment executionWhenthe experiment plan is loaded by the EMT the file namesand model descriptions of models and data are listed Withthe support of cluster configuration models can be mappedto the nodes in the GUI according to the computationrequirements EMT also provides the setting of experimentsitself including the running times start time and end time
533 Models and Data Deployment Models are stored inthe agent population and services repository According tothe settings in experiment plan the customized models aredownloaded from the repository first then the models suchas agent model environment model emergent events modeland intervention measures model are integrated to be thecompositionmodels of artificial BeijingThemodels and datalocations in the nodes are configured in the ldquomodel pathrdquoand ldquodata pathrdquo in the GUI of central controller Finallythe models are uploaded to the nodes in cluster with theconsideration of partition The partition of artificial society
Mathematical Problems in Engineering 21
models is different from the multiagent system in [52]Models in artificial society are divided into three levels agentsand environments grids and city Agents and environmentsare encapsulated in grids which compose the whole cityThus the partition of artificial society becomes the partitionof grids The parallelism degrees are found to support thepartition of models agent independence and grid indepen-dence According to the parallelisms agents are partitioned toLPs according to the population distribution in grids Agentmovements among grids are simulated by the movementevents among grids in different LPs As a result a two-tierparallel architecture is proposed to support the simulation ofartificial society The architecture is detailed in [53]
534 Running of Experiments When the preparation isready the experiments can be done by the central controllerof EMT The running and control commands of the experi-ments are grouped into three aspects start pause or stop theexperiments get the running information of the nodes suchas the snapshot of the desktop of node With the help of thesecommands users can run the computational experimentsto do the research on the emergency management It isworth notifying that the simulation engine is optimized inorder to support the ten million agent simulation UsingGPU as coprocessor a two-tier parallel simulation engine isdesigned with support of MPI and OpenCL through phasedsynchronization [54] One-sided communication is used forreflection of remote simulation objects and message passingbetween processes A general kernel function prototype iselaborately designed and conditionally compiled for execu-tion on both CPU and GPU Moreover the optimizationoperations like load balance is developed under the instruc-tion of activity-based simulation [55] The densely populatedgrids are given more CPU and memory resource accordingto the activity predictions
54 Results of H1N1 Experiments with KD-ACP On thebasis of artificial society models and the intervention mea-sures mentioned in Section 512 a series of computationalexperiments are performed to study the H1N1 influenza inBeijing In our case of artificial Beijing 19610000 agentsare simulated in the cluster within 48 CPU cores and 128Gmemories the nodes are connected by kilomega networks Ittakes 18 hours to simulate a 250-day disease spreading
In order to illustrate how the optimal plans are obtainedbyKD-ACP the experiments are divided into four groups theexperiments ofmodel validation the experiments of sensitiv-ity analysis of vaccination rate the experiments of sensitivityanalysis of isolation and the experiments of the combinationof interventionmeasuresThese four groups show the generalprocess of the research on emergency management in H1N1influenza Firstly the models of H1N1 are validated withthe support of historical data Secondly the traditionalmedical interventions of pandemic such as vaccination areanalyzed to find the optimal plan Thirdly the traditionalnonmedical interventions of pandemic such as isolation areanalyzed to find the optimal plan Finally the combination ofinterventions is tested to show the combined effect
It is worth mentioning that all the experiments in ourwork are performed 100 times These experiments are ini-tialized with different settings of interventions in IMCT Theresults shown in the figures are the mean values
541 The Experiments of Model Validation In our artifi-cial Beijing the models of H1N1 influenza are built basedon the existing researches Wang and her colleagues hadstudied the H1N1 influenza by SEIR (SusceptibleExposedInfectiveRecovered) models [56] With the support of theirwork we build the models of artificial Beijing in the mannerof multiagents
In Figure 23 the control group shown in red line isdrawn based on the realistic statistical data collected in H1N1influenza in Beijing in 2009 The influenza lasts more thansix months and more than 170000 people are infectedComparing with the control group the simulated resultswithout interventions are drawn in blue It is obvious that thesimulated case fits the control group well before 70 days Butthe ascending of control group is slowed after 70 days andthe peak value (17883) is much less than the simulated results(25135) It is because the vaccination is activated at day 67 incontrol group according to the intervention measures usedin 2009 The restrain by vaccination becomes effective eightdays later The change of infected number after vaccinationis shown by red spotted line The trend of control group isdifferent from the simulation data after 75 days
However the simulated results (blue line) in compu-tational experiments fit the control group (as shown inFigure 23) in the first 70 days when no intervention mea-sures are executed The key features such as effective basicreproduction number (119877
0) are also almost the same in both
cases Therefore it can be concluded that the models ofartificial society built in microview are validated by therealistic statistical data in influenza in Beijing in 2009
In this group of experiments no interventions are exe-cuted in order to validate the models Therefore the simu-lated results of this extreme case which cannot happen todaycould be obtained As shown by blue line the number ofinfected agents without interventions reaches peak value atday 91 with 25135 consisting of 8052 in symptomatic phaseand 28950 in recovery phase It is obvious that the totalnumber of infected agents grows slowly at the beginningtime Then the number increases quickly when the influenzabreaks and it reaches its maximum value with 383870 in theend This phenomenon can be explained as in the followingAt the beginning of influenza due to the limited numberof infected agents the propagation of the disease remainsslow but many agents become susceptible When the numberof infected agents increases rapidly the spreading becomessignificant However the number of infected agents decreasesafter day 91 It is because the disease spreading relies on thesocial networks and spatial contact networks [49] It meansthat the infected agents can only infect agents either in thesocial relationship networks or in the spatially contacted
542 The Experiments of Sensitivity Analysis of VaccinationRate According to Section 512 vaccination is a typical
22 Mathematical Problems in Engineering
0
Infe
cted
agen
ts
0 50 100 150 200 250
Statistical dataSimulation data
25000
20000
15000
10000
5000
Day d
Figure 23 The results with no intervention executed and the com-parison with realistic statistical data in Beijing in 2009
medical intervention in influenza Traditionally the effective-ness of vaccination can only be evaluated after the influenzaand the key settings such as vaccination rate are determinedby the experience It is always difficult for emergency man-agement organizations to find the most optimal vaccinationrate in advanceTherefore the sensitivity of vaccination rate isanalyzed in this group of experiments The settings are listedin Table 7 average results are shown in Figure 24
Vaccination rates are set linearly (10 30 50 and70) to test the changes of influenza The peak value ofinfected agents decreases from 25135 22392 15787 6907to 638 respectively It shows that the more the agents arevaccinated the smaller the peak value of the infected agentswill be In the case of 70 vaccination the total infected agentis only 13799 It means that the influenza in this case does nothappen due to the high vaccination rate
The change of peak values also shows an interestingphenomenon Vaccination rates increase linearly the peakvalues decrease slower than the ascending of vaccination ratesat beginning but the descending becomes more and morefaster with the linear increase of vaccination rate It meansthat the small scale of vaccination (less than 30) does notmake senseWith the increase of vaccination rate the numberof susceptible agents decreases significantly In sequencethe possibility of infections is lowered greatly Additionallyit could be found that the decrease of peak values comeslater than the increase of the vaccination rates This is quitereasonable according to Mei et alrsquos work in [57]
As a result it is concluded that high vaccination rate iseffective in the control of H1N1 influenza though costly
543 The Experiments of Sensitivity Analysis of IsolationAccording to Section 512 isolation is a typical nonmedicalintervention It is the most used intervention measure in
Table 7 The initial settings of experiments of vaccination
Parameters ValueAgent count 19610000Initial infected count 40Interventions activated time(119879Intervention)
0
Vaccination rate (119877Vaccination) 0 10 30 50 70
0
Infe
cted
agen
ts
Loading intervention of vaccination
No vaccination10 vaccination30 vaccination
50 vaccination70 vaccination
0 50 100 150 200 250
25000
20000
15000
10000
5000
Day d
Figure 24 The results of loading intervention of vaccination
the emergency management in influenza As mentionedbefore a problem also exists that the measure cannot bevalidated before the intervention Therefore the sensitivityof isolation is analyzed in this group of experiments It isworth mentioning that the isolation is usually combinedwith the hospital policy It means the infected individual issent to hospital meanwhile the people who contact withthe infected are all isolated As a result parameters delay tohospital and isolation duration are designed together in thesettings of experiment Delay to hospital stands for the periodfrom infection time to hospital time while Isolation durationrepresents the period when an agent cannot contact others ifhe has contact with an infected agent The settings of theseparameters are listed in Table 8 average results are shown inFigure 25
Firstly the parameter delay to hospital is simulated aloneto test the sensitivity The simulated results are shown insolid line in Figure 25 The peak value decreases from 6708to 2218 when the delay to hospital decreases from 15 daysto 1 day The half day change brings the 67 reductionMoreover the peak value almost approximates to zero inthe case of 05 days The influenza does not happen in thiscase It can be concluded that the earlier the infected agents
Mathematical Problems in Engineering 23
Table 8 The initial settings of experiments of isolation
Parameters ValueAgent count 19610000Initial infected count 40Interventions activated time (119879Intervention) 80Delay to hospital (119879HDelay) isolation duration (119879IDuration) (15 0) (1 0) (05 0) (15 3) (1 3) (05 3)
0 50 100 150 200 2500
1000
2000
3000
4000
5000
6000
7000
Infe
cted
agen
ts
Loading intervention of isolation
Delay to hospital = 15
Delay to hospital = 10
Delay to hospital = 05
Delay to hospital = 15 isolation duration = 3
Delay to hospital = 10 isolation duration = 3
Delay to hospital = 05 isolation duration = 3
Day d
Figure 25 The results of loading intervention of isolation
are sent to hospital the less the agents will be infectedSo the construction of effective mechanism to find infectedindividuals is important in the emergency management ofinfluenza
Secondly the parameter isolation duration is added tosimulate the isolation of interventionObviously the isolationmeasure decreases the infected agents greatly Compared tothe parameter delay to hospital of 15-day case the peak valuedecreases from 6708 to 362when the isolation is added in theinterventionThe reduction of 6346 agents comes from the 3-day isolation of agents who had contacted with the infectedagents However the decrease of parameter delay to hospitalcannot bring the similar reduction in the 3-day isolationcases According to the figure the peak value decreases from362 to 51 from 15-day case to 05-day case only 311 agents aresaved from infection As a result the sensitivity of simulatedresults does not change obviously with delay to hospital inthe case of isolation It can be concluded that isolation is amore effective measure Though it is not easy for emergencymanagement organizations to decrease the delay to hospitaldue to the current monitoring mechanism the influenza canstill be controlled by isolation
544 The Experiments of Combination of Intervention Mea-sures Isolation is not the only nonmedical intervention ininfluenza close workspace is another common measureWhen close workspace is activated the workspaces are allclosed agents can only stay home at the closed time Luckilyisolation and close workspace are independent from oneanother It is possible for emergency management organi-zations to use these two measures together So this groupof experiments is designed to obtain the effectiveness of thecombination of intervention measures The parameters areinitialized with different settings listed in Table 9
Parameters delay to hospital and isolation duration is setin the middle values (1 day 3 days) according to the lastsection According to the experience close duration is set asseven days It is known that close workspace is a really costintervention measure the activated time is a key parameterWith the fixed parameters mentioned before the changeof interventions activated time is analyzed in this group ofcomputational experimentsThe average results are shown inFigure 26
Compared with the simulated results of experimentsdiscussed above Figure 26 strongly illustrates that influenzais ceased when the intervention measures are combinedtogether The peak value of infected agents is 141 in the ldquo80thdays of close-workspace and isolation with 1 day delay tohospitalrdquo case while 25135 in ldquono interventionrdquo case 2218 inldquo1 day delay to hospitalrdquo case 203 in ldquo1 day delay to hospitaland isolationrdquo case The interventions bring a great reductionof infection the peak values decrease from ten thousandslevel to hundred level
On the other hand the peak value even decreases to 108when the execution time of close workspace is 10 days earlierThe date of maximum value of infected agents is advancedfrom the 135th day to the 108th day The whole circle of H1N1influenza is reducedThe results of this group of experimentsshow that intervention measures should be activated as soonas possible and the compositions of interventions are moreeffective in emergency management of H1N1 influenza
In summary based on the general process lots of exper-iments could be done by KD-ACP to study emergency man-agement in H1N1 influenza Many conclusions are obtainedbased on the analysis of the experiments in this sectionAccording to these conclusions themost optimal response inour case study is the case of ldquo70th days of close-workspace andisolation with 1 day delay to hospitalrdquo It is because this casehas the fewest infected agents relatively But the modelling ofH1N1 influenza is not sufficient Economic and social factorsof interventions such as the money cost by vaccination andisolation the influence of the close-workspace in society
24 Mathematical Problems in Engineering
Table 9 The initial settings of experiments of isolation and closeworkspace
Parameters ValueAgent count 19610000Initial infected count 40Interventions activated time (119879Intervention) 70th day 80th dayDelay to hospital (119879HDelay) 1 dayIsolation duration (119879IDuration) 3 daysClose workspace duration (119879CWDuration) 7 days
50 100 150 200 2500
50
100
150
200
Infe
cted
agen
ts
Loading intervention of isolation and close workspace
Delay to hospital = 1 isolation duration = 3Start close time = 70 close duration time = 7Delay to hospital = 1 isolation duration = 3Start close time = 80 close duration time = 7
Day d
Figure 26The results of loading interventions of isolation and closeworkspace
and even opinion transmission of H1N1 in internet are notconsidered in our case study Without the analysis of thesefactors it does not make sense to talk about the most optimalintervention measures However KD-ACP had provided asoftware framework and general process of the research onemergency management Supported by KD-ACP the expertsof emergency management are able to study emergent eventssuch as H1N1 influenza as detailed as possible Based on theexperiments and analysis again and again the most optimalintervention measures will be found ultimately
55 Parallel Execution by ASST and OsdRT As discussed inSection 35 ASST andOsdRT are used to support the parallelexecution of artificial Beijing The statistical data is collectedand stored in artificial society runtime database at runtimeASST downloads the infection data from the database anddisplays it in the map As shown in Figure 27 the infectedagents are distributed in the districts in BeijingThe snapshotgives the geospatial distribution of the H1N1 epidemic Thelegends of the agents in different status are listed in the left
right corner in Figure 27 Five statuses are shown in the figurehealth without vaccination health with vaccination incuba-tion being symptomatic and convalescence ASST providesa very direct feeling for the emergency experts Partly withthe situation from ASST Emergency Decision Organizationmakes the emergency response plans to the response of theemergency scenario With the help of plans the epidemicin actual society can be restrained In the meantime theinfluenza information is reflected in the Internet Accordingto Figure 8 the data acquisition data extraction and datastandardization of H1N1 influenza are processed in OsdRTOsdRT is implemented by Dr Caorsquos team from Institute ofAutomation from Chinese Academy of Sciences [58] Thetool provides not only the reports of H1N1 SARS hand-foot-mouth disease and intestinal diseases are all monitored Theweb page is shown in Figure 28 H1N1 reports from Internetare collected first Based on the data extraction of the reportsH1N1 case of distribution is obtained and shown in theincidencemap As a result H1N1 incidence data is regarded asthe knowledge which is sent to EECT and IMCTThe settingsof emergent events and intervention measures are modifiedin order to make artificial society approach actual society innext turn of computational experiments With the help ofOsdRT the loop of parallel execution process integrates ACPapproach
6 Conclusion
This work provides an integrated software framework forsocial Computing in emergencymanagement From a systemperspective KD-ACP provides a reliable flexible low costand effective platform for the scientists to do the research onemergency response problems The analysis and predictionof these problems with inherent complexity can be solvedby the repetition of possible alternative experiments onthe software framework Therefore KD-ACP which is anattempt to implement ACP approach seemed as an effectivecomputational framework to support the decision making inemergency response
Currently KD-ACP is used to study the H1N1 epidemicin Beijing in 2009 The most optimal compositions of inter-vention measures are testified to support the response tonext influenza However it is still a long way ahead beforethe automation systematization and practicalization of KD-ACP A lot of work will be carried out along several directionsas in the following
(i) The parallel execution process in KD-ACP is actuallynot strongly connected with the emergency responseorganization Traditional experience and theory stillplay a dominant role in the decision of emergencymanagement It is necessary to propose a mechanismof hall for workshop to facilitate the parallel executionloop in actual society
(ii) Section 33 introduces the principle and process ofmodelling artificial society Metamodels and modelsof artificial Beijing are built by GME Model transfor-mation and code generation are used to implementcodes from FSM based models But many details
Mathematical Problems in Engineering 25
Fangshan district Daxing district Fengtai district Xicheng district
Mentougou district Haidian districtShijingshan district Dongcheng district
Incubation
Symptomatic
Convalescence
Health withoutvaccination
Health withvaccinationLegend of agent
health status
Figure 27 The situation of artificial Beijing by ASST
H1N1
Districts of Beijing
H1N1 case ofdistribution
Incidence map
Distribution ofH1N1 reportsfrom internet
Incidence of H1N1
Fangshan districtLiangxiang town
Population 19000Incidence 135789
(1100000)
Figure 28 The implementation of OsdRT by Institute of Automation from Chinese Academy of Sciences
are still not illustrated due to the topic of thispaper The key techniques such as template of codeframework and model transformations from specificdomains to FSM will be focused and introduced indetail in our next paper
(iii) The performance of computational experiments inKD-ACP is still slower than expectation The opti-mization of simulation engine and models structureis necessary to improve the performance So it ispossible to run artificial society with the real worldin parallel
In summary KD-ACP paves a new way for scientists inboth emergencymanagement and computation simulation to
collaborate with each other in solving the emergency man-agement problems by social computing
Research Highlights
(i) KD-ACP is implemented based on ACP approach
(ii) Artificial Beijing is built with the help of KD-ACP
(iii) H1N1 is modeled to simulate the influenza in Beijingin 2009
(iv) Computational experiments testify the effectivenessof intervention measures
26 Mathematical Problems in Engineering
Conflict of Interests
The authors declare that there is no conflict of interestsregarding the publication of this paper
Acknowledgment
This work was supported in part by the National ScienceFoundation of China (nos 91024030 71303252 61403402and 91324013)
References
[1] W Mao and F-Y Wang Advances in Intelligence and SecurityInformatics Academic Press Oxford UK 2012
[2] E Bonabeau ldquoAgent-based modeling methods and techniquesfor simulating human systemsrdquo Proceedings of the NationalAcademy of Sciences of the United States of America vol 99 no3 pp 7280ndash7287 2002
[3] K M Carley D B Fridsma E Casman et al ldquoBioWar scalableagent-basedmodel of bioattacksrdquo IEEETransactions on SystemsMan and Cybernetics A Systems and Humans vol 36 no 2 pp252ndash265 2006
[4] J M Epstein ldquoModelling to contain pandemicsrdquo Nature vol460 no 7256 p 687 2009
[5] Los Alamos National Laboratory Reports about EPISIMS LosAlamos National Laboratory 2003
[6] C H Builder and S C Bankes ldquoArtificial societies a conceptfor basic research on the societal impacts of informationtechnologyrdquo RAND Report 149 RAND Corp 1991
[7] F-Y Wang ldquoComputational theory and method on complexsystemrdquo China Basic Science vol 6 no 5 pp 3ndash10 2004
[8] F-Y Wang ldquoToward a paradigm shift in social computing theACP approachrdquo IEEE Intelligent Systems vol 22 no 5 pp 65ndash67 2007
[9] F-Y Wang and S Tang ldquoArtificial societies for integratedand sustainable development of metropolitan systemsrdquo IEEEIntelligent Systems vol 19 no 4 pp 82ndash87 2004
[10] D Wen Y Yuan and X-R Li ldquoArtificial societies computa-tional experiments and parallel systems an investigation ona computational theory for complex socioeconomic systemsrdquoIEEE Transactions on Services Computing vol 6 no 2 pp 177ndash185 2013
[11] F Wang F Zeng and Y Yuan ldquoAn ACP-based approach forcomplexity analysis of E-commerce systemrdquo Complex Systemsand Complexity Science vol 5 no 3 pp 1ndash8 2008 (Chinese)
[12] F-Y Wang and P K Wong ldquoResearch commentary intelligentsystems and technology for integrative and predictivemedicinean ACP approachrdquoACMTransactions on Intelligent Systems andTechnology vol 4 no 2 pp 1ndash6 2013
[13] J Sifeng X Gang F Dong and H Chunpeng ldquoStudy on theemergency rescue decision support system of petrochemicalplant based on ACP theoryrdquo in Proceedings of the 6th Manage-ment Annual Meeting (MAM rsquo11) 2011 (Chinese)
[14] X Dong D Fan G Xiong F Zhu Z Zhang and Y Yao ldquoPar-allel Bus Rapid Transit (BRT) operation management systembased on ACP approachrdquo in Proceedings of the 6th ManagementAnnual Meeting (MAM rsquo11) Beijing China 2011
[15] B Ning F-Y Wang H-R Dong R-M Li D Wen and L LildquoParallel systems for Urban rail transportation based on ACP
approachrdquo Journal of Transportation Systems Engineering andInformation Technology vol 10 no 6 pp 23ndash28 2010 (Chinese)
[16] F-Y Wang J Zhao and S-X Lun ldquoArtificial power systemsfor the operation and management of complex power gridsrdquoSouthern Power System Technology vol 2 no 3 pp 1ndash11 2008(Chinese)
[17] M J North T R Howe N T Collier and J R Vos ldquoRepastsimphony development environmentrdquo in Proceedings of theAgent 2005 Conference on Generative Social Processes Modelsand Mechanisms 2005
[18] M J North T RHowe N T Collier and J R Vos ldquoRepast Sim-phony runtime systemrdquo in Proceedings of the Agent Conferenceon Generative Social Processes Models and Mechanisms 2005
[19] SWARM Development Group Swarm 211 Reference Guide2000 httpwwwswarmorg
[20] S Tisue and U Wilensky ldquoNetLogo a simple environment formodeling complexityrdquo in International Conference on ComplexSystems pp 16ndash21 2004
[21] S Luke C Cioffi-Revilla L Panait K Sullivan and G BalanldquoMASON a multiagent simulation environmentrdquo Simulationvol 81 no 7 pp 517ndash527 2005
[22] M Ichikawa H Tanuma Y Koyama andH Deguchi ldquoSOARSintroduction as a social microscope for simulations of socialinteractions and gamingrdquo in Organizing and Learning throughGaming and Simulation Proceedings of the 38th Conference ofthe International Simulation and Gaming Association (ISAGArsquo07) pp 149ndash158 2007
[23] D S Burke J M Epstein D A T Cummings et al ldquoIndividual-based computational modeling of smallpox epidemic controlstrategiesrdquo Academic Emergency Medicine vol 13 no 11 pp1142ndash1149 2006
[24] F-Y Wang ldquoParallel control and management for intelligenttransportation systems concepts architectures and applica-tionsrdquo IEEE Transactions on Intelligent Transportation Systemsvol 11 no 3 pp 630ndash638 2010
[25] Z Molnar D Balasubramanian and A Ledeczi ldquoAn introduc-tion to the genericmodeling environmentrdquo in Proceedings of theTOOLS Europe 2007 Workshop on Model-Driven DevelopmentTool Implementers Forum Zurich Switzerland 2007
[26] S Kelly and J-P Tolvanen Domain Framework in Domain-Specific Modeling Enabling Full Code Generation JohnWiley ampSons Hoboken NJ USA 2007
[27] B P Zeigler H Praehofer and T G Kim Theory of Modelingand Simulation Integrating Discrete Event and ContinuousComplex Dynamic Systems Academic Press San Diego CalifUSA 2000
[28] G Booch J Rumbaugh and I JacobsonThe Unified ModelingLanguage User Guide Addison-Wesley Reading Mass USA1999
[29] F Wagner R Schmuki T Wagne and P Wolstenholme Mod-eling Software with Finite State Machines A Practical ApproachCRC amp Taylor amp Francis Boca Raton Fla USA 2006
[30] T Espiner ldquoChina builds worldrsquos fastest supercomputerrdquo ZDNetUK October 2010
[31] B Chen and G Guo ldquoA two-tier parallel architecture for arti-ficial society simulationrdquo in Proceedings of the ACMIEEESCS26th Workshop on Principles of Advanced and DistributedSimulation (PADS rsquo12) pp 184ndash186 July 2012
[32] G Guo B Chen X G Qiu and Z Li ldquoParallel simulation oflarge-scale artificial society on CPUGPU mixed architecturerdquo
Mathematical Problems in Engineering 27
in Proceedings of the ACMIEEESCS 26th Workshop on Prin-ciples of Advanced and Distributed Simulation (PADS rsquo12) pp174ndash177 July 2012
[33] S-X Lun ldquoResearch on the classification of parallel executionmodes of ACP theoryrdquo Acta Automatica Sinica vol 38 no 10pp 1602ndash1608 2012
[34] F-YWang ldquoA framework for social signal processing and anal-ysis from social sensing networks to computational dialecticalanalyticsrdquo Science China vol 43 no 12 pp 1598ndash1611 2013
[35] G GuoOne Model Userrsquos Manual version 12 National Univer-sity of Defense University 2011
[36] Y Ge R Meng Z Cao X Qiu and K Huang ldquoVirtual citymdashanindividual-based digital environment for human mobility andinteractive behaviorrdquo Simulation Transactions of the Society forthe Modeling and Simulation International 2014
[37] M Kretzschmar and R T Mikolajczyk ldquoContact profiles ineight European countries and implications for modelling thespread of airborne infectious diseasesrdquo PLoS ONE vol 4 no6 Article ID e5931 2009
[38] W Duan and X Qiu ldquoFostering artificial societies using sociallearning and social control in parallel emergency managementsystemsrdquo Frontiers in Computer Science vol 6 no 5 pp 604ndash610 2012
[39] W J Edmunds C J OrsquoCallaghan and D J Nokes ldquoWhomixes with whom Amethod to determine the contact patternsof adults that may lead to the spread of airborne infectionsrdquoProceedings of the Royal Society B Biological Sciences vol 264no 1384 pp 949ndash957 1997
[40] J Banks J S Carson B L Neison and D M Nicol Discrete-Event System Simulation Prentice Hall Englewood Cliffs NJUSA 4th edition 2007
[41] H Yongxia ldquoInvestigation and analysis of college studentsrsquo workand restrdquo Value Engineering vol 31 no 3 2012 (Chinese)
[42] W Duan Z Cao Y Wang et al ldquoAn ACP approach to publichealth emergency management using a campus outbreak ofh1n1 influenza as a case studyrdquo IEEE Transactions on SystemsMan and Cybernetics Part A Systems and Humans vol 43 no5 pp 1028ndash1041 2013
[43] HWHethcote ldquoThemathematics of infectious diseasesrdquo SIAMReview vol 42 no 4 pp 599ndash653 2000
[44] G Chowell E Shim F Brauer and et al ldquoModelling thetransmission dynamics of acute haemorrhagic conjunctivitisapplication to the 2003 outbreak in Mexicordquo Statistics inMedicine vol 25 no 11 pp 1840ndash1857 2006
[45] Y Ge L Liu B Chen X Qiu and K Huang ldquoAgent-basedmodeling for InfluenzaH1N1 in an artificial classroomrdquo SystemsEngineering Procedia vol 2 pp 94ndash104 2011
[46] A R Tuite A L Greer MWhelan et al ldquoEstimated epidemio-logic parameters andmorbidity associatedwith pandemicH1N1influenzardquo CMAJ vol 182 no 2 pp 131ndash136 2010
[47] L Brouwers B Cakici M Camitz A Tegnell and MBoman ldquoEconomic consequences to society of pandemic H1N1influenza 2009mdashpreliminary results for Swedenrdquo Euro Surveil-lance vol 14 no 37 pp 1ndash7 2009
[48] Y Ge L Liu X Qiu H Song Y Wang and K Huang ldquoAframework of multilayer social networks for communicationbehavior with agent-based modelingrdquo Simulation Transactionsof the Society for the Modeling and Simulation International vol89 no 7 pp 810ndash828 2013
[49] W Duan X Qiu Z Cao X Zheng and K Cui ldquoHeterogeneousand stochastic agent-based models for analyzing infectious
diseasesrsquo super spreadersrdquo IEEE Intelligent Systems vol 28 no4 pp 18ndash25 2013
[50] The Preparation and Emergency Response Plans to PandemicPublished by Ministry of Health Ministry of Health 2005(Chinese)
[51] R M Fujimoto Parallel and Distributed Simulation SystemsWiley Interscience 2000
[52] F Cicirelli A Giordano and L Nigro ldquoDistributed simulationof situated multi-agent systemsrdquo in Proceedings of the 15thIEEEACM International Symposium on Distributed Simulationand Real Time Applications (DS-RT rsquo11) pp 28ndash35 September2011
[53] B Chen and G Guo ldquoA two-tier parallel architecturefor artificial society simulationrdquo in Proceedings of the 26thACMIEEESCS Workshop on Principles of Advanced and Dis-tributed Simulation (PADS rsquo12) pp 184ndash186 July 2012
[54] G Guo B Chen and X Qiu ldquoParallel simulation of large-scale artificial society with GPU as coprocessorrdquo InternationalJournal of Modeling Simulation and Scientific Computing vol4 no 2 Article ID 1350005 2013
[55] B Chen L-B Zhang X-C Liu and H Vangheluwe ldquoActivity-based simulation using DEVS increasing performance by anactivity model in parallel DEVS simulationrdquo Journal of ZhejiangUniversity Science C vol 15 no 1 pp 13ndash30 2014
[56] X-L Wang P Yang Z-D Cao et al ldquoQuantitative evaluationon the effectiveness of prevention and control measures againstpandemic influenza A (H1N1) in Beijing 2009rdquoChinese Journalof Epidemiology vol 31 no 12 pp 1374ndash1378 2010 (Chinese)
[57] S Mei A D van de Vijver L Xuan Y Zhu and P M ASloot ldquoQuantitatively evaluating interventions in the influenzaA (H1N1) epidemic on China campus grounded on individual-based simulationsrdquo Procedia Computer Science vol 1 no 1 pp1675ndash1682 2010
[58] X Qiu Z Cao and Z Li Mid-Term Reports of China NationalScience Foundation 91024030 2014
Submit your manuscripts athttpwwwhindawicom
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
MathematicsJournal of
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Mathematical Problems in Engineering
Hindawi Publishing Corporationhttpwwwhindawicom
Differential EquationsInternational Journal of
Volume 2014
Applied MathematicsJournal of
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Probability and StatisticsHindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Journal of
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Mathematical PhysicsAdvances in
Complex AnalysisJournal of
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
OptimizationJournal of
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
CombinatoricsHindawi Publishing Corporationhttpwwwhindawicom Volume 2014
International Journal of
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Operations ResearchAdvances in
Journal of
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Function Spaces
Abstract and Applied AnalysisHindawi Publishing Corporationhttpwwwhindawicom Volume 2014
International Journal of Mathematics and Mathematical Sciences
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
The Scientific World JournalHindawi Publishing Corporation httpwwwhindawicom Volume 2014
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Algebra
Discrete Dynamics in Nature and Society
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Decision SciencesAdvances in
Discrete MathematicsJournal of
Hindawi Publishing Corporationhttpwwwhindawicom
Volume 2014 Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Stochastic AnalysisInternational Journal of
Mathematical Problems in Engineering 21
models is different from the multiagent system in [52]Models in artificial society are divided into three levels agentsand environments grids and city Agents and environmentsare encapsulated in grids which compose the whole cityThus the partition of artificial society becomes the partitionof grids The parallelism degrees are found to support thepartition of models agent independence and grid indepen-dence According to the parallelisms agents are partitioned toLPs according to the population distribution in grids Agentmovements among grids are simulated by the movementevents among grids in different LPs As a result a two-tierparallel architecture is proposed to support the simulation ofartificial society The architecture is detailed in [53]
534 Running of Experiments When the preparation isready the experiments can be done by the central controllerof EMT The running and control commands of the experi-ments are grouped into three aspects start pause or stop theexperiments get the running information of the nodes suchas the snapshot of the desktop of node With the help of thesecommands users can run the computational experimentsto do the research on the emergency management It isworth notifying that the simulation engine is optimized inorder to support the ten million agent simulation UsingGPU as coprocessor a two-tier parallel simulation engine isdesigned with support of MPI and OpenCL through phasedsynchronization [54] One-sided communication is used forreflection of remote simulation objects and message passingbetween processes A general kernel function prototype iselaborately designed and conditionally compiled for execu-tion on both CPU and GPU Moreover the optimizationoperations like load balance is developed under the instruc-tion of activity-based simulation [55] The densely populatedgrids are given more CPU and memory resource accordingto the activity predictions
54 Results of H1N1 Experiments with KD-ACP On thebasis of artificial society models and the intervention mea-sures mentioned in Section 512 a series of computationalexperiments are performed to study the H1N1 influenza inBeijing In our case of artificial Beijing 19610000 agentsare simulated in the cluster within 48 CPU cores and 128Gmemories the nodes are connected by kilomega networks Ittakes 18 hours to simulate a 250-day disease spreading
In order to illustrate how the optimal plans are obtainedbyKD-ACP the experiments are divided into four groups theexperiments ofmodel validation the experiments of sensitiv-ity analysis of vaccination rate the experiments of sensitivityanalysis of isolation and the experiments of the combinationof interventionmeasuresThese four groups show the generalprocess of the research on emergency management in H1N1influenza Firstly the models of H1N1 are validated withthe support of historical data Secondly the traditionalmedical interventions of pandemic such as vaccination areanalyzed to find the optimal plan Thirdly the traditionalnonmedical interventions of pandemic such as isolation areanalyzed to find the optimal plan Finally the combination ofinterventions is tested to show the combined effect
It is worth mentioning that all the experiments in ourwork are performed 100 times These experiments are ini-tialized with different settings of interventions in IMCT Theresults shown in the figures are the mean values
541 The Experiments of Model Validation In our artifi-cial Beijing the models of H1N1 influenza are built basedon the existing researches Wang and her colleagues hadstudied the H1N1 influenza by SEIR (SusceptibleExposedInfectiveRecovered) models [56] With the support of theirwork we build the models of artificial Beijing in the mannerof multiagents
In Figure 23 the control group shown in red line isdrawn based on the realistic statistical data collected in H1N1influenza in Beijing in 2009 The influenza lasts more thansix months and more than 170000 people are infectedComparing with the control group the simulated resultswithout interventions are drawn in blue It is obvious that thesimulated case fits the control group well before 70 days Butthe ascending of control group is slowed after 70 days andthe peak value (17883) is much less than the simulated results(25135) It is because the vaccination is activated at day 67 incontrol group according to the intervention measures usedin 2009 The restrain by vaccination becomes effective eightdays later The change of infected number after vaccinationis shown by red spotted line The trend of control group isdifferent from the simulation data after 75 days
However the simulated results (blue line) in compu-tational experiments fit the control group (as shown inFigure 23) in the first 70 days when no intervention mea-sures are executed The key features such as effective basicreproduction number (119877
0) are also almost the same in both
cases Therefore it can be concluded that the models ofartificial society built in microview are validated by therealistic statistical data in influenza in Beijing in 2009
In this group of experiments no interventions are exe-cuted in order to validate the models Therefore the simu-lated results of this extreme case which cannot happen todaycould be obtained As shown by blue line the number ofinfected agents without interventions reaches peak value atday 91 with 25135 consisting of 8052 in symptomatic phaseand 28950 in recovery phase It is obvious that the totalnumber of infected agents grows slowly at the beginningtime Then the number increases quickly when the influenzabreaks and it reaches its maximum value with 383870 in theend This phenomenon can be explained as in the followingAt the beginning of influenza due to the limited numberof infected agents the propagation of the disease remainsslow but many agents become susceptible When the numberof infected agents increases rapidly the spreading becomessignificant However the number of infected agents decreasesafter day 91 It is because the disease spreading relies on thesocial networks and spatial contact networks [49] It meansthat the infected agents can only infect agents either in thesocial relationship networks or in the spatially contacted
542 The Experiments of Sensitivity Analysis of VaccinationRate According to Section 512 vaccination is a typical
22 Mathematical Problems in Engineering
0
Infe
cted
agen
ts
0 50 100 150 200 250
Statistical dataSimulation data
25000
20000
15000
10000
5000
Day d
Figure 23 The results with no intervention executed and the com-parison with realistic statistical data in Beijing in 2009
medical intervention in influenza Traditionally the effective-ness of vaccination can only be evaluated after the influenzaand the key settings such as vaccination rate are determinedby the experience It is always difficult for emergency man-agement organizations to find the most optimal vaccinationrate in advanceTherefore the sensitivity of vaccination rate isanalyzed in this group of experiments The settings are listedin Table 7 average results are shown in Figure 24
Vaccination rates are set linearly (10 30 50 and70) to test the changes of influenza The peak value ofinfected agents decreases from 25135 22392 15787 6907to 638 respectively It shows that the more the agents arevaccinated the smaller the peak value of the infected agentswill be In the case of 70 vaccination the total infected agentis only 13799 It means that the influenza in this case does nothappen due to the high vaccination rate
The change of peak values also shows an interestingphenomenon Vaccination rates increase linearly the peakvalues decrease slower than the ascending of vaccination ratesat beginning but the descending becomes more and morefaster with the linear increase of vaccination rate It meansthat the small scale of vaccination (less than 30) does notmake senseWith the increase of vaccination rate the numberof susceptible agents decreases significantly In sequencethe possibility of infections is lowered greatly Additionallyit could be found that the decrease of peak values comeslater than the increase of the vaccination rates This is quitereasonable according to Mei et alrsquos work in [57]
As a result it is concluded that high vaccination rate iseffective in the control of H1N1 influenza though costly
543 The Experiments of Sensitivity Analysis of IsolationAccording to Section 512 isolation is a typical nonmedicalintervention It is the most used intervention measure in
Table 7 The initial settings of experiments of vaccination
Parameters ValueAgent count 19610000Initial infected count 40Interventions activated time(119879Intervention)
0
Vaccination rate (119877Vaccination) 0 10 30 50 70
0
Infe
cted
agen
ts
Loading intervention of vaccination
No vaccination10 vaccination30 vaccination
50 vaccination70 vaccination
0 50 100 150 200 250
25000
20000
15000
10000
5000
Day d
Figure 24 The results of loading intervention of vaccination
the emergency management in influenza As mentionedbefore a problem also exists that the measure cannot bevalidated before the intervention Therefore the sensitivityof isolation is analyzed in this group of experiments It isworth mentioning that the isolation is usually combinedwith the hospital policy It means the infected individual issent to hospital meanwhile the people who contact withthe infected are all isolated As a result parameters delay tohospital and isolation duration are designed together in thesettings of experiment Delay to hospital stands for the periodfrom infection time to hospital time while Isolation durationrepresents the period when an agent cannot contact others ifhe has contact with an infected agent The settings of theseparameters are listed in Table 8 average results are shown inFigure 25
Firstly the parameter delay to hospital is simulated aloneto test the sensitivity The simulated results are shown insolid line in Figure 25 The peak value decreases from 6708to 2218 when the delay to hospital decreases from 15 daysto 1 day The half day change brings the 67 reductionMoreover the peak value almost approximates to zero inthe case of 05 days The influenza does not happen in thiscase It can be concluded that the earlier the infected agents
Mathematical Problems in Engineering 23
Table 8 The initial settings of experiments of isolation
Parameters ValueAgent count 19610000Initial infected count 40Interventions activated time (119879Intervention) 80Delay to hospital (119879HDelay) isolation duration (119879IDuration) (15 0) (1 0) (05 0) (15 3) (1 3) (05 3)
0 50 100 150 200 2500
1000
2000
3000
4000
5000
6000
7000
Infe
cted
agen
ts
Loading intervention of isolation
Delay to hospital = 15
Delay to hospital = 10
Delay to hospital = 05
Delay to hospital = 15 isolation duration = 3
Delay to hospital = 10 isolation duration = 3
Delay to hospital = 05 isolation duration = 3
Day d
Figure 25 The results of loading intervention of isolation
are sent to hospital the less the agents will be infectedSo the construction of effective mechanism to find infectedindividuals is important in the emergency management ofinfluenza
Secondly the parameter isolation duration is added tosimulate the isolation of interventionObviously the isolationmeasure decreases the infected agents greatly Compared tothe parameter delay to hospital of 15-day case the peak valuedecreases from 6708 to 362when the isolation is added in theinterventionThe reduction of 6346 agents comes from the 3-day isolation of agents who had contacted with the infectedagents However the decrease of parameter delay to hospitalcannot bring the similar reduction in the 3-day isolationcases According to the figure the peak value decreases from362 to 51 from 15-day case to 05-day case only 311 agents aresaved from infection As a result the sensitivity of simulatedresults does not change obviously with delay to hospital inthe case of isolation It can be concluded that isolation is amore effective measure Though it is not easy for emergencymanagement organizations to decrease the delay to hospitaldue to the current monitoring mechanism the influenza canstill be controlled by isolation
544 The Experiments of Combination of Intervention Mea-sures Isolation is not the only nonmedical intervention ininfluenza close workspace is another common measureWhen close workspace is activated the workspaces are allclosed agents can only stay home at the closed time Luckilyisolation and close workspace are independent from oneanother It is possible for emergency management organi-zations to use these two measures together So this groupof experiments is designed to obtain the effectiveness of thecombination of intervention measures The parameters areinitialized with different settings listed in Table 9
Parameters delay to hospital and isolation duration is setin the middle values (1 day 3 days) according to the lastsection According to the experience close duration is set asseven days It is known that close workspace is a really costintervention measure the activated time is a key parameterWith the fixed parameters mentioned before the changeof interventions activated time is analyzed in this group ofcomputational experimentsThe average results are shown inFigure 26
Compared with the simulated results of experimentsdiscussed above Figure 26 strongly illustrates that influenzais ceased when the intervention measures are combinedtogether The peak value of infected agents is 141 in the ldquo80thdays of close-workspace and isolation with 1 day delay tohospitalrdquo case while 25135 in ldquono interventionrdquo case 2218 inldquo1 day delay to hospitalrdquo case 203 in ldquo1 day delay to hospitaland isolationrdquo case The interventions bring a great reductionof infection the peak values decrease from ten thousandslevel to hundred level
On the other hand the peak value even decreases to 108when the execution time of close workspace is 10 days earlierThe date of maximum value of infected agents is advancedfrom the 135th day to the 108th day The whole circle of H1N1influenza is reducedThe results of this group of experimentsshow that intervention measures should be activated as soonas possible and the compositions of interventions are moreeffective in emergency management of H1N1 influenza
In summary based on the general process lots of exper-iments could be done by KD-ACP to study emergency man-agement in H1N1 influenza Many conclusions are obtainedbased on the analysis of the experiments in this sectionAccording to these conclusions themost optimal response inour case study is the case of ldquo70th days of close-workspace andisolation with 1 day delay to hospitalrdquo It is because this casehas the fewest infected agents relatively But the modelling ofH1N1 influenza is not sufficient Economic and social factorsof interventions such as the money cost by vaccination andisolation the influence of the close-workspace in society
24 Mathematical Problems in Engineering
Table 9 The initial settings of experiments of isolation and closeworkspace
Parameters ValueAgent count 19610000Initial infected count 40Interventions activated time (119879Intervention) 70th day 80th dayDelay to hospital (119879HDelay) 1 dayIsolation duration (119879IDuration) 3 daysClose workspace duration (119879CWDuration) 7 days
50 100 150 200 2500
50
100
150
200
Infe
cted
agen
ts
Loading intervention of isolation and close workspace
Delay to hospital = 1 isolation duration = 3Start close time = 70 close duration time = 7Delay to hospital = 1 isolation duration = 3Start close time = 80 close duration time = 7
Day d
Figure 26The results of loading interventions of isolation and closeworkspace
and even opinion transmission of H1N1 in internet are notconsidered in our case study Without the analysis of thesefactors it does not make sense to talk about the most optimalintervention measures However KD-ACP had provided asoftware framework and general process of the research onemergency management Supported by KD-ACP the expertsof emergency management are able to study emergent eventssuch as H1N1 influenza as detailed as possible Based on theexperiments and analysis again and again the most optimalintervention measures will be found ultimately
55 Parallel Execution by ASST and OsdRT As discussed inSection 35 ASST andOsdRT are used to support the parallelexecution of artificial Beijing The statistical data is collectedand stored in artificial society runtime database at runtimeASST downloads the infection data from the database anddisplays it in the map As shown in Figure 27 the infectedagents are distributed in the districts in BeijingThe snapshotgives the geospatial distribution of the H1N1 epidemic Thelegends of the agents in different status are listed in the left
right corner in Figure 27 Five statuses are shown in the figurehealth without vaccination health with vaccination incuba-tion being symptomatic and convalescence ASST providesa very direct feeling for the emergency experts Partly withthe situation from ASST Emergency Decision Organizationmakes the emergency response plans to the response of theemergency scenario With the help of plans the epidemicin actual society can be restrained In the meantime theinfluenza information is reflected in the Internet Accordingto Figure 8 the data acquisition data extraction and datastandardization of H1N1 influenza are processed in OsdRTOsdRT is implemented by Dr Caorsquos team from Institute ofAutomation from Chinese Academy of Sciences [58] Thetool provides not only the reports of H1N1 SARS hand-foot-mouth disease and intestinal diseases are all monitored Theweb page is shown in Figure 28 H1N1 reports from Internetare collected first Based on the data extraction of the reportsH1N1 case of distribution is obtained and shown in theincidencemap As a result H1N1 incidence data is regarded asthe knowledge which is sent to EECT and IMCTThe settingsof emergent events and intervention measures are modifiedin order to make artificial society approach actual society innext turn of computational experiments With the help ofOsdRT the loop of parallel execution process integrates ACPapproach
6 Conclusion
This work provides an integrated software framework forsocial Computing in emergencymanagement From a systemperspective KD-ACP provides a reliable flexible low costand effective platform for the scientists to do the research onemergency response problems The analysis and predictionof these problems with inherent complexity can be solvedby the repetition of possible alternative experiments onthe software framework Therefore KD-ACP which is anattempt to implement ACP approach seemed as an effectivecomputational framework to support the decision making inemergency response
Currently KD-ACP is used to study the H1N1 epidemicin Beijing in 2009 The most optimal compositions of inter-vention measures are testified to support the response tonext influenza However it is still a long way ahead beforethe automation systematization and practicalization of KD-ACP A lot of work will be carried out along several directionsas in the following
(i) The parallel execution process in KD-ACP is actuallynot strongly connected with the emergency responseorganization Traditional experience and theory stillplay a dominant role in the decision of emergencymanagement It is necessary to propose a mechanismof hall for workshop to facilitate the parallel executionloop in actual society
(ii) Section 33 introduces the principle and process ofmodelling artificial society Metamodels and modelsof artificial Beijing are built by GME Model transfor-mation and code generation are used to implementcodes from FSM based models But many details
Mathematical Problems in Engineering 25
Fangshan district Daxing district Fengtai district Xicheng district
Mentougou district Haidian districtShijingshan district Dongcheng district
Incubation
Symptomatic
Convalescence
Health withoutvaccination
Health withvaccinationLegend of agent
health status
Figure 27 The situation of artificial Beijing by ASST
H1N1
Districts of Beijing
H1N1 case ofdistribution
Incidence map
Distribution ofH1N1 reportsfrom internet
Incidence of H1N1
Fangshan districtLiangxiang town
Population 19000Incidence 135789
(1100000)
Figure 28 The implementation of OsdRT by Institute of Automation from Chinese Academy of Sciences
are still not illustrated due to the topic of thispaper The key techniques such as template of codeframework and model transformations from specificdomains to FSM will be focused and introduced indetail in our next paper
(iii) The performance of computational experiments inKD-ACP is still slower than expectation The opti-mization of simulation engine and models structureis necessary to improve the performance So it ispossible to run artificial society with the real worldin parallel
In summary KD-ACP paves a new way for scientists inboth emergencymanagement and computation simulation to
collaborate with each other in solving the emergency man-agement problems by social computing
Research Highlights
(i) KD-ACP is implemented based on ACP approach
(ii) Artificial Beijing is built with the help of KD-ACP
(iii) H1N1 is modeled to simulate the influenza in Beijingin 2009
(iv) Computational experiments testify the effectivenessof intervention measures
26 Mathematical Problems in Engineering
Conflict of Interests
The authors declare that there is no conflict of interestsregarding the publication of this paper
Acknowledgment
This work was supported in part by the National ScienceFoundation of China (nos 91024030 71303252 61403402and 91324013)
References
[1] W Mao and F-Y Wang Advances in Intelligence and SecurityInformatics Academic Press Oxford UK 2012
[2] E Bonabeau ldquoAgent-based modeling methods and techniquesfor simulating human systemsrdquo Proceedings of the NationalAcademy of Sciences of the United States of America vol 99 no3 pp 7280ndash7287 2002
[3] K M Carley D B Fridsma E Casman et al ldquoBioWar scalableagent-basedmodel of bioattacksrdquo IEEETransactions on SystemsMan and Cybernetics A Systems and Humans vol 36 no 2 pp252ndash265 2006
[4] J M Epstein ldquoModelling to contain pandemicsrdquo Nature vol460 no 7256 p 687 2009
[5] Los Alamos National Laboratory Reports about EPISIMS LosAlamos National Laboratory 2003
[6] C H Builder and S C Bankes ldquoArtificial societies a conceptfor basic research on the societal impacts of informationtechnologyrdquo RAND Report 149 RAND Corp 1991
[7] F-Y Wang ldquoComputational theory and method on complexsystemrdquo China Basic Science vol 6 no 5 pp 3ndash10 2004
[8] F-Y Wang ldquoToward a paradigm shift in social computing theACP approachrdquo IEEE Intelligent Systems vol 22 no 5 pp 65ndash67 2007
[9] F-Y Wang and S Tang ldquoArtificial societies for integratedand sustainable development of metropolitan systemsrdquo IEEEIntelligent Systems vol 19 no 4 pp 82ndash87 2004
[10] D Wen Y Yuan and X-R Li ldquoArtificial societies computa-tional experiments and parallel systems an investigation ona computational theory for complex socioeconomic systemsrdquoIEEE Transactions on Services Computing vol 6 no 2 pp 177ndash185 2013
[11] F Wang F Zeng and Y Yuan ldquoAn ACP-based approach forcomplexity analysis of E-commerce systemrdquo Complex Systemsand Complexity Science vol 5 no 3 pp 1ndash8 2008 (Chinese)
[12] F-Y Wang and P K Wong ldquoResearch commentary intelligentsystems and technology for integrative and predictivemedicinean ACP approachrdquoACMTransactions on Intelligent Systems andTechnology vol 4 no 2 pp 1ndash6 2013
[13] J Sifeng X Gang F Dong and H Chunpeng ldquoStudy on theemergency rescue decision support system of petrochemicalplant based on ACP theoryrdquo in Proceedings of the 6th Manage-ment Annual Meeting (MAM rsquo11) 2011 (Chinese)
[14] X Dong D Fan G Xiong F Zhu Z Zhang and Y Yao ldquoPar-allel Bus Rapid Transit (BRT) operation management systembased on ACP approachrdquo in Proceedings of the 6th ManagementAnnual Meeting (MAM rsquo11) Beijing China 2011
[15] B Ning F-Y Wang H-R Dong R-M Li D Wen and L LildquoParallel systems for Urban rail transportation based on ACP
approachrdquo Journal of Transportation Systems Engineering andInformation Technology vol 10 no 6 pp 23ndash28 2010 (Chinese)
[16] F-Y Wang J Zhao and S-X Lun ldquoArtificial power systemsfor the operation and management of complex power gridsrdquoSouthern Power System Technology vol 2 no 3 pp 1ndash11 2008(Chinese)
[17] M J North T R Howe N T Collier and J R Vos ldquoRepastsimphony development environmentrdquo in Proceedings of theAgent 2005 Conference on Generative Social Processes Modelsand Mechanisms 2005
[18] M J North T RHowe N T Collier and J R Vos ldquoRepast Sim-phony runtime systemrdquo in Proceedings of the Agent Conferenceon Generative Social Processes Models and Mechanisms 2005
[19] SWARM Development Group Swarm 211 Reference Guide2000 httpwwwswarmorg
[20] S Tisue and U Wilensky ldquoNetLogo a simple environment formodeling complexityrdquo in International Conference on ComplexSystems pp 16ndash21 2004
[21] S Luke C Cioffi-Revilla L Panait K Sullivan and G BalanldquoMASON a multiagent simulation environmentrdquo Simulationvol 81 no 7 pp 517ndash527 2005
[22] M Ichikawa H Tanuma Y Koyama andH Deguchi ldquoSOARSintroduction as a social microscope for simulations of socialinteractions and gamingrdquo in Organizing and Learning throughGaming and Simulation Proceedings of the 38th Conference ofthe International Simulation and Gaming Association (ISAGArsquo07) pp 149ndash158 2007
[23] D S Burke J M Epstein D A T Cummings et al ldquoIndividual-based computational modeling of smallpox epidemic controlstrategiesrdquo Academic Emergency Medicine vol 13 no 11 pp1142ndash1149 2006
[24] F-Y Wang ldquoParallel control and management for intelligenttransportation systems concepts architectures and applica-tionsrdquo IEEE Transactions on Intelligent Transportation Systemsvol 11 no 3 pp 630ndash638 2010
[25] Z Molnar D Balasubramanian and A Ledeczi ldquoAn introduc-tion to the genericmodeling environmentrdquo in Proceedings of theTOOLS Europe 2007 Workshop on Model-Driven DevelopmentTool Implementers Forum Zurich Switzerland 2007
[26] S Kelly and J-P Tolvanen Domain Framework in Domain-Specific Modeling Enabling Full Code Generation JohnWiley ampSons Hoboken NJ USA 2007
[27] B P Zeigler H Praehofer and T G Kim Theory of Modelingand Simulation Integrating Discrete Event and ContinuousComplex Dynamic Systems Academic Press San Diego CalifUSA 2000
[28] G Booch J Rumbaugh and I JacobsonThe Unified ModelingLanguage User Guide Addison-Wesley Reading Mass USA1999
[29] F Wagner R Schmuki T Wagne and P Wolstenholme Mod-eling Software with Finite State Machines A Practical ApproachCRC amp Taylor amp Francis Boca Raton Fla USA 2006
[30] T Espiner ldquoChina builds worldrsquos fastest supercomputerrdquo ZDNetUK October 2010
[31] B Chen and G Guo ldquoA two-tier parallel architecture for arti-ficial society simulationrdquo in Proceedings of the ACMIEEESCS26th Workshop on Principles of Advanced and DistributedSimulation (PADS rsquo12) pp 184ndash186 July 2012
[32] G Guo B Chen X G Qiu and Z Li ldquoParallel simulation oflarge-scale artificial society on CPUGPU mixed architecturerdquo
Mathematical Problems in Engineering 27
in Proceedings of the ACMIEEESCS 26th Workshop on Prin-ciples of Advanced and Distributed Simulation (PADS rsquo12) pp174ndash177 July 2012
[33] S-X Lun ldquoResearch on the classification of parallel executionmodes of ACP theoryrdquo Acta Automatica Sinica vol 38 no 10pp 1602ndash1608 2012
[34] F-YWang ldquoA framework for social signal processing and anal-ysis from social sensing networks to computational dialecticalanalyticsrdquo Science China vol 43 no 12 pp 1598ndash1611 2013
[35] G GuoOne Model Userrsquos Manual version 12 National Univer-sity of Defense University 2011
[36] Y Ge R Meng Z Cao X Qiu and K Huang ldquoVirtual citymdashanindividual-based digital environment for human mobility andinteractive behaviorrdquo Simulation Transactions of the Society forthe Modeling and Simulation International 2014
[37] M Kretzschmar and R T Mikolajczyk ldquoContact profiles ineight European countries and implications for modelling thespread of airborne infectious diseasesrdquo PLoS ONE vol 4 no6 Article ID e5931 2009
[38] W Duan and X Qiu ldquoFostering artificial societies using sociallearning and social control in parallel emergency managementsystemsrdquo Frontiers in Computer Science vol 6 no 5 pp 604ndash610 2012
[39] W J Edmunds C J OrsquoCallaghan and D J Nokes ldquoWhomixes with whom Amethod to determine the contact patternsof adults that may lead to the spread of airborne infectionsrdquoProceedings of the Royal Society B Biological Sciences vol 264no 1384 pp 949ndash957 1997
[40] J Banks J S Carson B L Neison and D M Nicol Discrete-Event System Simulation Prentice Hall Englewood Cliffs NJUSA 4th edition 2007
[41] H Yongxia ldquoInvestigation and analysis of college studentsrsquo workand restrdquo Value Engineering vol 31 no 3 2012 (Chinese)
[42] W Duan Z Cao Y Wang et al ldquoAn ACP approach to publichealth emergency management using a campus outbreak ofh1n1 influenza as a case studyrdquo IEEE Transactions on SystemsMan and Cybernetics Part A Systems and Humans vol 43 no5 pp 1028ndash1041 2013
[43] HWHethcote ldquoThemathematics of infectious diseasesrdquo SIAMReview vol 42 no 4 pp 599ndash653 2000
[44] G Chowell E Shim F Brauer and et al ldquoModelling thetransmission dynamics of acute haemorrhagic conjunctivitisapplication to the 2003 outbreak in Mexicordquo Statistics inMedicine vol 25 no 11 pp 1840ndash1857 2006
[45] Y Ge L Liu B Chen X Qiu and K Huang ldquoAgent-basedmodeling for InfluenzaH1N1 in an artificial classroomrdquo SystemsEngineering Procedia vol 2 pp 94ndash104 2011
[46] A R Tuite A L Greer MWhelan et al ldquoEstimated epidemio-logic parameters andmorbidity associatedwith pandemicH1N1influenzardquo CMAJ vol 182 no 2 pp 131ndash136 2010
[47] L Brouwers B Cakici M Camitz A Tegnell and MBoman ldquoEconomic consequences to society of pandemic H1N1influenza 2009mdashpreliminary results for Swedenrdquo Euro Surveil-lance vol 14 no 37 pp 1ndash7 2009
[48] Y Ge L Liu X Qiu H Song Y Wang and K Huang ldquoAframework of multilayer social networks for communicationbehavior with agent-based modelingrdquo Simulation Transactionsof the Society for the Modeling and Simulation International vol89 no 7 pp 810ndash828 2013
[49] W Duan X Qiu Z Cao X Zheng and K Cui ldquoHeterogeneousand stochastic agent-based models for analyzing infectious
diseasesrsquo super spreadersrdquo IEEE Intelligent Systems vol 28 no4 pp 18ndash25 2013
[50] The Preparation and Emergency Response Plans to PandemicPublished by Ministry of Health Ministry of Health 2005(Chinese)
[51] R M Fujimoto Parallel and Distributed Simulation SystemsWiley Interscience 2000
[52] F Cicirelli A Giordano and L Nigro ldquoDistributed simulationof situated multi-agent systemsrdquo in Proceedings of the 15thIEEEACM International Symposium on Distributed Simulationand Real Time Applications (DS-RT rsquo11) pp 28ndash35 September2011
[53] B Chen and G Guo ldquoA two-tier parallel architecturefor artificial society simulationrdquo in Proceedings of the 26thACMIEEESCS Workshop on Principles of Advanced and Dis-tributed Simulation (PADS rsquo12) pp 184ndash186 July 2012
[54] G Guo B Chen and X Qiu ldquoParallel simulation of large-scale artificial society with GPU as coprocessorrdquo InternationalJournal of Modeling Simulation and Scientific Computing vol4 no 2 Article ID 1350005 2013
[55] B Chen L-B Zhang X-C Liu and H Vangheluwe ldquoActivity-based simulation using DEVS increasing performance by anactivity model in parallel DEVS simulationrdquo Journal of ZhejiangUniversity Science C vol 15 no 1 pp 13ndash30 2014
[56] X-L Wang P Yang Z-D Cao et al ldquoQuantitative evaluationon the effectiveness of prevention and control measures againstpandemic influenza A (H1N1) in Beijing 2009rdquoChinese Journalof Epidemiology vol 31 no 12 pp 1374ndash1378 2010 (Chinese)
[57] S Mei A D van de Vijver L Xuan Y Zhu and P M ASloot ldquoQuantitatively evaluating interventions in the influenzaA (H1N1) epidemic on China campus grounded on individual-based simulationsrdquo Procedia Computer Science vol 1 no 1 pp1675ndash1682 2010
[58] X Qiu Z Cao and Z Li Mid-Term Reports of China NationalScience Foundation 91024030 2014
Submit your manuscripts athttpwwwhindawicom
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
MathematicsJournal of
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Mathematical Problems in Engineering
Hindawi Publishing Corporationhttpwwwhindawicom
Differential EquationsInternational Journal of
Volume 2014
Applied MathematicsJournal of
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Probability and StatisticsHindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Journal of
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Mathematical PhysicsAdvances in
Complex AnalysisJournal of
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
OptimizationJournal of
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
CombinatoricsHindawi Publishing Corporationhttpwwwhindawicom Volume 2014
International Journal of
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Operations ResearchAdvances in
Journal of
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Function Spaces
Abstract and Applied AnalysisHindawi Publishing Corporationhttpwwwhindawicom Volume 2014
International Journal of Mathematics and Mathematical Sciences
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
The Scientific World JournalHindawi Publishing Corporation httpwwwhindawicom Volume 2014
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Algebra
Discrete Dynamics in Nature and Society
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Decision SciencesAdvances in
Discrete MathematicsJournal of
Hindawi Publishing Corporationhttpwwwhindawicom
Volume 2014 Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Stochastic AnalysisInternational Journal of
22 Mathematical Problems in Engineering
0
Infe
cted
agen
ts
0 50 100 150 200 250
Statistical dataSimulation data
25000
20000
15000
10000
5000
Day d
Figure 23 The results with no intervention executed and the com-parison with realistic statistical data in Beijing in 2009
medical intervention in influenza Traditionally the effective-ness of vaccination can only be evaluated after the influenzaand the key settings such as vaccination rate are determinedby the experience It is always difficult for emergency man-agement organizations to find the most optimal vaccinationrate in advanceTherefore the sensitivity of vaccination rate isanalyzed in this group of experiments The settings are listedin Table 7 average results are shown in Figure 24
Vaccination rates are set linearly (10 30 50 and70) to test the changes of influenza The peak value ofinfected agents decreases from 25135 22392 15787 6907to 638 respectively It shows that the more the agents arevaccinated the smaller the peak value of the infected agentswill be In the case of 70 vaccination the total infected agentis only 13799 It means that the influenza in this case does nothappen due to the high vaccination rate
The change of peak values also shows an interestingphenomenon Vaccination rates increase linearly the peakvalues decrease slower than the ascending of vaccination ratesat beginning but the descending becomes more and morefaster with the linear increase of vaccination rate It meansthat the small scale of vaccination (less than 30) does notmake senseWith the increase of vaccination rate the numberof susceptible agents decreases significantly In sequencethe possibility of infections is lowered greatly Additionallyit could be found that the decrease of peak values comeslater than the increase of the vaccination rates This is quitereasonable according to Mei et alrsquos work in [57]
As a result it is concluded that high vaccination rate iseffective in the control of H1N1 influenza though costly
543 The Experiments of Sensitivity Analysis of IsolationAccording to Section 512 isolation is a typical nonmedicalintervention It is the most used intervention measure in
Table 7 The initial settings of experiments of vaccination
Parameters ValueAgent count 19610000Initial infected count 40Interventions activated time(119879Intervention)
0
Vaccination rate (119877Vaccination) 0 10 30 50 70
0
Infe
cted
agen
ts
Loading intervention of vaccination
No vaccination10 vaccination30 vaccination
50 vaccination70 vaccination
0 50 100 150 200 250
25000
20000
15000
10000
5000
Day d
Figure 24 The results of loading intervention of vaccination
the emergency management in influenza As mentionedbefore a problem also exists that the measure cannot bevalidated before the intervention Therefore the sensitivityof isolation is analyzed in this group of experiments It isworth mentioning that the isolation is usually combinedwith the hospital policy It means the infected individual issent to hospital meanwhile the people who contact withthe infected are all isolated As a result parameters delay tohospital and isolation duration are designed together in thesettings of experiment Delay to hospital stands for the periodfrom infection time to hospital time while Isolation durationrepresents the period when an agent cannot contact others ifhe has contact with an infected agent The settings of theseparameters are listed in Table 8 average results are shown inFigure 25
Firstly the parameter delay to hospital is simulated aloneto test the sensitivity The simulated results are shown insolid line in Figure 25 The peak value decreases from 6708to 2218 when the delay to hospital decreases from 15 daysto 1 day The half day change brings the 67 reductionMoreover the peak value almost approximates to zero inthe case of 05 days The influenza does not happen in thiscase It can be concluded that the earlier the infected agents
Mathematical Problems in Engineering 23
Table 8 The initial settings of experiments of isolation
Parameters ValueAgent count 19610000Initial infected count 40Interventions activated time (119879Intervention) 80Delay to hospital (119879HDelay) isolation duration (119879IDuration) (15 0) (1 0) (05 0) (15 3) (1 3) (05 3)
0 50 100 150 200 2500
1000
2000
3000
4000
5000
6000
7000
Infe
cted
agen
ts
Loading intervention of isolation
Delay to hospital = 15
Delay to hospital = 10
Delay to hospital = 05
Delay to hospital = 15 isolation duration = 3
Delay to hospital = 10 isolation duration = 3
Delay to hospital = 05 isolation duration = 3
Day d
Figure 25 The results of loading intervention of isolation
are sent to hospital the less the agents will be infectedSo the construction of effective mechanism to find infectedindividuals is important in the emergency management ofinfluenza
Secondly the parameter isolation duration is added tosimulate the isolation of interventionObviously the isolationmeasure decreases the infected agents greatly Compared tothe parameter delay to hospital of 15-day case the peak valuedecreases from 6708 to 362when the isolation is added in theinterventionThe reduction of 6346 agents comes from the 3-day isolation of agents who had contacted with the infectedagents However the decrease of parameter delay to hospitalcannot bring the similar reduction in the 3-day isolationcases According to the figure the peak value decreases from362 to 51 from 15-day case to 05-day case only 311 agents aresaved from infection As a result the sensitivity of simulatedresults does not change obviously with delay to hospital inthe case of isolation It can be concluded that isolation is amore effective measure Though it is not easy for emergencymanagement organizations to decrease the delay to hospitaldue to the current monitoring mechanism the influenza canstill be controlled by isolation
544 The Experiments of Combination of Intervention Mea-sures Isolation is not the only nonmedical intervention ininfluenza close workspace is another common measureWhen close workspace is activated the workspaces are allclosed agents can only stay home at the closed time Luckilyisolation and close workspace are independent from oneanother It is possible for emergency management organi-zations to use these two measures together So this groupof experiments is designed to obtain the effectiveness of thecombination of intervention measures The parameters areinitialized with different settings listed in Table 9
Parameters delay to hospital and isolation duration is setin the middle values (1 day 3 days) according to the lastsection According to the experience close duration is set asseven days It is known that close workspace is a really costintervention measure the activated time is a key parameterWith the fixed parameters mentioned before the changeof interventions activated time is analyzed in this group ofcomputational experimentsThe average results are shown inFigure 26
Compared with the simulated results of experimentsdiscussed above Figure 26 strongly illustrates that influenzais ceased when the intervention measures are combinedtogether The peak value of infected agents is 141 in the ldquo80thdays of close-workspace and isolation with 1 day delay tohospitalrdquo case while 25135 in ldquono interventionrdquo case 2218 inldquo1 day delay to hospitalrdquo case 203 in ldquo1 day delay to hospitaland isolationrdquo case The interventions bring a great reductionof infection the peak values decrease from ten thousandslevel to hundred level
On the other hand the peak value even decreases to 108when the execution time of close workspace is 10 days earlierThe date of maximum value of infected agents is advancedfrom the 135th day to the 108th day The whole circle of H1N1influenza is reducedThe results of this group of experimentsshow that intervention measures should be activated as soonas possible and the compositions of interventions are moreeffective in emergency management of H1N1 influenza
In summary based on the general process lots of exper-iments could be done by KD-ACP to study emergency man-agement in H1N1 influenza Many conclusions are obtainedbased on the analysis of the experiments in this sectionAccording to these conclusions themost optimal response inour case study is the case of ldquo70th days of close-workspace andisolation with 1 day delay to hospitalrdquo It is because this casehas the fewest infected agents relatively But the modelling ofH1N1 influenza is not sufficient Economic and social factorsof interventions such as the money cost by vaccination andisolation the influence of the close-workspace in society
24 Mathematical Problems in Engineering
Table 9 The initial settings of experiments of isolation and closeworkspace
Parameters ValueAgent count 19610000Initial infected count 40Interventions activated time (119879Intervention) 70th day 80th dayDelay to hospital (119879HDelay) 1 dayIsolation duration (119879IDuration) 3 daysClose workspace duration (119879CWDuration) 7 days
50 100 150 200 2500
50
100
150
200
Infe
cted
agen
ts
Loading intervention of isolation and close workspace
Delay to hospital = 1 isolation duration = 3Start close time = 70 close duration time = 7Delay to hospital = 1 isolation duration = 3Start close time = 80 close duration time = 7
Day d
Figure 26The results of loading interventions of isolation and closeworkspace
and even opinion transmission of H1N1 in internet are notconsidered in our case study Without the analysis of thesefactors it does not make sense to talk about the most optimalintervention measures However KD-ACP had provided asoftware framework and general process of the research onemergency management Supported by KD-ACP the expertsof emergency management are able to study emergent eventssuch as H1N1 influenza as detailed as possible Based on theexperiments and analysis again and again the most optimalintervention measures will be found ultimately
55 Parallel Execution by ASST and OsdRT As discussed inSection 35 ASST andOsdRT are used to support the parallelexecution of artificial Beijing The statistical data is collectedand stored in artificial society runtime database at runtimeASST downloads the infection data from the database anddisplays it in the map As shown in Figure 27 the infectedagents are distributed in the districts in BeijingThe snapshotgives the geospatial distribution of the H1N1 epidemic Thelegends of the agents in different status are listed in the left
right corner in Figure 27 Five statuses are shown in the figurehealth without vaccination health with vaccination incuba-tion being symptomatic and convalescence ASST providesa very direct feeling for the emergency experts Partly withthe situation from ASST Emergency Decision Organizationmakes the emergency response plans to the response of theemergency scenario With the help of plans the epidemicin actual society can be restrained In the meantime theinfluenza information is reflected in the Internet Accordingto Figure 8 the data acquisition data extraction and datastandardization of H1N1 influenza are processed in OsdRTOsdRT is implemented by Dr Caorsquos team from Institute ofAutomation from Chinese Academy of Sciences [58] Thetool provides not only the reports of H1N1 SARS hand-foot-mouth disease and intestinal diseases are all monitored Theweb page is shown in Figure 28 H1N1 reports from Internetare collected first Based on the data extraction of the reportsH1N1 case of distribution is obtained and shown in theincidencemap As a result H1N1 incidence data is regarded asthe knowledge which is sent to EECT and IMCTThe settingsof emergent events and intervention measures are modifiedin order to make artificial society approach actual society innext turn of computational experiments With the help ofOsdRT the loop of parallel execution process integrates ACPapproach
6 Conclusion
This work provides an integrated software framework forsocial Computing in emergencymanagement From a systemperspective KD-ACP provides a reliable flexible low costand effective platform for the scientists to do the research onemergency response problems The analysis and predictionof these problems with inherent complexity can be solvedby the repetition of possible alternative experiments onthe software framework Therefore KD-ACP which is anattempt to implement ACP approach seemed as an effectivecomputational framework to support the decision making inemergency response
Currently KD-ACP is used to study the H1N1 epidemicin Beijing in 2009 The most optimal compositions of inter-vention measures are testified to support the response tonext influenza However it is still a long way ahead beforethe automation systematization and practicalization of KD-ACP A lot of work will be carried out along several directionsas in the following
(i) The parallel execution process in KD-ACP is actuallynot strongly connected with the emergency responseorganization Traditional experience and theory stillplay a dominant role in the decision of emergencymanagement It is necessary to propose a mechanismof hall for workshop to facilitate the parallel executionloop in actual society
(ii) Section 33 introduces the principle and process ofmodelling artificial society Metamodels and modelsof artificial Beijing are built by GME Model transfor-mation and code generation are used to implementcodes from FSM based models But many details
Mathematical Problems in Engineering 25
Fangshan district Daxing district Fengtai district Xicheng district
Mentougou district Haidian districtShijingshan district Dongcheng district
Incubation
Symptomatic
Convalescence
Health withoutvaccination
Health withvaccinationLegend of agent
health status
Figure 27 The situation of artificial Beijing by ASST
H1N1
Districts of Beijing
H1N1 case ofdistribution
Incidence map
Distribution ofH1N1 reportsfrom internet
Incidence of H1N1
Fangshan districtLiangxiang town
Population 19000Incidence 135789
(1100000)
Figure 28 The implementation of OsdRT by Institute of Automation from Chinese Academy of Sciences
are still not illustrated due to the topic of thispaper The key techniques such as template of codeframework and model transformations from specificdomains to FSM will be focused and introduced indetail in our next paper
(iii) The performance of computational experiments inKD-ACP is still slower than expectation The opti-mization of simulation engine and models structureis necessary to improve the performance So it ispossible to run artificial society with the real worldin parallel
In summary KD-ACP paves a new way for scientists inboth emergencymanagement and computation simulation to
collaborate with each other in solving the emergency man-agement problems by social computing
Research Highlights
(i) KD-ACP is implemented based on ACP approach
(ii) Artificial Beijing is built with the help of KD-ACP
(iii) H1N1 is modeled to simulate the influenza in Beijingin 2009
(iv) Computational experiments testify the effectivenessof intervention measures
26 Mathematical Problems in Engineering
Conflict of Interests
The authors declare that there is no conflict of interestsregarding the publication of this paper
Acknowledgment
This work was supported in part by the National ScienceFoundation of China (nos 91024030 71303252 61403402and 91324013)
References
[1] W Mao and F-Y Wang Advances in Intelligence and SecurityInformatics Academic Press Oxford UK 2012
[2] E Bonabeau ldquoAgent-based modeling methods and techniquesfor simulating human systemsrdquo Proceedings of the NationalAcademy of Sciences of the United States of America vol 99 no3 pp 7280ndash7287 2002
[3] K M Carley D B Fridsma E Casman et al ldquoBioWar scalableagent-basedmodel of bioattacksrdquo IEEETransactions on SystemsMan and Cybernetics A Systems and Humans vol 36 no 2 pp252ndash265 2006
[4] J M Epstein ldquoModelling to contain pandemicsrdquo Nature vol460 no 7256 p 687 2009
[5] Los Alamos National Laboratory Reports about EPISIMS LosAlamos National Laboratory 2003
[6] C H Builder and S C Bankes ldquoArtificial societies a conceptfor basic research on the societal impacts of informationtechnologyrdquo RAND Report 149 RAND Corp 1991
[7] F-Y Wang ldquoComputational theory and method on complexsystemrdquo China Basic Science vol 6 no 5 pp 3ndash10 2004
[8] F-Y Wang ldquoToward a paradigm shift in social computing theACP approachrdquo IEEE Intelligent Systems vol 22 no 5 pp 65ndash67 2007
[9] F-Y Wang and S Tang ldquoArtificial societies for integratedand sustainable development of metropolitan systemsrdquo IEEEIntelligent Systems vol 19 no 4 pp 82ndash87 2004
[10] D Wen Y Yuan and X-R Li ldquoArtificial societies computa-tional experiments and parallel systems an investigation ona computational theory for complex socioeconomic systemsrdquoIEEE Transactions on Services Computing vol 6 no 2 pp 177ndash185 2013
[11] F Wang F Zeng and Y Yuan ldquoAn ACP-based approach forcomplexity analysis of E-commerce systemrdquo Complex Systemsand Complexity Science vol 5 no 3 pp 1ndash8 2008 (Chinese)
[12] F-Y Wang and P K Wong ldquoResearch commentary intelligentsystems and technology for integrative and predictivemedicinean ACP approachrdquoACMTransactions on Intelligent Systems andTechnology vol 4 no 2 pp 1ndash6 2013
[13] J Sifeng X Gang F Dong and H Chunpeng ldquoStudy on theemergency rescue decision support system of petrochemicalplant based on ACP theoryrdquo in Proceedings of the 6th Manage-ment Annual Meeting (MAM rsquo11) 2011 (Chinese)
[14] X Dong D Fan G Xiong F Zhu Z Zhang and Y Yao ldquoPar-allel Bus Rapid Transit (BRT) operation management systembased on ACP approachrdquo in Proceedings of the 6th ManagementAnnual Meeting (MAM rsquo11) Beijing China 2011
[15] B Ning F-Y Wang H-R Dong R-M Li D Wen and L LildquoParallel systems for Urban rail transportation based on ACP
approachrdquo Journal of Transportation Systems Engineering andInformation Technology vol 10 no 6 pp 23ndash28 2010 (Chinese)
[16] F-Y Wang J Zhao and S-X Lun ldquoArtificial power systemsfor the operation and management of complex power gridsrdquoSouthern Power System Technology vol 2 no 3 pp 1ndash11 2008(Chinese)
[17] M J North T R Howe N T Collier and J R Vos ldquoRepastsimphony development environmentrdquo in Proceedings of theAgent 2005 Conference on Generative Social Processes Modelsand Mechanisms 2005
[18] M J North T RHowe N T Collier and J R Vos ldquoRepast Sim-phony runtime systemrdquo in Proceedings of the Agent Conferenceon Generative Social Processes Models and Mechanisms 2005
[19] SWARM Development Group Swarm 211 Reference Guide2000 httpwwwswarmorg
[20] S Tisue and U Wilensky ldquoNetLogo a simple environment formodeling complexityrdquo in International Conference on ComplexSystems pp 16ndash21 2004
[21] S Luke C Cioffi-Revilla L Panait K Sullivan and G BalanldquoMASON a multiagent simulation environmentrdquo Simulationvol 81 no 7 pp 517ndash527 2005
[22] M Ichikawa H Tanuma Y Koyama andH Deguchi ldquoSOARSintroduction as a social microscope for simulations of socialinteractions and gamingrdquo in Organizing and Learning throughGaming and Simulation Proceedings of the 38th Conference ofthe International Simulation and Gaming Association (ISAGArsquo07) pp 149ndash158 2007
[23] D S Burke J M Epstein D A T Cummings et al ldquoIndividual-based computational modeling of smallpox epidemic controlstrategiesrdquo Academic Emergency Medicine vol 13 no 11 pp1142ndash1149 2006
[24] F-Y Wang ldquoParallel control and management for intelligenttransportation systems concepts architectures and applica-tionsrdquo IEEE Transactions on Intelligent Transportation Systemsvol 11 no 3 pp 630ndash638 2010
[25] Z Molnar D Balasubramanian and A Ledeczi ldquoAn introduc-tion to the genericmodeling environmentrdquo in Proceedings of theTOOLS Europe 2007 Workshop on Model-Driven DevelopmentTool Implementers Forum Zurich Switzerland 2007
[26] S Kelly and J-P Tolvanen Domain Framework in Domain-Specific Modeling Enabling Full Code Generation JohnWiley ampSons Hoboken NJ USA 2007
[27] B P Zeigler H Praehofer and T G Kim Theory of Modelingand Simulation Integrating Discrete Event and ContinuousComplex Dynamic Systems Academic Press San Diego CalifUSA 2000
[28] G Booch J Rumbaugh and I JacobsonThe Unified ModelingLanguage User Guide Addison-Wesley Reading Mass USA1999
[29] F Wagner R Schmuki T Wagne and P Wolstenholme Mod-eling Software with Finite State Machines A Practical ApproachCRC amp Taylor amp Francis Boca Raton Fla USA 2006
[30] T Espiner ldquoChina builds worldrsquos fastest supercomputerrdquo ZDNetUK October 2010
[31] B Chen and G Guo ldquoA two-tier parallel architecture for arti-ficial society simulationrdquo in Proceedings of the ACMIEEESCS26th Workshop on Principles of Advanced and DistributedSimulation (PADS rsquo12) pp 184ndash186 July 2012
[32] G Guo B Chen X G Qiu and Z Li ldquoParallel simulation oflarge-scale artificial society on CPUGPU mixed architecturerdquo
Mathematical Problems in Engineering 27
in Proceedings of the ACMIEEESCS 26th Workshop on Prin-ciples of Advanced and Distributed Simulation (PADS rsquo12) pp174ndash177 July 2012
[33] S-X Lun ldquoResearch on the classification of parallel executionmodes of ACP theoryrdquo Acta Automatica Sinica vol 38 no 10pp 1602ndash1608 2012
[34] F-YWang ldquoA framework for social signal processing and anal-ysis from social sensing networks to computational dialecticalanalyticsrdquo Science China vol 43 no 12 pp 1598ndash1611 2013
[35] G GuoOne Model Userrsquos Manual version 12 National Univer-sity of Defense University 2011
[36] Y Ge R Meng Z Cao X Qiu and K Huang ldquoVirtual citymdashanindividual-based digital environment for human mobility andinteractive behaviorrdquo Simulation Transactions of the Society forthe Modeling and Simulation International 2014
[37] M Kretzschmar and R T Mikolajczyk ldquoContact profiles ineight European countries and implications for modelling thespread of airborne infectious diseasesrdquo PLoS ONE vol 4 no6 Article ID e5931 2009
[38] W Duan and X Qiu ldquoFostering artificial societies using sociallearning and social control in parallel emergency managementsystemsrdquo Frontiers in Computer Science vol 6 no 5 pp 604ndash610 2012
[39] W J Edmunds C J OrsquoCallaghan and D J Nokes ldquoWhomixes with whom Amethod to determine the contact patternsof adults that may lead to the spread of airborne infectionsrdquoProceedings of the Royal Society B Biological Sciences vol 264no 1384 pp 949ndash957 1997
[40] J Banks J S Carson B L Neison and D M Nicol Discrete-Event System Simulation Prentice Hall Englewood Cliffs NJUSA 4th edition 2007
[41] H Yongxia ldquoInvestigation and analysis of college studentsrsquo workand restrdquo Value Engineering vol 31 no 3 2012 (Chinese)
[42] W Duan Z Cao Y Wang et al ldquoAn ACP approach to publichealth emergency management using a campus outbreak ofh1n1 influenza as a case studyrdquo IEEE Transactions on SystemsMan and Cybernetics Part A Systems and Humans vol 43 no5 pp 1028ndash1041 2013
[43] HWHethcote ldquoThemathematics of infectious diseasesrdquo SIAMReview vol 42 no 4 pp 599ndash653 2000
[44] G Chowell E Shim F Brauer and et al ldquoModelling thetransmission dynamics of acute haemorrhagic conjunctivitisapplication to the 2003 outbreak in Mexicordquo Statistics inMedicine vol 25 no 11 pp 1840ndash1857 2006
[45] Y Ge L Liu B Chen X Qiu and K Huang ldquoAgent-basedmodeling for InfluenzaH1N1 in an artificial classroomrdquo SystemsEngineering Procedia vol 2 pp 94ndash104 2011
[46] A R Tuite A L Greer MWhelan et al ldquoEstimated epidemio-logic parameters andmorbidity associatedwith pandemicH1N1influenzardquo CMAJ vol 182 no 2 pp 131ndash136 2010
[47] L Brouwers B Cakici M Camitz A Tegnell and MBoman ldquoEconomic consequences to society of pandemic H1N1influenza 2009mdashpreliminary results for Swedenrdquo Euro Surveil-lance vol 14 no 37 pp 1ndash7 2009
[48] Y Ge L Liu X Qiu H Song Y Wang and K Huang ldquoAframework of multilayer social networks for communicationbehavior with agent-based modelingrdquo Simulation Transactionsof the Society for the Modeling and Simulation International vol89 no 7 pp 810ndash828 2013
[49] W Duan X Qiu Z Cao X Zheng and K Cui ldquoHeterogeneousand stochastic agent-based models for analyzing infectious
diseasesrsquo super spreadersrdquo IEEE Intelligent Systems vol 28 no4 pp 18ndash25 2013
[50] The Preparation and Emergency Response Plans to PandemicPublished by Ministry of Health Ministry of Health 2005(Chinese)
[51] R M Fujimoto Parallel and Distributed Simulation SystemsWiley Interscience 2000
[52] F Cicirelli A Giordano and L Nigro ldquoDistributed simulationof situated multi-agent systemsrdquo in Proceedings of the 15thIEEEACM International Symposium on Distributed Simulationand Real Time Applications (DS-RT rsquo11) pp 28ndash35 September2011
[53] B Chen and G Guo ldquoA two-tier parallel architecturefor artificial society simulationrdquo in Proceedings of the 26thACMIEEESCS Workshop on Principles of Advanced and Dis-tributed Simulation (PADS rsquo12) pp 184ndash186 July 2012
[54] G Guo B Chen and X Qiu ldquoParallel simulation of large-scale artificial society with GPU as coprocessorrdquo InternationalJournal of Modeling Simulation and Scientific Computing vol4 no 2 Article ID 1350005 2013
[55] B Chen L-B Zhang X-C Liu and H Vangheluwe ldquoActivity-based simulation using DEVS increasing performance by anactivity model in parallel DEVS simulationrdquo Journal of ZhejiangUniversity Science C vol 15 no 1 pp 13ndash30 2014
[56] X-L Wang P Yang Z-D Cao et al ldquoQuantitative evaluationon the effectiveness of prevention and control measures againstpandemic influenza A (H1N1) in Beijing 2009rdquoChinese Journalof Epidemiology vol 31 no 12 pp 1374ndash1378 2010 (Chinese)
[57] S Mei A D van de Vijver L Xuan Y Zhu and P M ASloot ldquoQuantitatively evaluating interventions in the influenzaA (H1N1) epidemic on China campus grounded on individual-based simulationsrdquo Procedia Computer Science vol 1 no 1 pp1675ndash1682 2010
[58] X Qiu Z Cao and Z Li Mid-Term Reports of China NationalScience Foundation 91024030 2014
Submit your manuscripts athttpwwwhindawicom
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
MathematicsJournal of
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Mathematical Problems in Engineering
Hindawi Publishing Corporationhttpwwwhindawicom
Differential EquationsInternational Journal of
Volume 2014
Applied MathematicsJournal of
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Probability and StatisticsHindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Journal of
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Mathematical PhysicsAdvances in
Complex AnalysisJournal of
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
OptimizationJournal of
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
CombinatoricsHindawi Publishing Corporationhttpwwwhindawicom Volume 2014
International Journal of
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Operations ResearchAdvances in
Journal of
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Function Spaces
Abstract and Applied AnalysisHindawi Publishing Corporationhttpwwwhindawicom Volume 2014
International Journal of Mathematics and Mathematical Sciences
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
The Scientific World JournalHindawi Publishing Corporation httpwwwhindawicom Volume 2014
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Algebra
Discrete Dynamics in Nature and Society
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Decision SciencesAdvances in
Discrete MathematicsJournal of
Hindawi Publishing Corporationhttpwwwhindawicom
Volume 2014 Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Stochastic AnalysisInternational Journal of
Mathematical Problems in Engineering 23
Table 8 The initial settings of experiments of isolation
Parameters ValueAgent count 19610000Initial infected count 40Interventions activated time (119879Intervention) 80Delay to hospital (119879HDelay) isolation duration (119879IDuration) (15 0) (1 0) (05 0) (15 3) (1 3) (05 3)
0 50 100 150 200 2500
1000
2000
3000
4000
5000
6000
7000
Infe
cted
agen
ts
Loading intervention of isolation
Delay to hospital = 15
Delay to hospital = 10
Delay to hospital = 05
Delay to hospital = 15 isolation duration = 3
Delay to hospital = 10 isolation duration = 3
Delay to hospital = 05 isolation duration = 3
Day d
Figure 25 The results of loading intervention of isolation
are sent to hospital the less the agents will be infectedSo the construction of effective mechanism to find infectedindividuals is important in the emergency management ofinfluenza
Secondly the parameter isolation duration is added tosimulate the isolation of interventionObviously the isolationmeasure decreases the infected agents greatly Compared tothe parameter delay to hospital of 15-day case the peak valuedecreases from 6708 to 362when the isolation is added in theinterventionThe reduction of 6346 agents comes from the 3-day isolation of agents who had contacted with the infectedagents However the decrease of parameter delay to hospitalcannot bring the similar reduction in the 3-day isolationcases According to the figure the peak value decreases from362 to 51 from 15-day case to 05-day case only 311 agents aresaved from infection As a result the sensitivity of simulatedresults does not change obviously with delay to hospital inthe case of isolation It can be concluded that isolation is amore effective measure Though it is not easy for emergencymanagement organizations to decrease the delay to hospitaldue to the current monitoring mechanism the influenza canstill be controlled by isolation
544 The Experiments of Combination of Intervention Mea-sures Isolation is not the only nonmedical intervention ininfluenza close workspace is another common measureWhen close workspace is activated the workspaces are allclosed agents can only stay home at the closed time Luckilyisolation and close workspace are independent from oneanother It is possible for emergency management organi-zations to use these two measures together So this groupof experiments is designed to obtain the effectiveness of thecombination of intervention measures The parameters areinitialized with different settings listed in Table 9
Parameters delay to hospital and isolation duration is setin the middle values (1 day 3 days) according to the lastsection According to the experience close duration is set asseven days It is known that close workspace is a really costintervention measure the activated time is a key parameterWith the fixed parameters mentioned before the changeof interventions activated time is analyzed in this group ofcomputational experimentsThe average results are shown inFigure 26
Compared with the simulated results of experimentsdiscussed above Figure 26 strongly illustrates that influenzais ceased when the intervention measures are combinedtogether The peak value of infected agents is 141 in the ldquo80thdays of close-workspace and isolation with 1 day delay tohospitalrdquo case while 25135 in ldquono interventionrdquo case 2218 inldquo1 day delay to hospitalrdquo case 203 in ldquo1 day delay to hospitaland isolationrdquo case The interventions bring a great reductionof infection the peak values decrease from ten thousandslevel to hundred level
On the other hand the peak value even decreases to 108when the execution time of close workspace is 10 days earlierThe date of maximum value of infected agents is advancedfrom the 135th day to the 108th day The whole circle of H1N1influenza is reducedThe results of this group of experimentsshow that intervention measures should be activated as soonas possible and the compositions of interventions are moreeffective in emergency management of H1N1 influenza
In summary based on the general process lots of exper-iments could be done by KD-ACP to study emergency man-agement in H1N1 influenza Many conclusions are obtainedbased on the analysis of the experiments in this sectionAccording to these conclusions themost optimal response inour case study is the case of ldquo70th days of close-workspace andisolation with 1 day delay to hospitalrdquo It is because this casehas the fewest infected agents relatively But the modelling ofH1N1 influenza is not sufficient Economic and social factorsof interventions such as the money cost by vaccination andisolation the influence of the close-workspace in society
24 Mathematical Problems in Engineering
Table 9 The initial settings of experiments of isolation and closeworkspace
Parameters ValueAgent count 19610000Initial infected count 40Interventions activated time (119879Intervention) 70th day 80th dayDelay to hospital (119879HDelay) 1 dayIsolation duration (119879IDuration) 3 daysClose workspace duration (119879CWDuration) 7 days
50 100 150 200 2500
50
100
150
200
Infe
cted
agen
ts
Loading intervention of isolation and close workspace
Delay to hospital = 1 isolation duration = 3Start close time = 70 close duration time = 7Delay to hospital = 1 isolation duration = 3Start close time = 80 close duration time = 7
Day d
Figure 26The results of loading interventions of isolation and closeworkspace
and even opinion transmission of H1N1 in internet are notconsidered in our case study Without the analysis of thesefactors it does not make sense to talk about the most optimalintervention measures However KD-ACP had provided asoftware framework and general process of the research onemergency management Supported by KD-ACP the expertsof emergency management are able to study emergent eventssuch as H1N1 influenza as detailed as possible Based on theexperiments and analysis again and again the most optimalintervention measures will be found ultimately
55 Parallel Execution by ASST and OsdRT As discussed inSection 35 ASST andOsdRT are used to support the parallelexecution of artificial Beijing The statistical data is collectedand stored in artificial society runtime database at runtimeASST downloads the infection data from the database anddisplays it in the map As shown in Figure 27 the infectedagents are distributed in the districts in BeijingThe snapshotgives the geospatial distribution of the H1N1 epidemic Thelegends of the agents in different status are listed in the left
right corner in Figure 27 Five statuses are shown in the figurehealth without vaccination health with vaccination incuba-tion being symptomatic and convalescence ASST providesa very direct feeling for the emergency experts Partly withthe situation from ASST Emergency Decision Organizationmakes the emergency response plans to the response of theemergency scenario With the help of plans the epidemicin actual society can be restrained In the meantime theinfluenza information is reflected in the Internet Accordingto Figure 8 the data acquisition data extraction and datastandardization of H1N1 influenza are processed in OsdRTOsdRT is implemented by Dr Caorsquos team from Institute ofAutomation from Chinese Academy of Sciences [58] Thetool provides not only the reports of H1N1 SARS hand-foot-mouth disease and intestinal diseases are all monitored Theweb page is shown in Figure 28 H1N1 reports from Internetare collected first Based on the data extraction of the reportsH1N1 case of distribution is obtained and shown in theincidencemap As a result H1N1 incidence data is regarded asthe knowledge which is sent to EECT and IMCTThe settingsof emergent events and intervention measures are modifiedin order to make artificial society approach actual society innext turn of computational experiments With the help ofOsdRT the loop of parallel execution process integrates ACPapproach
6 Conclusion
This work provides an integrated software framework forsocial Computing in emergencymanagement From a systemperspective KD-ACP provides a reliable flexible low costand effective platform for the scientists to do the research onemergency response problems The analysis and predictionof these problems with inherent complexity can be solvedby the repetition of possible alternative experiments onthe software framework Therefore KD-ACP which is anattempt to implement ACP approach seemed as an effectivecomputational framework to support the decision making inemergency response
Currently KD-ACP is used to study the H1N1 epidemicin Beijing in 2009 The most optimal compositions of inter-vention measures are testified to support the response tonext influenza However it is still a long way ahead beforethe automation systematization and practicalization of KD-ACP A lot of work will be carried out along several directionsas in the following
(i) The parallel execution process in KD-ACP is actuallynot strongly connected with the emergency responseorganization Traditional experience and theory stillplay a dominant role in the decision of emergencymanagement It is necessary to propose a mechanismof hall for workshop to facilitate the parallel executionloop in actual society
(ii) Section 33 introduces the principle and process ofmodelling artificial society Metamodels and modelsof artificial Beijing are built by GME Model transfor-mation and code generation are used to implementcodes from FSM based models But many details
Mathematical Problems in Engineering 25
Fangshan district Daxing district Fengtai district Xicheng district
Mentougou district Haidian districtShijingshan district Dongcheng district
Incubation
Symptomatic
Convalescence
Health withoutvaccination
Health withvaccinationLegend of agent
health status
Figure 27 The situation of artificial Beijing by ASST
H1N1
Districts of Beijing
H1N1 case ofdistribution
Incidence map
Distribution ofH1N1 reportsfrom internet
Incidence of H1N1
Fangshan districtLiangxiang town
Population 19000Incidence 135789
(1100000)
Figure 28 The implementation of OsdRT by Institute of Automation from Chinese Academy of Sciences
are still not illustrated due to the topic of thispaper The key techniques such as template of codeframework and model transformations from specificdomains to FSM will be focused and introduced indetail in our next paper
(iii) The performance of computational experiments inKD-ACP is still slower than expectation The opti-mization of simulation engine and models structureis necessary to improve the performance So it ispossible to run artificial society with the real worldin parallel
In summary KD-ACP paves a new way for scientists inboth emergencymanagement and computation simulation to
collaborate with each other in solving the emergency man-agement problems by social computing
Research Highlights
(i) KD-ACP is implemented based on ACP approach
(ii) Artificial Beijing is built with the help of KD-ACP
(iii) H1N1 is modeled to simulate the influenza in Beijingin 2009
(iv) Computational experiments testify the effectivenessof intervention measures
26 Mathematical Problems in Engineering
Conflict of Interests
The authors declare that there is no conflict of interestsregarding the publication of this paper
Acknowledgment
This work was supported in part by the National ScienceFoundation of China (nos 91024030 71303252 61403402and 91324013)
References
[1] W Mao and F-Y Wang Advances in Intelligence and SecurityInformatics Academic Press Oxford UK 2012
[2] E Bonabeau ldquoAgent-based modeling methods and techniquesfor simulating human systemsrdquo Proceedings of the NationalAcademy of Sciences of the United States of America vol 99 no3 pp 7280ndash7287 2002
[3] K M Carley D B Fridsma E Casman et al ldquoBioWar scalableagent-basedmodel of bioattacksrdquo IEEETransactions on SystemsMan and Cybernetics A Systems and Humans vol 36 no 2 pp252ndash265 2006
[4] J M Epstein ldquoModelling to contain pandemicsrdquo Nature vol460 no 7256 p 687 2009
[5] Los Alamos National Laboratory Reports about EPISIMS LosAlamos National Laboratory 2003
[6] C H Builder and S C Bankes ldquoArtificial societies a conceptfor basic research on the societal impacts of informationtechnologyrdquo RAND Report 149 RAND Corp 1991
[7] F-Y Wang ldquoComputational theory and method on complexsystemrdquo China Basic Science vol 6 no 5 pp 3ndash10 2004
[8] F-Y Wang ldquoToward a paradigm shift in social computing theACP approachrdquo IEEE Intelligent Systems vol 22 no 5 pp 65ndash67 2007
[9] F-Y Wang and S Tang ldquoArtificial societies for integratedand sustainable development of metropolitan systemsrdquo IEEEIntelligent Systems vol 19 no 4 pp 82ndash87 2004
[10] D Wen Y Yuan and X-R Li ldquoArtificial societies computa-tional experiments and parallel systems an investigation ona computational theory for complex socioeconomic systemsrdquoIEEE Transactions on Services Computing vol 6 no 2 pp 177ndash185 2013
[11] F Wang F Zeng and Y Yuan ldquoAn ACP-based approach forcomplexity analysis of E-commerce systemrdquo Complex Systemsand Complexity Science vol 5 no 3 pp 1ndash8 2008 (Chinese)
[12] F-Y Wang and P K Wong ldquoResearch commentary intelligentsystems and technology for integrative and predictivemedicinean ACP approachrdquoACMTransactions on Intelligent Systems andTechnology vol 4 no 2 pp 1ndash6 2013
[13] J Sifeng X Gang F Dong and H Chunpeng ldquoStudy on theemergency rescue decision support system of petrochemicalplant based on ACP theoryrdquo in Proceedings of the 6th Manage-ment Annual Meeting (MAM rsquo11) 2011 (Chinese)
[14] X Dong D Fan G Xiong F Zhu Z Zhang and Y Yao ldquoPar-allel Bus Rapid Transit (BRT) operation management systembased on ACP approachrdquo in Proceedings of the 6th ManagementAnnual Meeting (MAM rsquo11) Beijing China 2011
[15] B Ning F-Y Wang H-R Dong R-M Li D Wen and L LildquoParallel systems for Urban rail transportation based on ACP
approachrdquo Journal of Transportation Systems Engineering andInformation Technology vol 10 no 6 pp 23ndash28 2010 (Chinese)
[16] F-Y Wang J Zhao and S-X Lun ldquoArtificial power systemsfor the operation and management of complex power gridsrdquoSouthern Power System Technology vol 2 no 3 pp 1ndash11 2008(Chinese)
[17] M J North T R Howe N T Collier and J R Vos ldquoRepastsimphony development environmentrdquo in Proceedings of theAgent 2005 Conference on Generative Social Processes Modelsand Mechanisms 2005
[18] M J North T RHowe N T Collier and J R Vos ldquoRepast Sim-phony runtime systemrdquo in Proceedings of the Agent Conferenceon Generative Social Processes Models and Mechanisms 2005
[19] SWARM Development Group Swarm 211 Reference Guide2000 httpwwwswarmorg
[20] S Tisue and U Wilensky ldquoNetLogo a simple environment formodeling complexityrdquo in International Conference on ComplexSystems pp 16ndash21 2004
[21] S Luke C Cioffi-Revilla L Panait K Sullivan and G BalanldquoMASON a multiagent simulation environmentrdquo Simulationvol 81 no 7 pp 517ndash527 2005
[22] M Ichikawa H Tanuma Y Koyama andH Deguchi ldquoSOARSintroduction as a social microscope for simulations of socialinteractions and gamingrdquo in Organizing and Learning throughGaming and Simulation Proceedings of the 38th Conference ofthe International Simulation and Gaming Association (ISAGArsquo07) pp 149ndash158 2007
[23] D S Burke J M Epstein D A T Cummings et al ldquoIndividual-based computational modeling of smallpox epidemic controlstrategiesrdquo Academic Emergency Medicine vol 13 no 11 pp1142ndash1149 2006
[24] F-Y Wang ldquoParallel control and management for intelligenttransportation systems concepts architectures and applica-tionsrdquo IEEE Transactions on Intelligent Transportation Systemsvol 11 no 3 pp 630ndash638 2010
[25] Z Molnar D Balasubramanian and A Ledeczi ldquoAn introduc-tion to the genericmodeling environmentrdquo in Proceedings of theTOOLS Europe 2007 Workshop on Model-Driven DevelopmentTool Implementers Forum Zurich Switzerland 2007
[26] S Kelly and J-P Tolvanen Domain Framework in Domain-Specific Modeling Enabling Full Code Generation JohnWiley ampSons Hoboken NJ USA 2007
[27] B P Zeigler H Praehofer and T G Kim Theory of Modelingand Simulation Integrating Discrete Event and ContinuousComplex Dynamic Systems Academic Press San Diego CalifUSA 2000
[28] G Booch J Rumbaugh and I JacobsonThe Unified ModelingLanguage User Guide Addison-Wesley Reading Mass USA1999
[29] F Wagner R Schmuki T Wagne and P Wolstenholme Mod-eling Software with Finite State Machines A Practical ApproachCRC amp Taylor amp Francis Boca Raton Fla USA 2006
[30] T Espiner ldquoChina builds worldrsquos fastest supercomputerrdquo ZDNetUK October 2010
[31] B Chen and G Guo ldquoA two-tier parallel architecture for arti-ficial society simulationrdquo in Proceedings of the ACMIEEESCS26th Workshop on Principles of Advanced and DistributedSimulation (PADS rsquo12) pp 184ndash186 July 2012
[32] G Guo B Chen X G Qiu and Z Li ldquoParallel simulation oflarge-scale artificial society on CPUGPU mixed architecturerdquo
Mathematical Problems in Engineering 27
in Proceedings of the ACMIEEESCS 26th Workshop on Prin-ciples of Advanced and Distributed Simulation (PADS rsquo12) pp174ndash177 July 2012
[33] S-X Lun ldquoResearch on the classification of parallel executionmodes of ACP theoryrdquo Acta Automatica Sinica vol 38 no 10pp 1602ndash1608 2012
[34] F-YWang ldquoA framework for social signal processing and anal-ysis from social sensing networks to computational dialecticalanalyticsrdquo Science China vol 43 no 12 pp 1598ndash1611 2013
[35] G GuoOne Model Userrsquos Manual version 12 National Univer-sity of Defense University 2011
[36] Y Ge R Meng Z Cao X Qiu and K Huang ldquoVirtual citymdashanindividual-based digital environment for human mobility andinteractive behaviorrdquo Simulation Transactions of the Society forthe Modeling and Simulation International 2014
[37] M Kretzschmar and R T Mikolajczyk ldquoContact profiles ineight European countries and implications for modelling thespread of airborne infectious diseasesrdquo PLoS ONE vol 4 no6 Article ID e5931 2009
[38] W Duan and X Qiu ldquoFostering artificial societies using sociallearning and social control in parallel emergency managementsystemsrdquo Frontiers in Computer Science vol 6 no 5 pp 604ndash610 2012
[39] W J Edmunds C J OrsquoCallaghan and D J Nokes ldquoWhomixes with whom Amethod to determine the contact patternsof adults that may lead to the spread of airborne infectionsrdquoProceedings of the Royal Society B Biological Sciences vol 264no 1384 pp 949ndash957 1997
[40] J Banks J S Carson B L Neison and D M Nicol Discrete-Event System Simulation Prentice Hall Englewood Cliffs NJUSA 4th edition 2007
[41] H Yongxia ldquoInvestigation and analysis of college studentsrsquo workand restrdquo Value Engineering vol 31 no 3 2012 (Chinese)
[42] W Duan Z Cao Y Wang et al ldquoAn ACP approach to publichealth emergency management using a campus outbreak ofh1n1 influenza as a case studyrdquo IEEE Transactions on SystemsMan and Cybernetics Part A Systems and Humans vol 43 no5 pp 1028ndash1041 2013
[43] HWHethcote ldquoThemathematics of infectious diseasesrdquo SIAMReview vol 42 no 4 pp 599ndash653 2000
[44] G Chowell E Shim F Brauer and et al ldquoModelling thetransmission dynamics of acute haemorrhagic conjunctivitisapplication to the 2003 outbreak in Mexicordquo Statistics inMedicine vol 25 no 11 pp 1840ndash1857 2006
[45] Y Ge L Liu B Chen X Qiu and K Huang ldquoAgent-basedmodeling for InfluenzaH1N1 in an artificial classroomrdquo SystemsEngineering Procedia vol 2 pp 94ndash104 2011
[46] A R Tuite A L Greer MWhelan et al ldquoEstimated epidemio-logic parameters andmorbidity associatedwith pandemicH1N1influenzardquo CMAJ vol 182 no 2 pp 131ndash136 2010
[47] L Brouwers B Cakici M Camitz A Tegnell and MBoman ldquoEconomic consequences to society of pandemic H1N1influenza 2009mdashpreliminary results for Swedenrdquo Euro Surveil-lance vol 14 no 37 pp 1ndash7 2009
[48] Y Ge L Liu X Qiu H Song Y Wang and K Huang ldquoAframework of multilayer social networks for communicationbehavior with agent-based modelingrdquo Simulation Transactionsof the Society for the Modeling and Simulation International vol89 no 7 pp 810ndash828 2013
[49] W Duan X Qiu Z Cao X Zheng and K Cui ldquoHeterogeneousand stochastic agent-based models for analyzing infectious
diseasesrsquo super spreadersrdquo IEEE Intelligent Systems vol 28 no4 pp 18ndash25 2013
[50] The Preparation and Emergency Response Plans to PandemicPublished by Ministry of Health Ministry of Health 2005(Chinese)
[51] R M Fujimoto Parallel and Distributed Simulation SystemsWiley Interscience 2000
[52] F Cicirelli A Giordano and L Nigro ldquoDistributed simulationof situated multi-agent systemsrdquo in Proceedings of the 15thIEEEACM International Symposium on Distributed Simulationand Real Time Applications (DS-RT rsquo11) pp 28ndash35 September2011
[53] B Chen and G Guo ldquoA two-tier parallel architecturefor artificial society simulationrdquo in Proceedings of the 26thACMIEEESCS Workshop on Principles of Advanced and Dis-tributed Simulation (PADS rsquo12) pp 184ndash186 July 2012
[54] G Guo B Chen and X Qiu ldquoParallel simulation of large-scale artificial society with GPU as coprocessorrdquo InternationalJournal of Modeling Simulation and Scientific Computing vol4 no 2 Article ID 1350005 2013
[55] B Chen L-B Zhang X-C Liu and H Vangheluwe ldquoActivity-based simulation using DEVS increasing performance by anactivity model in parallel DEVS simulationrdquo Journal of ZhejiangUniversity Science C vol 15 no 1 pp 13ndash30 2014
[56] X-L Wang P Yang Z-D Cao et al ldquoQuantitative evaluationon the effectiveness of prevention and control measures againstpandemic influenza A (H1N1) in Beijing 2009rdquoChinese Journalof Epidemiology vol 31 no 12 pp 1374ndash1378 2010 (Chinese)
[57] S Mei A D van de Vijver L Xuan Y Zhu and P M ASloot ldquoQuantitatively evaluating interventions in the influenzaA (H1N1) epidemic on China campus grounded on individual-based simulationsrdquo Procedia Computer Science vol 1 no 1 pp1675ndash1682 2010
[58] X Qiu Z Cao and Z Li Mid-Term Reports of China NationalScience Foundation 91024030 2014
Submit your manuscripts athttpwwwhindawicom
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
MathematicsJournal of
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Mathematical Problems in Engineering
Hindawi Publishing Corporationhttpwwwhindawicom
Differential EquationsInternational Journal of
Volume 2014
Applied MathematicsJournal of
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Probability and StatisticsHindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Journal of
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Mathematical PhysicsAdvances in
Complex AnalysisJournal of
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
OptimizationJournal of
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
CombinatoricsHindawi Publishing Corporationhttpwwwhindawicom Volume 2014
International Journal of
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Operations ResearchAdvances in
Journal of
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Function Spaces
Abstract and Applied AnalysisHindawi Publishing Corporationhttpwwwhindawicom Volume 2014
International Journal of Mathematics and Mathematical Sciences
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
The Scientific World JournalHindawi Publishing Corporation httpwwwhindawicom Volume 2014
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Algebra
Discrete Dynamics in Nature and Society
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Decision SciencesAdvances in
Discrete MathematicsJournal of
Hindawi Publishing Corporationhttpwwwhindawicom
Volume 2014 Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Stochastic AnalysisInternational Journal of
24 Mathematical Problems in Engineering
Table 9 The initial settings of experiments of isolation and closeworkspace
Parameters ValueAgent count 19610000Initial infected count 40Interventions activated time (119879Intervention) 70th day 80th dayDelay to hospital (119879HDelay) 1 dayIsolation duration (119879IDuration) 3 daysClose workspace duration (119879CWDuration) 7 days
50 100 150 200 2500
50
100
150
200
Infe
cted
agen
ts
Loading intervention of isolation and close workspace
Delay to hospital = 1 isolation duration = 3Start close time = 70 close duration time = 7Delay to hospital = 1 isolation duration = 3Start close time = 80 close duration time = 7
Day d
Figure 26The results of loading interventions of isolation and closeworkspace
and even opinion transmission of H1N1 in internet are notconsidered in our case study Without the analysis of thesefactors it does not make sense to talk about the most optimalintervention measures However KD-ACP had provided asoftware framework and general process of the research onemergency management Supported by KD-ACP the expertsof emergency management are able to study emergent eventssuch as H1N1 influenza as detailed as possible Based on theexperiments and analysis again and again the most optimalintervention measures will be found ultimately
55 Parallel Execution by ASST and OsdRT As discussed inSection 35 ASST andOsdRT are used to support the parallelexecution of artificial Beijing The statistical data is collectedand stored in artificial society runtime database at runtimeASST downloads the infection data from the database anddisplays it in the map As shown in Figure 27 the infectedagents are distributed in the districts in BeijingThe snapshotgives the geospatial distribution of the H1N1 epidemic Thelegends of the agents in different status are listed in the left
right corner in Figure 27 Five statuses are shown in the figurehealth without vaccination health with vaccination incuba-tion being symptomatic and convalescence ASST providesa very direct feeling for the emergency experts Partly withthe situation from ASST Emergency Decision Organizationmakes the emergency response plans to the response of theemergency scenario With the help of plans the epidemicin actual society can be restrained In the meantime theinfluenza information is reflected in the Internet Accordingto Figure 8 the data acquisition data extraction and datastandardization of H1N1 influenza are processed in OsdRTOsdRT is implemented by Dr Caorsquos team from Institute ofAutomation from Chinese Academy of Sciences [58] Thetool provides not only the reports of H1N1 SARS hand-foot-mouth disease and intestinal diseases are all monitored Theweb page is shown in Figure 28 H1N1 reports from Internetare collected first Based on the data extraction of the reportsH1N1 case of distribution is obtained and shown in theincidencemap As a result H1N1 incidence data is regarded asthe knowledge which is sent to EECT and IMCTThe settingsof emergent events and intervention measures are modifiedin order to make artificial society approach actual society innext turn of computational experiments With the help ofOsdRT the loop of parallel execution process integrates ACPapproach
6 Conclusion
This work provides an integrated software framework forsocial Computing in emergencymanagement From a systemperspective KD-ACP provides a reliable flexible low costand effective platform for the scientists to do the research onemergency response problems The analysis and predictionof these problems with inherent complexity can be solvedby the repetition of possible alternative experiments onthe software framework Therefore KD-ACP which is anattempt to implement ACP approach seemed as an effectivecomputational framework to support the decision making inemergency response
Currently KD-ACP is used to study the H1N1 epidemicin Beijing in 2009 The most optimal compositions of inter-vention measures are testified to support the response tonext influenza However it is still a long way ahead beforethe automation systematization and practicalization of KD-ACP A lot of work will be carried out along several directionsas in the following
(i) The parallel execution process in KD-ACP is actuallynot strongly connected with the emergency responseorganization Traditional experience and theory stillplay a dominant role in the decision of emergencymanagement It is necessary to propose a mechanismof hall for workshop to facilitate the parallel executionloop in actual society
(ii) Section 33 introduces the principle and process ofmodelling artificial society Metamodels and modelsof artificial Beijing are built by GME Model transfor-mation and code generation are used to implementcodes from FSM based models But many details
Mathematical Problems in Engineering 25
Fangshan district Daxing district Fengtai district Xicheng district
Mentougou district Haidian districtShijingshan district Dongcheng district
Incubation
Symptomatic
Convalescence
Health withoutvaccination
Health withvaccinationLegend of agent
health status
Figure 27 The situation of artificial Beijing by ASST
H1N1
Districts of Beijing
H1N1 case ofdistribution
Incidence map
Distribution ofH1N1 reportsfrom internet
Incidence of H1N1
Fangshan districtLiangxiang town
Population 19000Incidence 135789
(1100000)
Figure 28 The implementation of OsdRT by Institute of Automation from Chinese Academy of Sciences
are still not illustrated due to the topic of thispaper The key techniques such as template of codeframework and model transformations from specificdomains to FSM will be focused and introduced indetail in our next paper
(iii) The performance of computational experiments inKD-ACP is still slower than expectation The opti-mization of simulation engine and models structureis necessary to improve the performance So it ispossible to run artificial society with the real worldin parallel
In summary KD-ACP paves a new way for scientists inboth emergencymanagement and computation simulation to
collaborate with each other in solving the emergency man-agement problems by social computing
Research Highlights
(i) KD-ACP is implemented based on ACP approach
(ii) Artificial Beijing is built with the help of KD-ACP
(iii) H1N1 is modeled to simulate the influenza in Beijingin 2009
(iv) Computational experiments testify the effectivenessof intervention measures
26 Mathematical Problems in Engineering
Conflict of Interests
The authors declare that there is no conflict of interestsregarding the publication of this paper
Acknowledgment
This work was supported in part by the National ScienceFoundation of China (nos 91024030 71303252 61403402and 91324013)
References
[1] W Mao and F-Y Wang Advances in Intelligence and SecurityInformatics Academic Press Oxford UK 2012
[2] E Bonabeau ldquoAgent-based modeling methods and techniquesfor simulating human systemsrdquo Proceedings of the NationalAcademy of Sciences of the United States of America vol 99 no3 pp 7280ndash7287 2002
[3] K M Carley D B Fridsma E Casman et al ldquoBioWar scalableagent-basedmodel of bioattacksrdquo IEEETransactions on SystemsMan and Cybernetics A Systems and Humans vol 36 no 2 pp252ndash265 2006
[4] J M Epstein ldquoModelling to contain pandemicsrdquo Nature vol460 no 7256 p 687 2009
[5] Los Alamos National Laboratory Reports about EPISIMS LosAlamos National Laboratory 2003
[6] C H Builder and S C Bankes ldquoArtificial societies a conceptfor basic research on the societal impacts of informationtechnologyrdquo RAND Report 149 RAND Corp 1991
[7] F-Y Wang ldquoComputational theory and method on complexsystemrdquo China Basic Science vol 6 no 5 pp 3ndash10 2004
[8] F-Y Wang ldquoToward a paradigm shift in social computing theACP approachrdquo IEEE Intelligent Systems vol 22 no 5 pp 65ndash67 2007
[9] F-Y Wang and S Tang ldquoArtificial societies for integratedand sustainable development of metropolitan systemsrdquo IEEEIntelligent Systems vol 19 no 4 pp 82ndash87 2004
[10] D Wen Y Yuan and X-R Li ldquoArtificial societies computa-tional experiments and parallel systems an investigation ona computational theory for complex socioeconomic systemsrdquoIEEE Transactions on Services Computing vol 6 no 2 pp 177ndash185 2013
[11] F Wang F Zeng and Y Yuan ldquoAn ACP-based approach forcomplexity analysis of E-commerce systemrdquo Complex Systemsand Complexity Science vol 5 no 3 pp 1ndash8 2008 (Chinese)
[12] F-Y Wang and P K Wong ldquoResearch commentary intelligentsystems and technology for integrative and predictivemedicinean ACP approachrdquoACMTransactions on Intelligent Systems andTechnology vol 4 no 2 pp 1ndash6 2013
[13] J Sifeng X Gang F Dong and H Chunpeng ldquoStudy on theemergency rescue decision support system of petrochemicalplant based on ACP theoryrdquo in Proceedings of the 6th Manage-ment Annual Meeting (MAM rsquo11) 2011 (Chinese)
[14] X Dong D Fan G Xiong F Zhu Z Zhang and Y Yao ldquoPar-allel Bus Rapid Transit (BRT) operation management systembased on ACP approachrdquo in Proceedings of the 6th ManagementAnnual Meeting (MAM rsquo11) Beijing China 2011
[15] B Ning F-Y Wang H-R Dong R-M Li D Wen and L LildquoParallel systems for Urban rail transportation based on ACP
approachrdquo Journal of Transportation Systems Engineering andInformation Technology vol 10 no 6 pp 23ndash28 2010 (Chinese)
[16] F-Y Wang J Zhao and S-X Lun ldquoArtificial power systemsfor the operation and management of complex power gridsrdquoSouthern Power System Technology vol 2 no 3 pp 1ndash11 2008(Chinese)
[17] M J North T R Howe N T Collier and J R Vos ldquoRepastsimphony development environmentrdquo in Proceedings of theAgent 2005 Conference on Generative Social Processes Modelsand Mechanisms 2005
[18] M J North T RHowe N T Collier and J R Vos ldquoRepast Sim-phony runtime systemrdquo in Proceedings of the Agent Conferenceon Generative Social Processes Models and Mechanisms 2005
[19] SWARM Development Group Swarm 211 Reference Guide2000 httpwwwswarmorg
[20] S Tisue and U Wilensky ldquoNetLogo a simple environment formodeling complexityrdquo in International Conference on ComplexSystems pp 16ndash21 2004
[21] S Luke C Cioffi-Revilla L Panait K Sullivan and G BalanldquoMASON a multiagent simulation environmentrdquo Simulationvol 81 no 7 pp 517ndash527 2005
[22] M Ichikawa H Tanuma Y Koyama andH Deguchi ldquoSOARSintroduction as a social microscope for simulations of socialinteractions and gamingrdquo in Organizing and Learning throughGaming and Simulation Proceedings of the 38th Conference ofthe International Simulation and Gaming Association (ISAGArsquo07) pp 149ndash158 2007
[23] D S Burke J M Epstein D A T Cummings et al ldquoIndividual-based computational modeling of smallpox epidemic controlstrategiesrdquo Academic Emergency Medicine vol 13 no 11 pp1142ndash1149 2006
[24] F-Y Wang ldquoParallel control and management for intelligenttransportation systems concepts architectures and applica-tionsrdquo IEEE Transactions on Intelligent Transportation Systemsvol 11 no 3 pp 630ndash638 2010
[25] Z Molnar D Balasubramanian and A Ledeczi ldquoAn introduc-tion to the genericmodeling environmentrdquo in Proceedings of theTOOLS Europe 2007 Workshop on Model-Driven DevelopmentTool Implementers Forum Zurich Switzerland 2007
[26] S Kelly and J-P Tolvanen Domain Framework in Domain-Specific Modeling Enabling Full Code Generation JohnWiley ampSons Hoboken NJ USA 2007
[27] B P Zeigler H Praehofer and T G Kim Theory of Modelingand Simulation Integrating Discrete Event and ContinuousComplex Dynamic Systems Academic Press San Diego CalifUSA 2000
[28] G Booch J Rumbaugh and I JacobsonThe Unified ModelingLanguage User Guide Addison-Wesley Reading Mass USA1999
[29] F Wagner R Schmuki T Wagne and P Wolstenholme Mod-eling Software with Finite State Machines A Practical ApproachCRC amp Taylor amp Francis Boca Raton Fla USA 2006
[30] T Espiner ldquoChina builds worldrsquos fastest supercomputerrdquo ZDNetUK October 2010
[31] B Chen and G Guo ldquoA two-tier parallel architecture for arti-ficial society simulationrdquo in Proceedings of the ACMIEEESCS26th Workshop on Principles of Advanced and DistributedSimulation (PADS rsquo12) pp 184ndash186 July 2012
[32] G Guo B Chen X G Qiu and Z Li ldquoParallel simulation oflarge-scale artificial society on CPUGPU mixed architecturerdquo
Mathematical Problems in Engineering 27
in Proceedings of the ACMIEEESCS 26th Workshop on Prin-ciples of Advanced and Distributed Simulation (PADS rsquo12) pp174ndash177 July 2012
[33] S-X Lun ldquoResearch on the classification of parallel executionmodes of ACP theoryrdquo Acta Automatica Sinica vol 38 no 10pp 1602ndash1608 2012
[34] F-YWang ldquoA framework for social signal processing and anal-ysis from social sensing networks to computational dialecticalanalyticsrdquo Science China vol 43 no 12 pp 1598ndash1611 2013
[35] G GuoOne Model Userrsquos Manual version 12 National Univer-sity of Defense University 2011
[36] Y Ge R Meng Z Cao X Qiu and K Huang ldquoVirtual citymdashanindividual-based digital environment for human mobility andinteractive behaviorrdquo Simulation Transactions of the Society forthe Modeling and Simulation International 2014
[37] M Kretzschmar and R T Mikolajczyk ldquoContact profiles ineight European countries and implications for modelling thespread of airborne infectious diseasesrdquo PLoS ONE vol 4 no6 Article ID e5931 2009
[38] W Duan and X Qiu ldquoFostering artificial societies using sociallearning and social control in parallel emergency managementsystemsrdquo Frontiers in Computer Science vol 6 no 5 pp 604ndash610 2012
[39] W J Edmunds C J OrsquoCallaghan and D J Nokes ldquoWhomixes with whom Amethod to determine the contact patternsof adults that may lead to the spread of airborne infectionsrdquoProceedings of the Royal Society B Biological Sciences vol 264no 1384 pp 949ndash957 1997
[40] J Banks J S Carson B L Neison and D M Nicol Discrete-Event System Simulation Prentice Hall Englewood Cliffs NJUSA 4th edition 2007
[41] H Yongxia ldquoInvestigation and analysis of college studentsrsquo workand restrdquo Value Engineering vol 31 no 3 2012 (Chinese)
[42] W Duan Z Cao Y Wang et al ldquoAn ACP approach to publichealth emergency management using a campus outbreak ofh1n1 influenza as a case studyrdquo IEEE Transactions on SystemsMan and Cybernetics Part A Systems and Humans vol 43 no5 pp 1028ndash1041 2013
[43] HWHethcote ldquoThemathematics of infectious diseasesrdquo SIAMReview vol 42 no 4 pp 599ndash653 2000
[44] G Chowell E Shim F Brauer and et al ldquoModelling thetransmission dynamics of acute haemorrhagic conjunctivitisapplication to the 2003 outbreak in Mexicordquo Statistics inMedicine vol 25 no 11 pp 1840ndash1857 2006
[45] Y Ge L Liu B Chen X Qiu and K Huang ldquoAgent-basedmodeling for InfluenzaH1N1 in an artificial classroomrdquo SystemsEngineering Procedia vol 2 pp 94ndash104 2011
[46] A R Tuite A L Greer MWhelan et al ldquoEstimated epidemio-logic parameters andmorbidity associatedwith pandemicH1N1influenzardquo CMAJ vol 182 no 2 pp 131ndash136 2010
[47] L Brouwers B Cakici M Camitz A Tegnell and MBoman ldquoEconomic consequences to society of pandemic H1N1influenza 2009mdashpreliminary results for Swedenrdquo Euro Surveil-lance vol 14 no 37 pp 1ndash7 2009
[48] Y Ge L Liu X Qiu H Song Y Wang and K Huang ldquoAframework of multilayer social networks for communicationbehavior with agent-based modelingrdquo Simulation Transactionsof the Society for the Modeling and Simulation International vol89 no 7 pp 810ndash828 2013
[49] W Duan X Qiu Z Cao X Zheng and K Cui ldquoHeterogeneousand stochastic agent-based models for analyzing infectious
diseasesrsquo super spreadersrdquo IEEE Intelligent Systems vol 28 no4 pp 18ndash25 2013
[50] The Preparation and Emergency Response Plans to PandemicPublished by Ministry of Health Ministry of Health 2005(Chinese)
[51] R M Fujimoto Parallel and Distributed Simulation SystemsWiley Interscience 2000
[52] F Cicirelli A Giordano and L Nigro ldquoDistributed simulationof situated multi-agent systemsrdquo in Proceedings of the 15thIEEEACM International Symposium on Distributed Simulationand Real Time Applications (DS-RT rsquo11) pp 28ndash35 September2011
[53] B Chen and G Guo ldquoA two-tier parallel architecturefor artificial society simulationrdquo in Proceedings of the 26thACMIEEESCS Workshop on Principles of Advanced and Dis-tributed Simulation (PADS rsquo12) pp 184ndash186 July 2012
[54] G Guo B Chen and X Qiu ldquoParallel simulation of large-scale artificial society with GPU as coprocessorrdquo InternationalJournal of Modeling Simulation and Scientific Computing vol4 no 2 Article ID 1350005 2013
[55] B Chen L-B Zhang X-C Liu and H Vangheluwe ldquoActivity-based simulation using DEVS increasing performance by anactivity model in parallel DEVS simulationrdquo Journal of ZhejiangUniversity Science C vol 15 no 1 pp 13ndash30 2014
[56] X-L Wang P Yang Z-D Cao et al ldquoQuantitative evaluationon the effectiveness of prevention and control measures againstpandemic influenza A (H1N1) in Beijing 2009rdquoChinese Journalof Epidemiology vol 31 no 12 pp 1374ndash1378 2010 (Chinese)
[57] S Mei A D van de Vijver L Xuan Y Zhu and P M ASloot ldquoQuantitatively evaluating interventions in the influenzaA (H1N1) epidemic on China campus grounded on individual-based simulationsrdquo Procedia Computer Science vol 1 no 1 pp1675ndash1682 2010
[58] X Qiu Z Cao and Z Li Mid-Term Reports of China NationalScience Foundation 91024030 2014
Submit your manuscripts athttpwwwhindawicom
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
MathematicsJournal of
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Mathematical Problems in Engineering
Hindawi Publishing Corporationhttpwwwhindawicom
Differential EquationsInternational Journal of
Volume 2014
Applied MathematicsJournal of
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Probability and StatisticsHindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Journal of
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Mathematical PhysicsAdvances in
Complex AnalysisJournal of
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
OptimizationJournal of
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
CombinatoricsHindawi Publishing Corporationhttpwwwhindawicom Volume 2014
International Journal of
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Operations ResearchAdvances in
Journal of
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Function Spaces
Abstract and Applied AnalysisHindawi Publishing Corporationhttpwwwhindawicom Volume 2014
International Journal of Mathematics and Mathematical Sciences
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
The Scientific World JournalHindawi Publishing Corporation httpwwwhindawicom Volume 2014
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Algebra
Discrete Dynamics in Nature and Society
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Decision SciencesAdvances in
Discrete MathematicsJournal of
Hindawi Publishing Corporationhttpwwwhindawicom
Volume 2014 Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Stochastic AnalysisInternational Journal of
Mathematical Problems in Engineering 25
Fangshan district Daxing district Fengtai district Xicheng district
Mentougou district Haidian districtShijingshan district Dongcheng district
Incubation
Symptomatic
Convalescence
Health withoutvaccination
Health withvaccinationLegend of agent
health status
Figure 27 The situation of artificial Beijing by ASST
H1N1
Districts of Beijing
H1N1 case ofdistribution
Incidence map
Distribution ofH1N1 reportsfrom internet
Incidence of H1N1
Fangshan districtLiangxiang town
Population 19000Incidence 135789
(1100000)
Figure 28 The implementation of OsdRT by Institute of Automation from Chinese Academy of Sciences
are still not illustrated due to the topic of thispaper The key techniques such as template of codeframework and model transformations from specificdomains to FSM will be focused and introduced indetail in our next paper
(iii) The performance of computational experiments inKD-ACP is still slower than expectation The opti-mization of simulation engine and models structureis necessary to improve the performance So it ispossible to run artificial society with the real worldin parallel
In summary KD-ACP paves a new way for scientists inboth emergencymanagement and computation simulation to
collaborate with each other in solving the emergency man-agement problems by social computing
Research Highlights
(i) KD-ACP is implemented based on ACP approach
(ii) Artificial Beijing is built with the help of KD-ACP
(iii) H1N1 is modeled to simulate the influenza in Beijingin 2009
(iv) Computational experiments testify the effectivenessof intervention measures
26 Mathematical Problems in Engineering
Conflict of Interests
The authors declare that there is no conflict of interestsregarding the publication of this paper
Acknowledgment
This work was supported in part by the National ScienceFoundation of China (nos 91024030 71303252 61403402and 91324013)
References
[1] W Mao and F-Y Wang Advances in Intelligence and SecurityInformatics Academic Press Oxford UK 2012
[2] E Bonabeau ldquoAgent-based modeling methods and techniquesfor simulating human systemsrdquo Proceedings of the NationalAcademy of Sciences of the United States of America vol 99 no3 pp 7280ndash7287 2002
[3] K M Carley D B Fridsma E Casman et al ldquoBioWar scalableagent-basedmodel of bioattacksrdquo IEEETransactions on SystemsMan and Cybernetics A Systems and Humans vol 36 no 2 pp252ndash265 2006
[4] J M Epstein ldquoModelling to contain pandemicsrdquo Nature vol460 no 7256 p 687 2009
[5] Los Alamos National Laboratory Reports about EPISIMS LosAlamos National Laboratory 2003
[6] C H Builder and S C Bankes ldquoArtificial societies a conceptfor basic research on the societal impacts of informationtechnologyrdquo RAND Report 149 RAND Corp 1991
[7] F-Y Wang ldquoComputational theory and method on complexsystemrdquo China Basic Science vol 6 no 5 pp 3ndash10 2004
[8] F-Y Wang ldquoToward a paradigm shift in social computing theACP approachrdquo IEEE Intelligent Systems vol 22 no 5 pp 65ndash67 2007
[9] F-Y Wang and S Tang ldquoArtificial societies for integratedand sustainable development of metropolitan systemsrdquo IEEEIntelligent Systems vol 19 no 4 pp 82ndash87 2004
[10] D Wen Y Yuan and X-R Li ldquoArtificial societies computa-tional experiments and parallel systems an investigation ona computational theory for complex socioeconomic systemsrdquoIEEE Transactions on Services Computing vol 6 no 2 pp 177ndash185 2013
[11] F Wang F Zeng and Y Yuan ldquoAn ACP-based approach forcomplexity analysis of E-commerce systemrdquo Complex Systemsand Complexity Science vol 5 no 3 pp 1ndash8 2008 (Chinese)
[12] F-Y Wang and P K Wong ldquoResearch commentary intelligentsystems and technology for integrative and predictivemedicinean ACP approachrdquoACMTransactions on Intelligent Systems andTechnology vol 4 no 2 pp 1ndash6 2013
[13] J Sifeng X Gang F Dong and H Chunpeng ldquoStudy on theemergency rescue decision support system of petrochemicalplant based on ACP theoryrdquo in Proceedings of the 6th Manage-ment Annual Meeting (MAM rsquo11) 2011 (Chinese)
[14] X Dong D Fan G Xiong F Zhu Z Zhang and Y Yao ldquoPar-allel Bus Rapid Transit (BRT) operation management systembased on ACP approachrdquo in Proceedings of the 6th ManagementAnnual Meeting (MAM rsquo11) Beijing China 2011
[15] B Ning F-Y Wang H-R Dong R-M Li D Wen and L LildquoParallel systems for Urban rail transportation based on ACP
approachrdquo Journal of Transportation Systems Engineering andInformation Technology vol 10 no 6 pp 23ndash28 2010 (Chinese)
[16] F-Y Wang J Zhao and S-X Lun ldquoArtificial power systemsfor the operation and management of complex power gridsrdquoSouthern Power System Technology vol 2 no 3 pp 1ndash11 2008(Chinese)
[17] M J North T R Howe N T Collier and J R Vos ldquoRepastsimphony development environmentrdquo in Proceedings of theAgent 2005 Conference on Generative Social Processes Modelsand Mechanisms 2005
[18] M J North T RHowe N T Collier and J R Vos ldquoRepast Sim-phony runtime systemrdquo in Proceedings of the Agent Conferenceon Generative Social Processes Models and Mechanisms 2005
[19] SWARM Development Group Swarm 211 Reference Guide2000 httpwwwswarmorg
[20] S Tisue and U Wilensky ldquoNetLogo a simple environment formodeling complexityrdquo in International Conference on ComplexSystems pp 16ndash21 2004
[21] S Luke C Cioffi-Revilla L Panait K Sullivan and G BalanldquoMASON a multiagent simulation environmentrdquo Simulationvol 81 no 7 pp 517ndash527 2005
[22] M Ichikawa H Tanuma Y Koyama andH Deguchi ldquoSOARSintroduction as a social microscope for simulations of socialinteractions and gamingrdquo in Organizing and Learning throughGaming and Simulation Proceedings of the 38th Conference ofthe International Simulation and Gaming Association (ISAGArsquo07) pp 149ndash158 2007
[23] D S Burke J M Epstein D A T Cummings et al ldquoIndividual-based computational modeling of smallpox epidemic controlstrategiesrdquo Academic Emergency Medicine vol 13 no 11 pp1142ndash1149 2006
[24] F-Y Wang ldquoParallel control and management for intelligenttransportation systems concepts architectures and applica-tionsrdquo IEEE Transactions on Intelligent Transportation Systemsvol 11 no 3 pp 630ndash638 2010
[25] Z Molnar D Balasubramanian and A Ledeczi ldquoAn introduc-tion to the genericmodeling environmentrdquo in Proceedings of theTOOLS Europe 2007 Workshop on Model-Driven DevelopmentTool Implementers Forum Zurich Switzerland 2007
[26] S Kelly and J-P Tolvanen Domain Framework in Domain-Specific Modeling Enabling Full Code Generation JohnWiley ampSons Hoboken NJ USA 2007
[27] B P Zeigler H Praehofer and T G Kim Theory of Modelingand Simulation Integrating Discrete Event and ContinuousComplex Dynamic Systems Academic Press San Diego CalifUSA 2000
[28] G Booch J Rumbaugh and I JacobsonThe Unified ModelingLanguage User Guide Addison-Wesley Reading Mass USA1999
[29] F Wagner R Schmuki T Wagne and P Wolstenholme Mod-eling Software with Finite State Machines A Practical ApproachCRC amp Taylor amp Francis Boca Raton Fla USA 2006
[30] T Espiner ldquoChina builds worldrsquos fastest supercomputerrdquo ZDNetUK October 2010
[31] B Chen and G Guo ldquoA two-tier parallel architecture for arti-ficial society simulationrdquo in Proceedings of the ACMIEEESCS26th Workshop on Principles of Advanced and DistributedSimulation (PADS rsquo12) pp 184ndash186 July 2012
[32] G Guo B Chen X G Qiu and Z Li ldquoParallel simulation oflarge-scale artificial society on CPUGPU mixed architecturerdquo
Mathematical Problems in Engineering 27
in Proceedings of the ACMIEEESCS 26th Workshop on Prin-ciples of Advanced and Distributed Simulation (PADS rsquo12) pp174ndash177 July 2012
[33] S-X Lun ldquoResearch on the classification of parallel executionmodes of ACP theoryrdquo Acta Automatica Sinica vol 38 no 10pp 1602ndash1608 2012
[34] F-YWang ldquoA framework for social signal processing and anal-ysis from social sensing networks to computational dialecticalanalyticsrdquo Science China vol 43 no 12 pp 1598ndash1611 2013
[35] G GuoOne Model Userrsquos Manual version 12 National Univer-sity of Defense University 2011
[36] Y Ge R Meng Z Cao X Qiu and K Huang ldquoVirtual citymdashanindividual-based digital environment for human mobility andinteractive behaviorrdquo Simulation Transactions of the Society forthe Modeling and Simulation International 2014
[37] M Kretzschmar and R T Mikolajczyk ldquoContact profiles ineight European countries and implications for modelling thespread of airborne infectious diseasesrdquo PLoS ONE vol 4 no6 Article ID e5931 2009
[38] W Duan and X Qiu ldquoFostering artificial societies using sociallearning and social control in parallel emergency managementsystemsrdquo Frontiers in Computer Science vol 6 no 5 pp 604ndash610 2012
[39] W J Edmunds C J OrsquoCallaghan and D J Nokes ldquoWhomixes with whom Amethod to determine the contact patternsof adults that may lead to the spread of airborne infectionsrdquoProceedings of the Royal Society B Biological Sciences vol 264no 1384 pp 949ndash957 1997
[40] J Banks J S Carson B L Neison and D M Nicol Discrete-Event System Simulation Prentice Hall Englewood Cliffs NJUSA 4th edition 2007
[41] H Yongxia ldquoInvestigation and analysis of college studentsrsquo workand restrdquo Value Engineering vol 31 no 3 2012 (Chinese)
[42] W Duan Z Cao Y Wang et al ldquoAn ACP approach to publichealth emergency management using a campus outbreak ofh1n1 influenza as a case studyrdquo IEEE Transactions on SystemsMan and Cybernetics Part A Systems and Humans vol 43 no5 pp 1028ndash1041 2013
[43] HWHethcote ldquoThemathematics of infectious diseasesrdquo SIAMReview vol 42 no 4 pp 599ndash653 2000
[44] G Chowell E Shim F Brauer and et al ldquoModelling thetransmission dynamics of acute haemorrhagic conjunctivitisapplication to the 2003 outbreak in Mexicordquo Statistics inMedicine vol 25 no 11 pp 1840ndash1857 2006
[45] Y Ge L Liu B Chen X Qiu and K Huang ldquoAgent-basedmodeling for InfluenzaH1N1 in an artificial classroomrdquo SystemsEngineering Procedia vol 2 pp 94ndash104 2011
[46] A R Tuite A L Greer MWhelan et al ldquoEstimated epidemio-logic parameters andmorbidity associatedwith pandemicH1N1influenzardquo CMAJ vol 182 no 2 pp 131ndash136 2010
[47] L Brouwers B Cakici M Camitz A Tegnell and MBoman ldquoEconomic consequences to society of pandemic H1N1influenza 2009mdashpreliminary results for Swedenrdquo Euro Surveil-lance vol 14 no 37 pp 1ndash7 2009
[48] Y Ge L Liu X Qiu H Song Y Wang and K Huang ldquoAframework of multilayer social networks for communicationbehavior with agent-based modelingrdquo Simulation Transactionsof the Society for the Modeling and Simulation International vol89 no 7 pp 810ndash828 2013
[49] W Duan X Qiu Z Cao X Zheng and K Cui ldquoHeterogeneousand stochastic agent-based models for analyzing infectious
diseasesrsquo super spreadersrdquo IEEE Intelligent Systems vol 28 no4 pp 18ndash25 2013
[50] The Preparation and Emergency Response Plans to PandemicPublished by Ministry of Health Ministry of Health 2005(Chinese)
[51] R M Fujimoto Parallel and Distributed Simulation SystemsWiley Interscience 2000
[52] F Cicirelli A Giordano and L Nigro ldquoDistributed simulationof situated multi-agent systemsrdquo in Proceedings of the 15thIEEEACM International Symposium on Distributed Simulationand Real Time Applications (DS-RT rsquo11) pp 28ndash35 September2011
[53] B Chen and G Guo ldquoA two-tier parallel architecturefor artificial society simulationrdquo in Proceedings of the 26thACMIEEESCS Workshop on Principles of Advanced and Dis-tributed Simulation (PADS rsquo12) pp 184ndash186 July 2012
[54] G Guo B Chen and X Qiu ldquoParallel simulation of large-scale artificial society with GPU as coprocessorrdquo InternationalJournal of Modeling Simulation and Scientific Computing vol4 no 2 Article ID 1350005 2013
[55] B Chen L-B Zhang X-C Liu and H Vangheluwe ldquoActivity-based simulation using DEVS increasing performance by anactivity model in parallel DEVS simulationrdquo Journal of ZhejiangUniversity Science C vol 15 no 1 pp 13ndash30 2014
[56] X-L Wang P Yang Z-D Cao et al ldquoQuantitative evaluationon the effectiveness of prevention and control measures againstpandemic influenza A (H1N1) in Beijing 2009rdquoChinese Journalof Epidemiology vol 31 no 12 pp 1374ndash1378 2010 (Chinese)
[57] S Mei A D van de Vijver L Xuan Y Zhu and P M ASloot ldquoQuantitatively evaluating interventions in the influenzaA (H1N1) epidemic on China campus grounded on individual-based simulationsrdquo Procedia Computer Science vol 1 no 1 pp1675ndash1682 2010
[58] X Qiu Z Cao and Z Li Mid-Term Reports of China NationalScience Foundation 91024030 2014
Submit your manuscripts athttpwwwhindawicom
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
MathematicsJournal of
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Mathematical Problems in Engineering
Hindawi Publishing Corporationhttpwwwhindawicom
Differential EquationsInternational Journal of
Volume 2014
Applied MathematicsJournal of
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Probability and StatisticsHindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Journal of
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Mathematical PhysicsAdvances in
Complex AnalysisJournal of
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
OptimizationJournal of
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
CombinatoricsHindawi Publishing Corporationhttpwwwhindawicom Volume 2014
International Journal of
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Operations ResearchAdvances in
Journal of
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Function Spaces
Abstract and Applied AnalysisHindawi Publishing Corporationhttpwwwhindawicom Volume 2014
International Journal of Mathematics and Mathematical Sciences
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
The Scientific World JournalHindawi Publishing Corporation httpwwwhindawicom Volume 2014
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Algebra
Discrete Dynamics in Nature and Society
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Decision SciencesAdvances in
Discrete MathematicsJournal of
Hindawi Publishing Corporationhttpwwwhindawicom
Volume 2014 Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Stochastic AnalysisInternational Journal of
26 Mathematical Problems in Engineering
Conflict of Interests
The authors declare that there is no conflict of interestsregarding the publication of this paper
Acknowledgment
This work was supported in part by the National ScienceFoundation of China (nos 91024030 71303252 61403402and 91324013)
References
[1] W Mao and F-Y Wang Advances in Intelligence and SecurityInformatics Academic Press Oxford UK 2012
[2] E Bonabeau ldquoAgent-based modeling methods and techniquesfor simulating human systemsrdquo Proceedings of the NationalAcademy of Sciences of the United States of America vol 99 no3 pp 7280ndash7287 2002
[3] K M Carley D B Fridsma E Casman et al ldquoBioWar scalableagent-basedmodel of bioattacksrdquo IEEETransactions on SystemsMan and Cybernetics A Systems and Humans vol 36 no 2 pp252ndash265 2006
[4] J M Epstein ldquoModelling to contain pandemicsrdquo Nature vol460 no 7256 p 687 2009
[5] Los Alamos National Laboratory Reports about EPISIMS LosAlamos National Laboratory 2003
[6] C H Builder and S C Bankes ldquoArtificial societies a conceptfor basic research on the societal impacts of informationtechnologyrdquo RAND Report 149 RAND Corp 1991
[7] F-Y Wang ldquoComputational theory and method on complexsystemrdquo China Basic Science vol 6 no 5 pp 3ndash10 2004
[8] F-Y Wang ldquoToward a paradigm shift in social computing theACP approachrdquo IEEE Intelligent Systems vol 22 no 5 pp 65ndash67 2007
[9] F-Y Wang and S Tang ldquoArtificial societies for integratedand sustainable development of metropolitan systemsrdquo IEEEIntelligent Systems vol 19 no 4 pp 82ndash87 2004
[10] D Wen Y Yuan and X-R Li ldquoArtificial societies computa-tional experiments and parallel systems an investigation ona computational theory for complex socioeconomic systemsrdquoIEEE Transactions on Services Computing vol 6 no 2 pp 177ndash185 2013
[11] F Wang F Zeng and Y Yuan ldquoAn ACP-based approach forcomplexity analysis of E-commerce systemrdquo Complex Systemsand Complexity Science vol 5 no 3 pp 1ndash8 2008 (Chinese)
[12] F-Y Wang and P K Wong ldquoResearch commentary intelligentsystems and technology for integrative and predictivemedicinean ACP approachrdquoACMTransactions on Intelligent Systems andTechnology vol 4 no 2 pp 1ndash6 2013
[13] J Sifeng X Gang F Dong and H Chunpeng ldquoStudy on theemergency rescue decision support system of petrochemicalplant based on ACP theoryrdquo in Proceedings of the 6th Manage-ment Annual Meeting (MAM rsquo11) 2011 (Chinese)
[14] X Dong D Fan G Xiong F Zhu Z Zhang and Y Yao ldquoPar-allel Bus Rapid Transit (BRT) operation management systembased on ACP approachrdquo in Proceedings of the 6th ManagementAnnual Meeting (MAM rsquo11) Beijing China 2011
[15] B Ning F-Y Wang H-R Dong R-M Li D Wen and L LildquoParallel systems for Urban rail transportation based on ACP
approachrdquo Journal of Transportation Systems Engineering andInformation Technology vol 10 no 6 pp 23ndash28 2010 (Chinese)
[16] F-Y Wang J Zhao and S-X Lun ldquoArtificial power systemsfor the operation and management of complex power gridsrdquoSouthern Power System Technology vol 2 no 3 pp 1ndash11 2008(Chinese)
[17] M J North T R Howe N T Collier and J R Vos ldquoRepastsimphony development environmentrdquo in Proceedings of theAgent 2005 Conference on Generative Social Processes Modelsand Mechanisms 2005
[18] M J North T RHowe N T Collier and J R Vos ldquoRepast Sim-phony runtime systemrdquo in Proceedings of the Agent Conferenceon Generative Social Processes Models and Mechanisms 2005
[19] SWARM Development Group Swarm 211 Reference Guide2000 httpwwwswarmorg
[20] S Tisue and U Wilensky ldquoNetLogo a simple environment formodeling complexityrdquo in International Conference on ComplexSystems pp 16ndash21 2004
[21] S Luke C Cioffi-Revilla L Panait K Sullivan and G BalanldquoMASON a multiagent simulation environmentrdquo Simulationvol 81 no 7 pp 517ndash527 2005
[22] M Ichikawa H Tanuma Y Koyama andH Deguchi ldquoSOARSintroduction as a social microscope for simulations of socialinteractions and gamingrdquo in Organizing and Learning throughGaming and Simulation Proceedings of the 38th Conference ofthe International Simulation and Gaming Association (ISAGArsquo07) pp 149ndash158 2007
[23] D S Burke J M Epstein D A T Cummings et al ldquoIndividual-based computational modeling of smallpox epidemic controlstrategiesrdquo Academic Emergency Medicine vol 13 no 11 pp1142ndash1149 2006
[24] F-Y Wang ldquoParallel control and management for intelligenttransportation systems concepts architectures and applica-tionsrdquo IEEE Transactions on Intelligent Transportation Systemsvol 11 no 3 pp 630ndash638 2010
[25] Z Molnar D Balasubramanian and A Ledeczi ldquoAn introduc-tion to the genericmodeling environmentrdquo in Proceedings of theTOOLS Europe 2007 Workshop on Model-Driven DevelopmentTool Implementers Forum Zurich Switzerland 2007
[26] S Kelly and J-P Tolvanen Domain Framework in Domain-Specific Modeling Enabling Full Code Generation JohnWiley ampSons Hoboken NJ USA 2007
[27] B P Zeigler H Praehofer and T G Kim Theory of Modelingand Simulation Integrating Discrete Event and ContinuousComplex Dynamic Systems Academic Press San Diego CalifUSA 2000
[28] G Booch J Rumbaugh and I JacobsonThe Unified ModelingLanguage User Guide Addison-Wesley Reading Mass USA1999
[29] F Wagner R Schmuki T Wagne and P Wolstenholme Mod-eling Software with Finite State Machines A Practical ApproachCRC amp Taylor amp Francis Boca Raton Fla USA 2006
[30] T Espiner ldquoChina builds worldrsquos fastest supercomputerrdquo ZDNetUK October 2010
[31] B Chen and G Guo ldquoA two-tier parallel architecture for arti-ficial society simulationrdquo in Proceedings of the ACMIEEESCS26th Workshop on Principles of Advanced and DistributedSimulation (PADS rsquo12) pp 184ndash186 July 2012
[32] G Guo B Chen X G Qiu and Z Li ldquoParallel simulation oflarge-scale artificial society on CPUGPU mixed architecturerdquo
Mathematical Problems in Engineering 27
in Proceedings of the ACMIEEESCS 26th Workshop on Prin-ciples of Advanced and Distributed Simulation (PADS rsquo12) pp174ndash177 July 2012
[33] S-X Lun ldquoResearch on the classification of parallel executionmodes of ACP theoryrdquo Acta Automatica Sinica vol 38 no 10pp 1602ndash1608 2012
[34] F-YWang ldquoA framework for social signal processing and anal-ysis from social sensing networks to computational dialecticalanalyticsrdquo Science China vol 43 no 12 pp 1598ndash1611 2013
[35] G GuoOne Model Userrsquos Manual version 12 National Univer-sity of Defense University 2011
[36] Y Ge R Meng Z Cao X Qiu and K Huang ldquoVirtual citymdashanindividual-based digital environment for human mobility andinteractive behaviorrdquo Simulation Transactions of the Society forthe Modeling and Simulation International 2014
[37] M Kretzschmar and R T Mikolajczyk ldquoContact profiles ineight European countries and implications for modelling thespread of airborne infectious diseasesrdquo PLoS ONE vol 4 no6 Article ID e5931 2009
[38] W Duan and X Qiu ldquoFostering artificial societies using sociallearning and social control in parallel emergency managementsystemsrdquo Frontiers in Computer Science vol 6 no 5 pp 604ndash610 2012
[39] W J Edmunds C J OrsquoCallaghan and D J Nokes ldquoWhomixes with whom Amethod to determine the contact patternsof adults that may lead to the spread of airborne infectionsrdquoProceedings of the Royal Society B Biological Sciences vol 264no 1384 pp 949ndash957 1997
[40] J Banks J S Carson B L Neison and D M Nicol Discrete-Event System Simulation Prentice Hall Englewood Cliffs NJUSA 4th edition 2007
[41] H Yongxia ldquoInvestigation and analysis of college studentsrsquo workand restrdquo Value Engineering vol 31 no 3 2012 (Chinese)
[42] W Duan Z Cao Y Wang et al ldquoAn ACP approach to publichealth emergency management using a campus outbreak ofh1n1 influenza as a case studyrdquo IEEE Transactions on SystemsMan and Cybernetics Part A Systems and Humans vol 43 no5 pp 1028ndash1041 2013
[43] HWHethcote ldquoThemathematics of infectious diseasesrdquo SIAMReview vol 42 no 4 pp 599ndash653 2000
[44] G Chowell E Shim F Brauer and et al ldquoModelling thetransmission dynamics of acute haemorrhagic conjunctivitisapplication to the 2003 outbreak in Mexicordquo Statistics inMedicine vol 25 no 11 pp 1840ndash1857 2006
[45] Y Ge L Liu B Chen X Qiu and K Huang ldquoAgent-basedmodeling for InfluenzaH1N1 in an artificial classroomrdquo SystemsEngineering Procedia vol 2 pp 94ndash104 2011
[46] A R Tuite A L Greer MWhelan et al ldquoEstimated epidemio-logic parameters andmorbidity associatedwith pandemicH1N1influenzardquo CMAJ vol 182 no 2 pp 131ndash136 2010
[47] L Brouwers B Cakici M Camitz A Tegnell and MBoman ldquoEconomic consequences to society of pandemic H1N1influenza 2009mdashpreliminary results for Swedenrdquo Euro Surveil-lance vol 14 no 37 pp 1ndash7 2009
[48] Y Ge L Liu X Qiu H Song Y Wang and K Huang ldquoAframework of multilayer social networks for communicationbehavior with agent-based modelingrdquo Simulation Transactionsof the Society for the Modeling and Simulation International vol89 no 7 pp 810ndash828 2013
[49] W Duan X Qiu Z Cao X Zheng and K Cui ldquoHeterogeneousand stochastic agent-based models for analyzing infectious
diseasesrsquo super spreadersrdquo IEEE Intelligent Systems vol 28 no4 pp 18ndash25 2013
[50] The Preparation and Emergency Response Plans to PandemicPublished by Ministry of Health Ministry of Health 2005(Chinese)
[51] R M Fujimoto Parallel and Distributed Simulation SystemsWiley Interscience 2000
[52] F Cicirelli A Giordano and L Nigro ldquoDistributed simulationof situated multi-agent systemsrdquo in Proceedings of the 15thIEEEACM International Symposium on Distributed Simulationand Real Time Applications (DS-RT rsquo11) pp 28ndash35 September2011
[53] B Chen and G Guo ldquoA two-tier parallel architecturefor artificial society simulationrdquo in Proceedings of the 26thACMIEEESCS Workshop on Principles of Advanced and Dis-tributed Simulation (PADS rsquo12) pp 184ndash186 July 2012
[54] G Guo B Chen and X Qiu ldquoParallel simulation of large-scale artificial society with GPU as coprocessorrdquo InternationalJournal of Modeling Simulation and Scientific Computing vol4 no 2 Article ID 1350005 2013
[55] B Chen L-B Zhang X-C Liu and H Vangheluwe ldquoActivity-based simulation using DEVS increasing performance by anactivity model in parallel DEVS simulationrdquo Journal of ZhejiangUniversity Science C vol 15 no 1 pp 13ndash30 2014
[56] X-L Wang P Yang Z-D Cao et al ldquoQuantitative evaluationon the effectiveness of prevention and control measures againstpandemic influenza A (H1N1) in Beijing 2009rdquoChinese Journalof Epidemiology vol 31 no 12 pp 1374ndash1378 2010 (Chinese)
[57] S Mei A D van de Vijver L Xuan Y Zhu and P M ASloot ldquoQuantitatively evaluating interventions in the influenzaA (H1N1) epidemic on China campus grounded on individual-based simulationsrdquo Procedia Computer Science vol 1 no 1 pp1675ndash1682 2010
[58] X Qiu Z Cao and Z Li Mid-Term Reports of China NationalScience Foundation 91024030 2014
Submit your manuscripts athttpwwwhindawicom
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
MathematicsJournal of
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Mathematical Problems in Engineering
Hindawi Publishing Corporationhttpwwwhindawicom
Differential EquationsInternational Journal of
Volume 2014
Applied MathematicsJournal of
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Probability and StatisticsHindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Journal of
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Mathematical PhysicsAdvances in
Complex AnalysisJournal of
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
OptimizationJournal of
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
CombinatoricsHindawi Publishing Corporationhttpwwwhindawicom Volume 2014
International Journal of
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Operations ResearchAdvances in
Journal of
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Function Spaces
Abstract and Applied AnalysisHindawi Publishing Corporationhttpwwwhindawicom Volume 2014
International Journal of Mathematics and Mathematical Sciences
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
The Scientific World JournalHindawi Publishing Corporation httpwwwhindawicom Volume 2014
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Algebra
Discrete Dynamics in Nature and Society
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Decision SciencesAdvances in
Discrete MathematicsJournal of
Hindawi Publishing Corporationhttpwwwhindawicom
Volume 2014 Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Stochastic AnalysisInternational Journal of
Mathematical Problems in Engineering 27
in Proceedings of the ACMIEEESCS 26th Workshop on Prin-ciples of Advanced and Distributed Simulation (PADS rsquo12) pp174ndash177 July 2012
[33] S-X Lun ldquoResearch on the classification of parallel executionmodes of ACP theoryrdquo Acta Automatica Sinica vol 38 no 10pp 1602ndash1608 2012
[34] F-YWang ldquoA framework for social signal processing and anal-ysis from social sensing networks to computational dialecticalanalyticsrdquo Science China vol 43 no 12 pp 1598ndash1611 2013
[35] G GuoOne Model Userrsquos Manual version 12 National Univer-sity of Defense University 2011
[36] Y Ge R Meng Z Cao X Qiu and K Huang ldquoVirtual citymdashanindividual-based digital environment for human mobility andinteractive behaviorrdquo Simulation Transactions of the Society forthe Modeling and Simulation International 2014
[37] M Kretzschmar and R T Mikolajczyk ldquoContact profiles ineight European countries and implications for modelling thespread of airborne infectious diseasesrdquo PLoS ONE vol 4 no6 Article ID e5931 2009
[38] W Duan and X Qiu ldquoFostering artificial societies using sociallearning and social control in parallel emergency managementsystemsrdquo Frontiers in Computer Science vol 6 no 5 pp 604ndash610 2012
[39] W J Edmunds C J OrsquoCallaghan and D J Nokes ldquoWhomixes with whom Amethod to determine the contact patternsof adults that may lead to the spread of airborne infectionsrdquoProceedings of the Royal Society B Biological Sciences vol 264no 1384 pp 949ndash957 1997
[40] J Banks J S Carson B L Neison and D M Nicol Discrete-Event System Simulation Prentice Hall Englewood Cliffs NJUSA 4th edition 2007
[41] H Yongxia ldquoInvestigation and analysis of college studentsrsquo workand restrdquo Value Engineering vol 31 no 3 2012 (Chinese)
[42] W Duan Z Cao Y Wang et al ldquoAn ACP approach to publichealth emergency management using a campus outbreak ofh1n1 influenza as a case studyrdquo IEEE Transactions on SystemsMan and Cybernetics Part A Systems and Humans vol 43 no5 pp 1028ndash1041 2013
[43] HWHethcote ldquoThemathematics of infectious diseasesrdquo SIAMReview vol 42 no 4 pp 599ndash653 2000
[44] G Chowell E Shim F Brauer and et al ldquoModelling thetransmission dynamics of acute haemorrhagic conjunctivitisapplication to the 2003 outbreak in Mexicordquo Statistics inMedicine vol 25 no 11 pp 1840ndash1857 2006
[45] Y Ge L Liu B Chen X Qiu and K Huang ldquoAgent-basedmodeling for InfluenzaH1N1 in an artificial classroomrdquo SystemsEngineering Procedia vol 2 pp 94ndash104 2011
[46] A R Tuite A L Greer MWhelan et al ldquoEstimated epidemio-logic parameters andmorbidity associatedwith pandemicH1N1influenzardquo CMAJ vol 182 no 2 pp 131ndash136 2010
[47] L Brouwers B Cakici M Camitz A Tegnell and MBoman ldquoEconomic consequences to society of pandemic H1N1influenza 2009mdashpreliminary results for Swedenrdquo Euro Surveil-lance vol 14 no 37 pp 1ndash7 2009
[48] Y Ge L Liu X Qiu H Song Y Wang and K Huang ldquoAframework of multilayer social networks for communicationbehavior with agent-based modelingrdquo Simulation Transactionsof the Society for the Modeling and Simulation International vol89 no 7 pp 810ndash828 2013
[49] W Duan X Qiu Z Cao X Zheng and K Cui ldquoHeterogeneousand stochastic agent-based models for analyzing infectious
diseasesrsquo super spreadersrdquo IEEE Intelligent Systems vol 28 no4 pp 18ndash25 2013
[50] The Preparation and Emergency Response Plans to PandemicPublished by Ministry of Health Ministry of Health 2005(Chinese)
[51] R M Fujimoto Parallel and Distributed Simulation SystemsWiley Interscience 2000
[52] F Cicirelli A Giordano and L Nigro ldquoDistributed simulationof situated multi-agent systemsrdquo in Proceedings of the 15thIEEEACM International Symposium on Distributed Simulationand Real Time Applications (DS-RT rsquo11) pp 28ndash35 September2011
[53] B Chen and G Guo ldquoA two-tier parallel architecturefor artificial society simulationrdquo in Proceedings of the 26thACMIEEESCS Workshop on Principles of Advanced and Dis-tributed Simulation (PADS rsquo12) pp 184ndash186 July 2012
[54] G Guo B Chen and X Qiu ldquoParallel simulation of large-scale artificial society with GPU as coprocessorrdquo InternationalJournal of Modeling Simulation and Scientific Computing vol4 no 2 Article ID 1350005 2013
[55] B Chen L-B Zhang X-C Liu and H Vangheluwe ldquoActivity-based simulation using DEVS increasing performance by anactivity model in parallel DEVS simulationrdquo Journal of ZhejiangUniversity Science C vol 15 no 1 pp 13ndash30 2014
[56] X-L Wang P Yang Z-D Cao et al ldquoQuantitative evaluationon the effectiveness of prevention and control measures againstpandemic influenza A (H1N1) in Beijing 2009rdquoChinese Journalof Epidemiology vol 31 no 12 pp 1374ndash1378 2010 (Chinese)
[57] S Mei A D van de Vijver L Xuan Y Zhu and P M ASloot ldquoQuantitatively evaluating interventions in the influenzaA (H1N1) epidemic on China campus grounded on individual-based simulationsrdquo Procedia Computer Science vol 1 no 1 pp1675ndash1682 2010
[58] X Qiu Z Cao and Z Li Mid-Term Reports of China NationalScience Foundation 91024030 2014
Submit your manuscripts athttpwwwhindawicom
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
MathematicsJournal of
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Mathematical Problems in Engineering
Hindawi Publishing Corporationhttpwwwhindawicom
Differential EquationsInternational Journal of
Volume 2014
Applied MathematicsJournal of
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Probability and StatisticsHindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Journal of
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Mathematical PhysicsAdvances in
Complex AnalysisJournal of
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
OptimizationJournal of
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
CombinatoricsHindawi Publishing Corporationhttpwwwhindawicom Volume 2014
International Journal of
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Operations ResearchAdvances in
Journal of
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Function Spaces
Abstract and Applied AnalysisHindawi Publishing Corporationhttpwwwhindawicom Volume 2014
International Journal of Mathematics and Mathematical Sciences
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
The Scientific World JournalHindawi Publishing Corporation httpwwwhindawicom Volume 2014
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Algebra
Discrete Dynamics in Nature and Society
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Decision SciencesAdvances in
Discrete MathematicsJournal of
Hindawi Publishing Corporationhttpwwwhindawicom
Volume 2014 Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Stochastic AnalysisInternational Journal of
Submit your manuscripts athttpwwwhindawicom
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
MathematicsJournal of
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Mathematical Problems in Engineering
Hindawi Publishing Corporationhttpwwwhindawicom
Differential EquationsInternational Journal of
Volume 2014
Applied MathematicsJournal of
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Probability and StatisticsHindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Journal of
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Mathematical PhysicsAdvances in
Complex AnalysisJournal of
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
OptimizationJournal of
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
CombinatoricsHindawi Publishing Corporationhttpwwwhindawicom Volume 2014
International Journal of
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Operations ResearchAdvances in
Journal of
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Function Spaces
Abstract and Applied AnalysisHindawi Publishing Corporationhttpwwwhindawicom Volume 2014
International Journal of Mathematics and Mathematical Sciences
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
The Scientific World JournalHindawi Publishing Corporation httpwwwhindawicom Volume 2014
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Algebra
Discrete Dynamics in Nature and Society
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Decision SciencesAdvances in
Discrete MathematicsJournal of
Hindawi Publishing Corporationhttpwwwhindawicom
Volume 2014 Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Stochastic AnalysisInternational Journal of