An extended contract net mechanism for dynamic supply ...

Multiagent and Grid Systems – An International Journal 2 (2006) 183–207 183IOS Press

An extended contract net mechanism fordynamic supply chain formation and itsapplication in China petroleum supply chainmanagement

Jiang Tiana, Richard Foleya, Xin Yaob and Huaglory Tianfielda,∗aSchool of Computing and Mathematical Sciences, Glasgow Caledonian University, Glasgow G4 0BA, UKbSchool of Computer Science, University of Birmingham, Birmingham B15 2TT, UK

Received 11 November 2005

Revised 18 January 2006; 7 February 2006; 14 March 2006

Accepted 21 March 2006

Abstract. Software agents representing supply chain partners make it possible to automate supply chain management andparticularly can address the challenging problem of automating the process of dynamic supply chain formation. This paperputs forward an extended contract net mechanism for dynamic supply chain formation and applies it to China petroleum supplychain management, which is characterized by a semi-monopolized market, where conventional negotiation protocols are limitedbecause they are based on the assumption of a pure market. The proposed multi-agent negotiation mechanism is algorithmizedand validated in two scenarios of dynamic supply chain formation, i.e., semi-monopolized market and emergency, respectively.

Keywords: Dynamic supply chain formation, extended contract net mechanism, multi-agent system, supply chain management,multi-agent supply chain management

1. Introduction

Roughly speaking, computerization of enterprisesystems has evolved over three levels. At the basiclevel, departmental business processes and workshop-wide processes in an enterprise, such as design, pro-curement, production, assembling, and distribution, areautomated. For example, Computer Aided Design(CAD) automates the routine activities of design, Mate-rials Requirements Planning (MRP-I) automates mate-rial supply scheduling, Computer-Aided Process Plan-

∗Corresponding author: School of Computing and MathematicalSciences, Glasgow Caledonian University, 70 Cowcaddens Road,Glasgow, G4 0BA, United Kingdom. Tel.: +44 141 331 8025; Fax:+44 141 331 3608; E-mail: [email protected].

ning (CAPP) automates production planning and con-trol, and Customer Relationship Management (CRM)automates the management of customers profiles andthe interdependent relationships. At the middle level,enterprise-wide business processes and workflows areautomated. For example, Manufacturing ResourcePlanning (MRP-II) and Enterprise Resource Planning(ERP) automate the planning and monitoring of var-ious interdependent plans and improve the utilizationof various resources within an enterprise. The recenttrend for computerization of enterprise systems is at thetop level, where the automation of inter-enterprise busi-ness processes is particularly aimed at. Supply chainmanagement (SCM) is one of the most typical inter-enterprise problems. In today’s business environments,enterprises have to focus their own operations in thefields of their core competence. Multiple enterprises in

ISSN 1574-1702/06/$17.00 © 2006 – IOS Press and the authors. All rights reserved

184 J. Tian et al. / An extended contract net mechanism for dynamic supply chain formation

a supply chain have to come together to form a strategicconsortium which as a whole provides customers withwell-packaged products or services. SCM is basicallyto form strategic partnerships and achieve cooperationamong the multiple enterprises in a supply chain.

How to automate SCM processes has remained ahighly challenging problem. While information tech-nologies such as Electronic Data Interchange (EDI)and e-commerce have enhanced the information ex-change and workflow interoperation, enterprise cooper-ation among the multiple enterprises in a supply chainhas not been significantly evolved. Two main reasonsmay be attributable. First, the information infrastruc-ture as required to support the dynamism of SCM [7]is far from flexible, which largely limits the dynamicreconfiguration of a supply chain. Second, SCM stillhas to rely upon rigid, linear and centralized methods,and thus the cooperation between supply chain partnersis not optimized and agile in most cases.

Relationships between supply chain partners can beintricate. Suppliers or customers in a supply chainmay have their own suppliers or customers, and differ-ent supply chains may overlap when the same partnerssimultaneously participate in different supply chains.Overlapping of supply chains results in the partnersin one supply chain acting as competitors for anothersupply chain [2,3,13,33]. Most prominently, relation-ships between supply chain partners are inherently dy-namic. In real-world business environments, supplychain partners are independent enterprises, and havetheir own motivations and goals. Only coming to co-operate with their own interests, supply chain partnersmay join or leave a supply chain based on their ownjudgment without the obligation to remain with the sup-ply chain. Therefore, the cooperation in a supply chainis inherently dynamic [4].

For a complex product/service, the number of sup-pliers involved in a supply chain may reach hundred oreven thousand. Multiplied by the number of hundred oreven thousand of suppliers, the intricate and dynamicrelationships between supply chain partners will veryeasily entail the complexity of SCM to explode into anunmanageable level.

Today’s business environments are fiercely compet-itive. Changes in a market environment are inherentlyunpredictable and instantaneous, with an order as shortas a week, a day or even a couple of hours. In such cir-cumstances, dynamic supply chain formation is muchmore crucial and challenging for SCM than any otherproblems of SCM such as modeling, simulation, andoperation, etc.

Supply chain formation is basically a multi-objectivedecision making and reconfiguration process. Theremay be two scenarios of supply chain formation, i.e.,top-down and bottom-up [31]. In the top-down sce-nario, there is a principal enterprise in charge of theformation, maintenance, management and dissolutionof a supply chain. The principal enterprise selects thepartners strictly under one principle, which is whetherthe partners will increase the profit of the whole supplychain. In the bottom-up scenario, every supply chainpartner only uses local knowledge to form a supplychain. There need to be many interactions betweenpartners and the final partners are determined accord-ing to their own considerations and consensus on costand profit. Through these local demand-supply inter-actions, a supply chain formed in this scenario maynot be the best one, and the process may even end upwith a fail. While the top-down scenario may simulateplanning based economic environments, the bottom-upscenario simulates market or part-market based envi-ronments.

Dynamic supply chain formation is to form a supplychain in real time. It is simply unimaginable how man-ual processes can handle dynamic supply chain for-mation properly in face of unpredictable and instanta-neously changing market environments. For such situ-ations, automation of dynamic supply chain formationis undoubtedly desirable.

In recent years multi-agent system (MAS) has beenrecognized as a promising technology for the automa-tion of SCM. MAS essentially supports decentralizedmethods for SCM and can significantly improve thedynamic reconfigurability of supply chains and theagility of SCM. On the one hand, MAS provide naturalmetaphors for the modeling and simulation of SCM.On the other hand, if each supply chain partner is wellequipped with and trustworthily represented by onesoftware agent (normally a powerful one in terms of ca-pabilities, Sadeh etc. [20] presented a type of softwareagents representing supply chain partners), the SCMcan be fully automated in real time as a multi-agentcooperation problem. In particular, the interactions be-tween agents using an agent communication language,enable supply chain partners to be able to negotiate andcoordinate with one another effectively in real time inan automated manner for dynamic supply chain forma-tion and supply chain operation. Through the nego-tiation process performed among the software agentsrepresenting supply chain partners over a certain pe-riod, different objectives of supply chain partners canbe individually accommodated and various constraints

J. Tian et al. / An extended contract net mechanism for dynamic supply chain formation 185

can be met [1,4,27,28]. In fact, dynamic supply chainformation via automated negotiation is where softwareagents make unprecedented difference to SCM.

Automation of dynamic supply chain formation viaMAS requires an environment comprised of the follow-ing stacks (in a bottom-up order):

1. Inter-enterprise communication infrastructure,including internet, intranet, and/or EDI, etc.

2. E-commerce and workflow interoperation appli-cations, and/or grid computing environment;

3. Agent platform suitable for real-time applica-tions, including software agents, agent manage-ment system (authentication, resources, whitepages (naming)), directory facilitator (yellowpages), and agent communication system (agentcommunication language, content language, on-tology, and agent communication channel), etc.;

4. Agent interaction mechanism (or protocol) suit-able for real-time applications;

5. User interface, e.g., agent-user interface, etc.

This paper is concerned with agent interaction mech-anism suitable for real-time applications in SCM.

Negotiation is a kind of decision making where twoor more participants jointly search for a space of so-lutions with the goal of achieving consensus [4,19,27,28]. Specifically, negotiation is a process by whichtwo or more parties verbalize contradictory demandsand move toward an agreement by a series of conces-sion making and search for new alternatives. This pro-cess involves exchange of information between par-ticipants to reconcile their differences and produce asettlement [17,18,24]. There may be three outcomesfor negotiation: win-win, win-lose, and lose-lose. Thewin-win outcome meets the needs of both parties [5].

For the requirements in real-world supply chains,a negotiation mechanism should consider multiple at-tributes, uncertainties and particularly concession mak-ing upon the original proposals. Only a limited numberof multi-agent negotiation protocols are suitable for dy-namic supply chain formation, including contract netprotocol, third-party negotiation protocol, double bidbased protocol, simulated trading algorithm, and multi-attribute negotiation. Strictly speaking, these protocolsare not in full sense about negotiation, but only for re-source allocation or task assignment, because they in-volve least concession making upon the original pro-posals.

The main drawback of conventional negotiation pro-tocols is that only one bidder is selected in the negoti-ation outcome. To guarantee the existence of dynamic

supply chain formation, at least one supplier has to bechosen in any time by the consumers. Otherwise, thesupply chain will become nonexistent. Therefore, ne-gotiation mechanism should allow auctioneer to choosea ranked list of several candidates as the winning bid-ders. Moreover, all the conventional negotiation pro-tocols are based on the assumption that there is a puremarket for supply chain partners. In the reality, how-ever, supply chain partners sometimes have to subjectthemselves to non-market restrictions such as monop-olization. The conventional negotiation protocols areunable to effectively tackle the SCM problems in suchnon-market circumstances.

This paper puts forward an extended contractnet mechanism for dynamic supply chain forma-tion. Specifically, concession making is incorporatedinto the conventional contract net protocol. Thisstudy is based on a real-world problem domain, i.e.,China petroleum supply chain management (CP-SCM),which is uniquely characterized by a semi-monopolizedmarket.

2. The context and the overall framework of anextended contract net mechanism

Petroleum industry is typical of large-scale, com-plex supply chains. CP-SCM is constrained by themulti-layered administrations which involve the state,provinces, and national corporations. The distinc-tive characteristic of CP-SCM is the semi-monopolizedmarket.

There are three national petroleum corporationswhich monopolize the petroleum supply chains inChina, namely China National Petroleum Corporation(CNPC) [36], China Petroleum and Chemical Corpo-ration (Sinopec) [37], and China National OffshoreOil Corporation (CNOOC) [35]. These monopolis-tic corporations control (or try to control) their com-plete petroleum supply chains ranging from petroleumexploration and exploitation, transportation, refiningand petrochemical processing to distribution. CNPCand Sinopec occupy the upstream and downstreambusinesses in petroleum supply chains, while CNOOConly occupies exploration and exploitation. The com-petitions between the three corporations are limited.There is a rough regional division for petroleum sup-ply among the three corporations, i.e., CNPC controlsthe petroleum supply in North, Northeast, Northwest,and Southwest China, Sinopec controls the petroleumsupply in East, Central, and South China, and CNOOC


Fig. 1. Regional occupations of the three national corporations.

Exploration &Exploitation

Agent

TransportationAgentRefining AgentDistribution AgentCustomer Agent

ImportedPetroleum

Agent

CNPCRefinery

Agent Cluster

CNPCDistrict Sale

Agent Cluster

CNPCOil Field

Agent Cluster

SinopecOil Field

Agent Cluster

CNOOCOil Field

Agent Cluster

SinopecDistrict PetroleumCo. Agent Cluster

SimopecRefinery

Agent Cluster

ProvinceAgent Cluster

Import AgentCluster

Fig. 2. The multi-agent system of China petroleum supply chains.

controls the exploration and exploitation in the coastaland offshore areas of China, as sketched in Fig. 1.

CP-SCM can be modeled as a MAS where the sub-systems of CP-SCM are represented with agents, re-spectively, as shown in Fig. 2. Since the three cor-porations (CNPC, Sinopec and CNOOC) monopolizethe petroleum supply businesses from exploration andexploitation, transportation, refining, to distributions,each agent of the MAS is further expanded with aninstantiation by a cluster of specific agents. Substituteproducts are not considered in the petroleum supply

model, and in fact developing natural gas to substitutepetroleum has a similar supply chain to petroleum.

In exploration and exploitation subsystem, the tasksare carried out by a cluster of oil field agents, i.e., therelevant companies of CNPC, Sinopec and/or CNOOC.Exploration and Exploitation Agent allocates subtasksto these oil field agents, and negotiates with them andmonitors them to complete the exploration and ex-ploitation tasks. Furthermore, the oil fields can be in-stantiated into geology tectonics and even into a num-ber of oil wells. This paper confines itself only to the oil


Exploration & Exploitation Agent

Bo Sea Offshore Oil Field Agent

East Sea Offshore Oil Field Agent

South Sea Offshore Oil Field Agent

Huang Sea Offshore Oil Field Agent

CNOOC Oil Field Agent Cluster

Jianghan Oil Field Agent

Zhongyuan Oil Field Agent

Henan Oil Field Agent

Shengli Oil Field Agent

Jiangsu Oil Field Agent

Northwest Subordinate Co. Agent

Sinopec Oil Field Agent Cluster

Dagang Oil Field Agent

Huabei Oil Field Agent

Liaohe Oil Field Agent

Jilin Oil Field Agent

Daqing Oil Field Agent

Changqing Oil Field Agent

Xinjiang Oil Field Agent

Tarim Oil Field Agent

Tuha Oil Field Agent

Qinghai Oil Field Agent

Yumeng Oil Field Agent

Sichuan Natural Gas Field Agent

Yidong Oil Field Agent

CNPC Oil Field Agent Cluster

Fig. 3. The instantiation architecture of Exploration and Exploitation Agent.

Partner

Consumeri

Supplieri

Demand

Upstream Partner

Consumeri-1

Supplieri-1

Downstream Partner

Consumeri+1

Supplieri+1

Supply

Tier (i+1) Tier i Tier (i-1)

Fig. 4. Consumer-supplier relationships in petroleum supply chains.

field level. Figure 3 depicts the instantiation architec-ture of Exploration and Exploitation Agent. Similarly,Customer Agent, Distribution Agent, Refining Agentand Imported Petroleum Agent can have their instanti-ation architectures to carry out their tasks, respectively.

Since a petroleum supply chain is a systems oper-ation problem, it is impossible to analyze an isolatedprovince in China. A partner enterprise in a petroleumsupply chain is a consumer to upstream partners andat the same time a supplier to downstream partners.The consumer-supplier relationship, including the no-tations for tiers, is depicted in Fig. 4. Petroleum sup-ply chains are simultaneously formed depending onboth consumers and suppliers in the supply chains.For example, for Beijing, Dalian Petrochemical Ltd.in Liaoning Province is ideal petroleum supplier, butDalian Petrochemical Ltd. may be selected by Liaon-

ing Province because the geographical proximity in thelatter case can make lower supply cost.

Petroleum supply chain formation is achieved viamulti-agent negotiation between agents representingsupply chain partners. The formation of petroleum sup-ply chains is not a separated process, but a simultane-ous process in the negotiations between supply chainpartners to obtain or allocate the resource, as shownin Fig. 5. The consumers/suppliers in the same cor-poration are given the priority to form the petroleumsupply chain. The suppliers outside the corporation areconsidered only if the suppliers within the corporationfail to meet the demands of consumers. Furthermore,if domestic petroleum supply fails to meet the demandsfor petroleum, imported petroleum will be considered.Imported petroleum will be used to meet different re-gions’ demands according to the shorter transportation


Consider and Choose Bids fromSuppliers within Same Corporation

Y

SupplierPartners

ConsumerPartner

Processing

N

Decision

Partners

The Supplyin Corporation

Sufficient?

Consider and Choose Bids fromSupplier outside Corporation

Cal

l for

pro

posa

l

Res

pons

e fo

r the

pro

posa

l

Consider and Choose Bids from Import Supplier

Y

N

Domestic SupplySufficient?

Fig. 5. Dynamic formation of China petroleum supply chains.

CNPCOil Fields

CNPC�sCustomers

CNPCDistribution

SinopecDistribution

CNPCRefineries

SinopecRefineries

SinopecOil Fields

ImportedCrude Oil

CNOOCOil Fields

Supply

Demand

Sinepec�sCustomers

Fig. 6. Priority criterion for semi-monopolized market.

distances and the infrastructure conditions. For exam-ple, imported crude oil from Kazakstan and Russia canbe used to meet the demands in Northeast and North-west China because they are shorter transportation dis-tances and there are pipelines and railways in place,while imported petroleum from Middle East and Africacan be used to meet the demands in East and SouthChina, because it is shorter than in North China andthere are harbor infrastructures for imported petroleumby tanker fleets.

Basically there are two criteria for multi-agent nego-tiation in CP-SCM, i.e., priority to meet the demands ofthe consumers in the same corporation, and then min-imum cost to meet the demands of all the consumers.Suppliers in the same corporation are given priority to

those outside the corporation. Because Sinopec has avery high level of demand for crude oils, and CNOOC’soil fields are adjacent to Sinopec’s refineries, CNOOCbecomes an important upstream supplier for Sinopec.The priority criterion for semi-monopolized market isdepicted in Fig. 6.

The second negotiation criterion in CP-SCM is tominimize the overall cost in a supply chain. The sup-ply distance is an important factor contributing to thesupply chain wide cost. Petroleum companies even inthe same corporation are geologically dispersed acrossChina. If supply is excessive to demand, only the ad-jacent suppliers are selected. On the contrary, if de-mand is excessive to supply, the distant suppliers maybe considered depending on the balance of the cost in-


curred by distance. For example, Daqing Petrochem-ical Co. firstly requests Daqing Oil Field to supplycrude oil because the crude oil pipeline between is theshortest. If Daqing Oil Field can not meet DaqingPetrochemical Co.’s demands, distant oil fields suchas Liaohe Oil Field and Jilin Oil Field can be consid-ered by Daqing Petrochemical Co. Moreover, differentpetroleum products or crude oils with different proper-ties will also result in different supply costs. For exam-ple, the crude oils in different oil fields, which have dif-ferent properties, can cause different exploitation costs.

A multi-agent negotiation mechanism, based on theconventional contract net protocol, is put forward here.The mechanism takes account of the domain-specificrequirements in CP-SCM, and comprises four mainsteps as follows.

Step 1: Consumers/suppliers calculate their de-mands or supply-capacities and announcethe requests for bids;

Step 2: Considerations in the negotiation are notonly the quantitative factors such as sup-plier’s price and delivery date, but also thestrategic factor such as collaborative part-nership. For example, CNOOC is an impor-tant crude oil supplier for Sinopec becausethey have collaborative partnership in crudeoil supply;

Step 3: Both consumers and suppliers have to agreeon bids for their own benefits. For exam-ple, consumers will select the supplier en-tailing lower procurement price, while sup-pliers may agree with the consumers makinghigher demand quantity. This is a mutualprocess, which is not used in the conven-tional contract net protocol.

Step 4: After a negotiation round, consumers/supp-liers can revise their negotiation contents soas to settle the bids in the next negotiationround.

The proposed multi-agent negotiation mechanismfor dynamic supply chain formation in CP-SCM is fur-ther depicted in Fig. 7.

The proposed multi-agent negotiation mechanismare algorithimized and validated below in two sce-narios of dynamic supply chain formation, i.e., semi-monopolized market, and emergency of petroleum sup-ply, respectively. Semi-monopolized market is the mostdistinctive characteristic of CP-SCM, and emergency isa challenging case because existing petroleum supplychain in China is very weak and fragile to deal withemergency events.

3. Algorithmization of multi-agent negotiationmechanism for dynamic supply chain formationin a semi-monopolized market

Semi-monopolized market is a practical problem inCP-SCM. There are several ways to solve this prob-lem. For example, the announcements are only de-clared to a certain corporation agents; or by definingcorporation relationship between agents the negotia-tion protocol is only allowed to be adopted for thiscorporation relationship; or the negotiation protocol isonly adopted for predetermined agents. In this paperthe solution for semi-monopolized market is to restrictthe announcement only to certain corporation agents.Consumers and suppliers in the MAS announce theirdemands/supply-capacities within the corporation, andthe suppliers/consumers outside the corporation willnot be notified, as depicted in Fig. 8.

The multi-agent negotiation process for dynamicsupply chain formation in CP-SCM in a semi-monopolized market is designed as follows.

Step 1: Set i from 5 to 1. For Tier i, repeat steps2–8.

Step 2: Tier i and Tier i − 1 agents calculate theirdemands/supply-capacities, respectively. If the capac-ity of a Tier i − 1 supplier is approaching nil, the sup-plier quits the negotiation, even if it has a competitivesupply price.

Sept 3: Tier i consumers announce their demandswithin the same corporation. A request for bids con-tains product type, demand date, demand quantity, anddemand price, etc.

Step 4: Tier i − 1 suppliers formulate and submitbids to the requesting Tier i consumers according to thedemands. The supply price pi−1

supply of a Tier i− 1 sup-plier can be calculated by considering its procurementcost ci−1

procure, and its processing cost ci−1process including

refining and storing. In the simple case,

pi−1supply ⇐ ci−1

procure+ci−1process

qi−1,i−2

= pi−1procure +

ci−1process

qi−1,i−2

⎫⎬⎭ (1)

where qi−1,i−2 is the quantity of goods that the Tier i−1agent procures and processes. The cost relationshipsbetween adjacent tiers are depicted in Fig. 10. OnlyTier i − 1 suppliers with the required product typesubmit bids to the requesting consumers. For example,if a refinery requests heavy crude oil, oil fields withlight crude oil will not submit bids to this refinery.

If no Tier i − 1 suppliers submit bids, Go back toStep 3, Tier i consumers re-announce their demandsbut outside the corporation.


Consumers Suppliers

Agree on Pre-Offer

Provide Agreed Supply

Notify of Pre-Offer

Submit Bids

Announce Demands

Formulate Bids

Revise Supply Status

Consider Bids

Revise Demand Status

Consider Pre-Offer

Calculate Demands Calculate Supplies

Tier i Tier i-1

Confirm Offer

Fig. 7. The multi-agent negotiation mechanism for dynamic supply chain formation.


Agent

TransportationAgent

RefiningAgent

DistributionAgent

ConsumerAgent

CNPCDistribution

Agents

SinopecDistribution

Agents

CNPCRefineryAgents

SinopecRefineryAgents

CNPCOil FieldAgents

SinopecOil FieldAgents

CNPCTransportation

Agents

SinopecTransportation

Agents

CNPCProvinceAgents

CNOOCOil FieldAgents

ImportedPetroleum

Agent

SupplyDemand

SinopecProvinceAgents

Tier 5 Tier 4 Tier 3 Tier 2 Tier 1

Fig. 8. The multi-agent negotiation in a semi-monopolized market.

Step 5: Tier i consumers consider the bids receivedfrom Tier i − 1 suppliers.

(5.1) Tier i consumer pre-checks the bids receivedfrom Tier i-1 suppliers.

(a) If the product type does not match the require-ment, reject the bid. Otherwise, proceed withthe bid.

(b) If the supply date does not precede the demanddate, reject the bid. Otherwise, proceed with thebid.

(c) If the supply quantity is less than the demandquantity, reject the bid or let the human operator

decide. Otherwise, proceed with the bid.(d) If there is collaborative partnership between the

consumers and the supplier, let the human oper-ator decide. Otherwise, proceed with the bid.

(5.2) The Tier i procurement price p iprocure is cal-

culated by considering the supply price p i−1supply of the

Tier i − 1 supplier, and related costs such as the trans-portation cost from the Tier i− 1 suppliers to the Tier i

consumer. A transportation price can be calculated as

ci,i−1transport = di,i−1 × li,i−1 (2)


Tier i consumer announcesrequest for bids within the corporation

Tier i-1 supplier formulates bid, includingsupply quantity, supply date, supply price

Tier i-1 supplier submits bid to Tier i consumer

Tier i consumer pre-checks bids received, and calculates procurement prices

Tier i consumer selects the Tier i-1 supplier whomakes the lowest procurement price

for Tier i consumer, and notify of pre-offer

Tier i-1 supplier considers pre-offer, agreeswith the consumer with higher demand quantity

Selected Tier i-1 supplier provides agreedconsumer with agreed supply

Update demand/supply status

Tier i=i-1

i = 6. Initialize

Print out negotiation result at Tier i: product, supplier, quantity, date, costs

Yes

No

No

Yes

End

Tier i-1 supplier quits the negotiation

Yes

No

i=1

1,11

1,

−−−

−

−

−iii

SiS

iiiD

iD

qQQ

qQQ

5,,1,,5 L=jQQ jSD

0>iDQ

01 >−iSQ

Fig. 9. The negotiation process of dynamic supply chain formation in a semi-monopolized market.

where di,i−1 is the supply distance from Tier i−1 sup-plier to Tier i consumer, and l i,i−1 is the transportationprice. In the simple case,

piprocure ⇐ pi−1

supply +ci,i−1transport

qi,i−1(3)

By bringing Eq. (1) into Eq. (3), there is

piprocure ⇐ pi−1

procure +ci−1process

qi−1,i−2+

ci,i−1transport

qi,i−1(4)

See Fig. 10 for the cost relationships between tiers.(5.3) Arrange Tier i − 1 suppliers in an ascending

order according to the Tier i consumer’s procurementprices with regard to them. Select the Tier i−1 supplierthat makes the lowest procurement price for the Tier iconsumer.


1−iprocessc

1−iprocurec i

procurec1sup−i

plyp

iDQ1−i

DQ

1−iSQ

Tier (i-2) Tier (i-1) Tier i

2−iprocessc

2−iSQ

iprocessc

iSQ

iplypsup

2sup−i

plyp

1, −iiq2,1 −− iiq

Demand Demand

Fig. 10. Cost relationships between tiers.

Then Tier i consumer notifies the selected Tier i− 1supplier of the pre-offer.

If there is not a proper Tier i − 1 supplier for Tier iconsumer to select, go back to Step 3, Tier i consumersre-announce their demands but outside the corporation.

Step 6: The selected Tier i − 1 supplier considersthe pre-offers from Tier i consumers.

(6.1) If there is collaborative partnership betweenthe Tier i − 1 supplier and the Tier i consumer, let thehuman operator decide. Otherwise, proceed with thepre-offer.

(6.2) If there is more than one Tier i consumer ac-cepting the supply price, the Tier i consumer that hashigher demand quantity is to be agreed with so that thesupplier can sell more products. Otherwise, proceedwith the pre-offer.

(6.3) Agree with the Tier i consumer on the pre-offer.Step 7: The agreed Tier i consumer confirms the of-

fer to the selected Tier i−1 supplier. The procurementcost of the Tier i consumer can be calculated

ciprocure = pi

procure × qi,i−1 (5)

where qi,i−1 is the supply quantity from the Tier i − 1supplier. Print out the negotiation results.

Then the selected Tier i − 1 supplier provides theagreed Tier i consumer with the agreed supply at theagreed supply price.

Step 8: Tier i, Tier i − 1 agents update their de-mand/supply status after a successful negotiation, orthe unselected agents in the negotiation revise their de-mand/supply status.

The negotiation process can be diagramed as inFig. 9, and further depicted by state machines of agentsin a petroleum supply chain, as in Fig. 11.

The negotiations between supply chain partners si-multaneously take place at each tier of a petroleum sup-

ply chain. There are many rounds to find out the sup-plier that makes the lowest procurement price. Oncea round is complete, the demand/supply status will beupdated, e.g., the reduced demands and supply capac-ities. The negotiation processes at Tier i will not stopuntil the demands of all the Tier i consumers are fullymet, and the negotiation process in a supply chain willnot stop until the demands at all tiers are completelymet.

4. Algorithmization of multi-agent negotiationmechanisms for dynamic supply chainformation in emergency

In emergency events, e.g., petroleum pipeline isbroken, in order to meet the emergent demands forpetroleum and at the same time to provide steadypetroleum supplies in normal regions, first, the pro-curement time instead of the procurement cost becomesthe crucial factor for the multi-agent negotiation in apetroleum supply chain; second, the monopolistic re-strictions between different corporations should be dis-regarded, and agents in different corporations are in apure market, as shown in Fig. 12.

Supply distance is quite influential because theshorter a supply distance, the less the delivery time.First of all, the consumers can be supplied by the sup-pliers within the same province/city because this is thequick way for supply. As there is no more monopo-listic restriction between different corporations, the de-mands of the consumers will be announced to all agentsin a petroleum supply chain, and the agents from dif-ferent corporations will have the equal opportunity tosubmit bids. After all, the procurement cost will beconsidered, especially when two suppliers entail near


Costumer Agent

Announce

Consider bids

Update status

Startup

Distribution Agent

Announce

Agree on pre-offer & get offer

Consider bids

Distribute petroleum

Formulate bidStartup

Select supplier & notify of pre-offerPre-offer

Startup

Imported Petroleum Agent

Formulate bid

Supply crude oil

Bid

Demand

Select supplier & notify of pre-offer

Update status

Startup

Refining Agent

Formulate bid


Consider bids

Supply petroleum

Announce


Startup

Transportation Agent

Formulate bid


Consider bids

Transport crude oil

Announce


Bid

Demand

Pre-offer

Demand

Bid

Pre-offer

Update status


Bid

Pre-offer

Demand

Update status

Startup


Formulate bid

Supply crude oil


Bid

Pre-offer

Update status

Demand

Tier 3 Tier 2

Tier 1 Tier 1

Tier 5Tier 4

Fig. 11. The state machine graph of the negotiation for dynamic supply chain formation in a semi-monopolized market.

procurement dates for the consumer (e.g., when theyare located in the same city). When a supplier con-siders pre-offers, the consumer that requests a lowerquantity can be agreed with because resource is limitedin emergency.

In emergency, the procurement date becomes cru-cial, which is determined by processing time, includ-ing storing and wrapping, and the transportation timein a petroleum supply chain. The transportation timedepends upon the transportation speed. The transporta-tion speeds by different modes, i.e., pipeline, railway,tanker fleet, and truck, can be calculated as

timei,i−1transport =

di,i−1

300/λ(6)

where di,i−1 is the supply distance from Tier i − 1supplier to Tier i consumer, and λ is the transportationspeed coefficient. The standard (λ = 1) transportationspeed is assumed as 300 km/day by railway, and λ is as-sumed as 0.8, 1.2, and 1.4 by pipeline, truck and tankerfleet, respectively. Although different transportationmodes have their own supply distances, in this paper,supply distances are uniformly approximated using thegeographical survey coordinates of supply nodes.

The multi-agent negotiation process for dynamicsupply chain formation in emergency is designed asfollows.

Step 1: Set i from 5 to 1. For Tier i, repeat steps2–8.



Agent

TransportationAgent

RefiningAgent

DistributionAgent

ConsumerAgent

CNPCDistribution

Agents

SinopecDistribution

Agents

CNPCRefineryAgents

SinopecRefineryAgents

CNPCOil FieldAgents

SinopecOil FieldAgents

CNPCTransportation

Agents

SinopecTransportation

Agents

ProvinceConsumer

Agent

CNOOCOil FieldAgents

ImportedPetroleum

AgentSolid Arrow: Supply FlowDashed Arrow: Demand Flow

Tier 5 Tier 4 Tier 3 Tier 2 Tier 1

Fig. 12. The multi-agent negotiation in emergency.

Step 2: Tier i and Tier i − 1 agents calculate theirdemands/supply-capacities, respectively. The emer-gency circumstance is taken into consideration, e.g.,processing capacity may decrease because of the pro-longed repair of a refinery. If the capacity of a Tieri − 1 supplier is approaching nil, the supplier quits thenegotiation, even if it has a desirable supply date and/ora competitive supply price.

Sept 3: Tier i consumers announce their demands toTier i−1 suppliers. A request for bids contains producttype, demand date, demand quantity, and demand price,etc.

Step 4: Tier i − 1 suppliers formulate and submitbids to the requesting Tier i consumers according to thedemands. A bid contains product type, quantity, supplydate and supply price, etc. The supply price p i−1

supply ofa Tier i − 1 supplier can be calculated by consideringthe procurement cost ci−1

procure, and the processing costci−1process including refining, storing and wrapping, as

in Eq. (1). The supply date datei−1supply reflects the

processing time. In the simple case,

datei−1supply = datei−1

procure(7)

+timei−1proccess/store/wrap

Step 5: Tier i consumers consider the bids receivedfrom Tier i − 1 suppliers.

(5.1) Tier i consumer pre-checks the bids receivedfrom Tier i − 1 suppliers.

(a) If the product type does not match the require-ment, reject the bid. Otherwise, proceed withthe bid.

(b) If the supply quantity is less than the demandone, reject the bid or let the human operatordecide. Otherwise, proceed with the bid.

(5.2) The procurement date date iprocure of a Tier i

consumer is calculated by considering the supply datedatei−1

supply of the Tier i − 1 supplier, and the trans-portation time from the Tier i− 1 supplier to the Tier iconsumer, timei,i−1

transport. In the simple case,

dateiprocure = datei−1

supply + timei,i−1transport (8)

By bringing Eq. (7) into Eq. (8), there is

dateiprocure = datei−1

procure

+timei−1proccess/store/wrap (9)

+timei,i−1transport

(5.3) If there is more than one Tier i−1 supplier withnear supply date datei−1

supply , choose the supplier thatmakes the earliest procurement date datei

procure for theTier i consumer, which, in terms of Eq. (8), impliesthe Tier i − 1 supplier in the quickest transportation.Otherwise, proceed with the bid.

(5.4) The procurement price p iprocure of a Tier i con-

sumer is calculated by considering the supply pricepi−1supply of the Tier i − 1 supplier, and the related costs

such as transportation costs from the Tier i−1 supplierto the Tier i consumer, as shown in Eqs (3) and (4).

(5.5) Arrange Tier i − 1 suppliers in an ascendingorder according to the Tier i consumer’s procurementdates and prices with regard to them. Select the Tieri− 1 supplier that makes the earliest procurement dateand lower procurement price pi

procure for the Tier iconsumer.

Then Tier i consumer notifies the selected Tier i− 1supplier of the pre-offer.

Step 6: The selected Tier i − 1 supplier considersthe pre-offer.


Tier i consumer announces requestfor bids to all Tier i-1 suppliers

Tier i-1 supplier formulates bid, includingsupply quantity, supply date, supply price

Tier i-1 supplier submits bidto Tier i consumer

Tier i consumer pre-checks bids receivedand calculates procurement dates and

procurement prices

Tier i consumer selects the Tier i-1 supplier that makes the earliest procurement date, or makes the earliestprocurement date and lower procurement price

for Tier i consumer. Notify of pre-offer

Tier i-1 supplier considers pre-offer, agreeswith the consumer with lower demand quantity

Selected Tier i-1 supplier provides agreed consumerwith agreed supply before agreed supply date

Update demand/supply status

Tier i = i-1

i=1

i=6. Initialize

Print out negotiation result at Tier i: product, supplier, quantity, date, costs

Yes

NoNo

Yes

End

Yes

NoTier i-1 supplier quits the negotiation

5,,1,,5 L=jQQ jSD

0>iDQ

01 >−iSQ

1,11

1,

−−−

−

−

−iii

SiS

iiiD

iD

qQQ

qQQ

Fig. 13. The negotiation process of dynamic supply chain formation in emergency.

(6.1) If there is more than one Tier i consumer ac-cepting the supply date of the Tier i−1 supplier, the Tieri consumer that has lower demand quantity is agreed

with because resource is limited in emergency. Other-wise, proceed with the pre-offer.

(6.2) Agree with the Tier i consumer on the pre-offer.


Costumer Agent

Announce

Updated status

Startup

Distribution Agent

Announce

Consider bids

Distribute petroleum

Formulate bidStartup

Pre-offer

StartupImported Petroleum Agent

Formulate bid

Supply crude oil

Bid

Demand

Select supplier& notify of pre-offer

Updated status

Startup

Refining Agent

Formulate bid

Supply petroleum

Announce Startup

Transportation Agent

Formulate bid

Transport crude oil

Announce

Bid

Demand

Pre-offer

Demand

Bid

Pre-offer

Updated status

Agree on pre-offer& get offer

Bid

Pre-offer

Demand

Updated status

Startup


Formulate bid

Supply crude oil


Bid

Pre-offer

Updated status

Demand

Agree on pre-offer& get offerAgree on pre-offer

& get offer





Consider bids

Consider bidsConsider bids

Tier 3 Tier 2

Tier 1 Tier 1

Tier4 Tier 5

Fig. 14. The state machine graph of the negotiation for dynamic supply chain formation in emergency.

Step 7: The agreed Tier i consumer confirms the of-fer to the selected Tier i−1 supplier. The procurementcost ci

procure of the Tier i consumer can be calculatedas in Eq. (5). Print out the negotiation results.

Then the selected Tier i − 1 supplier provides theagreed Tier i consumer with the agreed supply beforethe agreed supply date and at the agreed supply price.

Step 8: Tier i, Tier i − 1 agents update their de-mand/supply status after a successful negotiation, orthe unselected agents in the negotiation revise their de-mand/supply status.

The negotiation process can be diagramed as inFig. 13, and further depicted by state machines as inFig. 14. The negotiation processes at Tier i will not stop

until the demands of all the Tier i consumers are fullymet, and the negotiation process in a supply chain willnot stop until the demands at all tiers are completelymet.

5. Case studies

A petroleum supply chain in Northeast China issimulated to demonstrate the proposed negotiationmechanism for dynamic supply chain formation in asemi-monopolized market and an emergency scenario.Northeast China region is one of the main petroleumsupply foundations in China. Monopolization and im-


Table 1Demands/supply-capacities of customer, refinery and oil field agents (104 tonnes)

CustomerAgent

PetroleumProductDemanded

RefineryAgent

PetroleumProductSupplied

ProcessingCrude OilDemanded

Oil FieldAgent

Crude OilSupplied

HeilongjiangProvince Agent

660 DaqingPetrochemicalAgent

433.96 602.72 Daqing OilField Agent

4640

Jilin ProvinceAgent

450 FushunPetrochemicalAgent

666.12 925.17 Liaohe OilField Agent

1435.65

LiaoningProvince Agent

1170 LiaoyangPetrochemicalAgent

540 750 Jilin OilField Agent

505.5

JinxiPetrochemicalAgent

432 600 CNOOCBo SeaOil Field

691.51

JinzhouPetrochemicalAgent

396 550 RussiaImportedAgent

1077

DalianPetrochemicalAgent

511.2 710

Total 2280 2979.28 4137.89 8349.66

(Source: CNPC: http://www.cnpc.com.cn/, CNOOC: http://www.cnooc.com.cn/).

Refining Agent

CNPC Refineries Agent Cluster

Liaoyang Petrochemical Agent

Fushun Petrochemical Agent

Daqing Petrochemical Agent

Jinzhou Petrochemical Ltd. Agent

Jinxi Petrochemical Agent

Dalian Petrochemical Agent

Exploration and Exploitation Agent

Liaohe Oil Field Agent

Jilin Oil Field Agent

Daqing Oil Field Agent

CNPC Oil Field Agent Cluster

Customer Agent

Customer Province Agents Cluster

Heilongjiang Province Agent

Jilin Province Agent

Liaoning Province Agent

Russia Imported Agent

CNOOC BoSea Oil Field Agent

Tier 3 Tier 2 Tier 1

Fig. 15. The multi-agent system of petroleum supply chain in Northeast China.

ported petroleum coexist in this region. Most of thepetroleum companies in this region are controlled byCNPC, while a few of them are import companies andfrom CNOOC. The MAS for simulation is constructed,as shown in Fig. 15. The demands and supply capac-ities of agents are shown in Table 1, and the locationsof individual agents are listed in Table 2. To simplifythe analysis without loss of correctness, the followingassumptions are made for the simulation of multi-agentnegotiation for dynamic supply chain formation in CP-SCM.

Assumption 1: Only three supply tiers are consid-ered in the simulation, i.e., Consumer, Refining, and

Exploitation and Exploration (or Imported Petroleum).Assumption 2: The final product is the Petroleum

Product, and the original material is the Crude Oil. Allpetroleum products or crude oils are available in anyrefinery or oil field, respectively.

Assumption 3: There is no price fluctuation of prod-ucts and services in the petroleum supply chain.

Assumption 4: The term of supply is one year. All theproduction capabilities, and transportation capabilitiesare based on a year. The settings for the simulation ofthe petroleum supply chain are the actual statistics in2004.


Table 2Locations of customer, refinery and oil field agents

City of agent X (km) Y (km)

Daqing, Daqing Oil Field Agent and Daqing Petrochemical Agent 13908.3 5842.6Jinzhou, Jinzhou Petrochemical Agent 13550 4700Fushun, Fushun Petrochemical Agent 13800 5100Dalian, Dalian Petrochemical Agent 13525 4675Liaoyang, Liaoyang Petrochemical Agent 13711.3 5023.5Jinxi, Jinxi Petrochemical Agent 13450 4950Panjin, Liaohe Oil Field Agent 13586.7 5010.8Songyuan (Fuyu), Jilin Oil Field Agent 13900 5625Tianjin, Bo Sea Oil Field Agent 13045 4713.1Angarsk, Russia Import Agent 11575 6875Shenyang, Liaoning Province Agent 13725 5100Changchun, Jilin Province Agent 13950 5425Harbin, Heilongjiang Province Agent 14100 5700

(Source: http://www.multimap.com/).

Fig. 16. Screenshot of agent types in Zeus.

Assumption 5: The refining rate for transferringcrude oils into petroleum products is 72%, the utiliza-tion rate of the processing capabilities in refineries is100%, and the loss rate in transportation and distribu-tion is 5%.

Assumption 6: The supply nodes are the individualpetroleum companies, and the distances between sup-ply nodes are calculated using the geographical surveycoordinates of the cities where petroleum companiesare located. The geographical distances are calculatedby function f(x, y) =

√(x2 − x1)2 + (y2 − y1)2,

where (x1, y1) and (x2, y2)are the geographical surveycoordinates of cities where the supply chain partnersare located.

Assumption 7: There is sufficient infrastructure fortransportation and distribution of petroleum or crudeoil.

The simulation for semi-monopolized market is im-plemented in Zeus 1.1, an agent platform developed byBritish Telecommunication plc. [34]. The agent typesin the system are shown in Fig. 16.

In a semi-monopolized market, all partners are re-


Fig. 17. Screenshot of the agent society in a semi-monopolized market.

stricted to supply chains within the corporation. Thusthe BoSeaOilField agent is excluded from the nego-tiation in Exploration and Exploration supply tier, asshown in Fig. 17. The statistics of the interaction be-tween agents is shown in Fig. 18.

Under the assumptions for simulation, the trans-portation distance is the only negotiation factor for pro-curement cost in a semi-monopolized market. Accord-ing to the negotiation algorithm, the shorter a trans-portation distance, the less the procurement cost, soa supplier with shorter transportation distance shouldfirst be selected. The negotiation outcomes betweencustomer, refinery and oil field agents in a semi-monopolized market are shown in Tables 3 and 4. The

deals leave remaining550.31×104 tonnes of petroleumproduct, of which 485.64×104 tonnes are from DalianPetrochemical Agent, and 64.67 × 104 tonnes fromJinzhou Petrochemical Agent; and 3, 137.36 × 104

tonnes of crude oil, of which 2, 114.21 × 104 tonnesare from Daqing Oil Field Agent, and 1, 023.15× 104

tonnes from Import Agent from Russia.In emergency, all partners in the supply chain can

negotiate with each other as monopolization is disre-garded. Thus the BoSeaOilField agent can participatein the negotiation in Exploration and Exploration sup-ply tier, as shown in Fig. 19. The statistics of the in-teractions between agents in emergency is shown inFig. 20.


Table 3Negotiation outcome between customer and refinery agents

Consumer Supplier Petroleum Product TransportationSupply (104 tonnes) Distance (km)

Liaoning Province Agent Fushun Petrochemical Agent 632.81 75Liaoning Province Agent Liaoyang Petrochemical Agent 513 77.72Helongjiang Province Agent Daqing Petrochemical Agent 412.26 238.92Liaoning Province Agent Jinxi Petrochemical Agent 24.19 313.25Jilin Province Agent Jinxi Petrochemical Agent 386.21 689.66Jilin Province Agent Jinzhou Petrochemical Agent 63.79 828.02Helongjiang Province Agent Jinzhou Petrochemical Agent 247.74 1,141.27

Fig. 18. Screenshot of the interaction statistics between agents in a semi-monopolized market.

In the simulation in emergency, under the assump-tions for simulation, the shorter a transportation dis-tance, the less the procurement time. Therefore, thesupplier with shorter transportation distance is firstlyselected. The negotiation outcomes between customer,refinery and oil field agents in emergency are similarto those in a semi-monopolized market. The main dif-ference in the negotiations is that the suppliers outsidethe monopolized corporation are selected, because themonopolization is disregarded in emergency, as shownin Table 5. The deals leave remaining 3, 828.87× 104

tonnes of crude oil, of which 2, 805.72 × 104 tonnes

are from Daqing Oil Field Agent, and 1, 023.15× 104

tonnes from Import Agent from Russia.

6. Related works

The conventional negotiation protocols suitable forsupply chain formation include contract net protocol,third-party negotiation protocol, double bid based pro-tocol, simulated trading algorithm, and multi-attributenegotiation, etc.


Table 4Negotiation outcome between refinery and oil field agents in a semi-monopolized market


Daqing Petrochemical Agent Daqing Oil Field Agent 602.72 1Liaoyang Petrochemical Agent Liaohe Oil Field Agent 750 125.25Jinxi Petrochemical Agent Liaohe Oil Field Agent 600 149.61Fushun Petrochemical Agent Liaohe Oil Field Agent 13.87 231.2Fushun Petrochemical Agent Jilin Oil Field Agent 480.23 534.44Fushun Petrochemical Agent Daqing Oil Field Agent 431.07 750.46Jinzhou Petrochemical Agent Daqing Oil Field Agent 550 1,197.46Dalian Petrochemical Agent Daqing Oil Field Agent 710 1,228.91

Fig. 19. Screenshot of the agent society in emergency.

6.1. Contract net protocol

A contract net protocol achieves cooperation throughtask sharing or task assignment in networks of com-municating problem solvers [21,22]. In a contract net

protocol a manager announces a task to a set of con-tractors. Based on their local cost estimates, the con-tractors formulate bids, and send them to the manager.The manager selects the best bid, rejects the others andgrants the task to the best bidder. This contractor is


Table 5Negotiation outcome between refinery and oil field agents in emergency


Daqing Petrochemical Agent Daqing Oil Field Agent 602.72 1Liaoyang Petrochemical Agent Liaohe Oil Field Agent 750 125.25Jinxi Petrochemical Agent Liaohe Oil Field Agent 600 149.61Fushun Petrochemical Agent Liaohe Oil Field Agent 13.87 231.2Dalian Petrochemical Agent Bo Sea Oil Field Agent 691.51 481.51Fushun Petrochemical Agent Jilin Oil Field Agent 480.23 534.44Fushun Petrochemical Agent Daqing Oil Field Agent 431.07 750.46Jinzhou Petrochemical Agent Daqing Oil Field Agent 550 1,197.46Dalian Petrochemical Agent Daqing Oil Field Agent 18.49 1,228.91 km

Fig. 20. Screenshot of the interaction statistics between agents in emergency.

committed to reporting the success or failure of the ex-ecution to the manager [6]. For a simple 2-tier supplychain, a contract net protocol can readily fit to a con-tract manager and its contractors. For multi-tier supplychains, a contractor may further announce requests forbids and award contracts to its own suppliers [14,15].

A contract net protocol is a simple method to find anallocation of a set of tasks to an agent society. However,since agents act independently during their decisionprocesses, the plans only become locally optimized [6].

One drawback of (extended) contract net protocol isthat the results are dependent upon the incoming orderof tasks. By changing the incoming order of tasks,the plans of agents can be completely different as wellas the result of the overall solution [25]. Contractsin contract net protocol are “binding” for the partnersinvolved, and there is no flip-flop of decisions oncecontracts are made. It is difficult to adjust the contractsto accommodate dynamic changes such as the arrivalsof urgent new orders [15].


6.2. Third-party negotiation protocols

A third-party negotiation protocol, which is alsocalled an auction protocol, can be described as themechanism between a general contractor and many spe-cialty contractors through an auctioneer. In a third-party negotiation protocol, the general contractor agent(seller) starts the negotiation by sending a message tothe auctioneer. This message contains the tasks thatit wants to sell, the highest desired price, and the pre-ferred auction type. After receiving the message, theauctioneer will broadcast it to the specialty contrac-tor agents (bidders) and organize an auction accordingto the requirements the seller submits. After severalrounds of conversation, the negotiation process willend with a deal reached between the seller and the bestbidder [12]. A third-party negotiation protocol is wellsuited for trading tasks where the number of specialtycontractors is too large to employ practically a simplebilateral negotiation protocol.

Common auctions protocols include Dutch, English,Vickrey, and First-Price-Sealed-Bid auctions [26,29].Although it is possible that, under some conditions, allthe four auction protocols are revenue equivalent for aprivate-value object, only English and Vickrey auctionsrely on agents playing their dominant strategies. AnEnglish auction is more robust with respect to chang-ing circumstances than a Vickrey auction. An Englishauction is still revenue superior in a very large class ofsituations, because the information revealed during theauction ensures that bidders will move closer to theirreservation prices [16].

6.3. Double bid based protocol

Normal auction protocols may have a problem whenthere are multiple calls for bids/tasks which use thesame resources belonging to a single bidder during thesame period of time. The most undesirable scenariomay be that a bidder fails to get any task assignmentalthough it has the sufficient capacity ready. A possiblesolution to this kind of problem is a double bid [23].

A double bid consists of two different bids, that is,one is a real bid computed according to the temporalresource schedule updated by the previous bid, and theother is a virtual bid which is better than the former andis generated based on the same resource schedule. Bothbids are evaluated by the auctioneer along with the bidsfrom other bidders. Suppose there are two auctioneersC1 and C2, and bidders S1 and S2, respectively. If thevirtual bid is determined as the winner, there are two

options for C2 to choose. One way is conservative, i.e.,to accept the real bid which is only inferior to the virtualbid; the other is more profitable but also more risky,i.e., to accept the best virtual bid and at the same timepreserve the second best real bid as a spare candidate.If C2 selects the latter, it must postpone the final winnerdetermination for a period of time to wait for the defini-tive real bid from S1 replacing the previous virtual oneand simultaneously send a ‘request for waiting’ mes-sage to the second best bidder. Once the update-bidof S1 arrives at C2 which offers no less competitiveconditions than what the earlier virtual bid does withinthe time limitation, auctioneer C2 will inform S1 ofan accept-bid. Otherwise, if the expected update-biddoesn’t return in time, the second best auctioneer willbe sent the final accept-bid message.

6.4. Simulated trading algorithm

To handle the enormous complexity of optimizinginterdependent plans in a supply chain, an optimizationprocedure is needed to improve the result after a firstvalid solution is found. Simulated trading is such acoordination mechanism. Trading is done in severalrounds, each of which consists of a number of decisioncycles. In each cycle, participants submit offers to sellor buy a task. At the end of each round a central man-ager agent tries to match the sell and buy offers of thecontractors such that the cost of the whole solution de-creases. Simulated trading procedure assumes a stablecooperative environment. An initial solution has to befound by using a contract net protocol before optimiza-tion algorithm can be used. Simulated trading is alsoan anytime algorithm, which is based on two nestedloops, i.e., trading rounds nested by decision cycles [6].

Each trading round represents a complete exchangeof tasks. After each trading round the agents’ plans arewell defined and valid again. The number of tradingrounds may be pre-specified before the optimizationstarts, if time is limited and an optimized solution isneeded at a certain point of time. Otherwise, the num-ber of trading rounds is unbounded, which means theoptimization never stops until someone stops it manu-ally.

Each trading round is then sub-divided into severaldecision cycles (which is usually � 10) and for eachone the agents generate a bid to sell or buy a task.When the current plan at the end of a trading round isbetter than the saved plan, the saved plan is replacedby the current plan. When the algorithm is terminatedor interrupted, it returns the saved plan which is the


best plan ever considered. Hence, the anytime propertyof monotonically growing quality of the solution isguaranteed.

6.5. Multi-attribute negotiation

A number of factors, such as cost, time, quality,safety, and environment, must be considered in thedecision-making process of SCM. The multi-attributenegotiation technique is developed based on the multi-attribute utility theory, which is an analytical tool formaking decisions involving multiple interdependentobjectives based on uncertainty and utility analysis, andan evaluation scheme for estimating various productsand performance [9–11,32].

A multi-attribute negotiation process contains fivesteps [8], i.e., (1) evaluation of the attributes of the ini-tial solutions made by the participants; (2) these evalua-tions are aggregated into overall utilities of these initialsolutions; (3) provision of the target utility; (4) basedon the target utility and the distribution of attributes,the values of the target attributes are determined, whichlead to a new round of decision making; and (5) foreach of the target attributes, an attribute value is chosenthat has an evaluation value as close as possible to thetarget evaluation value for the attribute.

7. Evaluations

In terms of the average procurement cost in apetroleum supply chain, the simulation results are notglobally optimal. The reason is that individual agentsonly consider their decision to select the supplier thatmakes the lowest procurement price or the nearest pro-curement date, and do not regard the effects for thecommon benefits of the society.

However, for CP-SCM, conventional negotiationprotocols are incapable due to the domain specific re-quirements of CP-SCM.

Auction protocols [26,29] are mainly suitable for thenegotiation for a relatively stable product or service,or their costs are easy to calculate. Because of thehuge demands worldwide for petroleum and high risksof petroleum exploration and exploitation, as well asmany unexpected factors, it is difficult for a bidderto submit an appropriate bid for auctions. Moreover,a contract net protocol directly takes place betweensupplier and consumer of a petroleum supply chain,thus the negotiation should be prompt and effectivesince there is no third-partner in the negotiation.

Although multi-attribute negotiation technology [30]considers the multiple factors which may influence thedecisions in a supply chain, it mainly depends upon theutilities of attributes. Because utilities often deviatefrom the values or costs of products or services, eventhe same products for the same consumers can have dif-ferent utilities after consuming. Petroleum consumersare very different from one another and thus the utili-ties of petroleum are diversified. Thus in practice anappropriate utility for petroleum during a certain periodis difficult to determine.

A simulated trading protocol [6] adopts a contractnet protocol and provides optimization mechanism byseveral rounds consisting of a number of decision cy-cles. However, it assumes that there is a stable coop-erative environment in a supply chain. If the demandsfor petroleum get alleviated this may be an ideal nego-tiation mechanism for petroleum SCM.

A double bid based protocol [23] considers that abidder can simultaneously respond to multiple calls forbids in auctions by providing a real bid and a virtualbid. However, there may be no essential differencebetween the real and the virtual bids, because they arecalculated based on the same resources and schedules.Very likely the real and the virtual bids will result inthe same petroleum supply chain formation.

Compared to conventional negotiation protocols, theproposed negotiation mechanism has the following ad-vantages.

First, the domain specific requirements of CP-SCMare taken into account. Conventional negotiation pro-tocols usually assume that there is a pure market, andthe bidder and the auctioneer are in an open market.

Second, the negotiation is carried out by many re-peated rounds, and any round will produce an ideal ne-gotiation result, i.e., to make the lowest procurementprice or nearest procurement date for consumers. Con-ventional negotiation protocols are usually completedby one round. For example, all auction protocols are tochoose one bidder based on their bids.

Third, multiple factors are considered, including notonly the quantitative factors such as price, date, andquantity, etc., but also the strategic ones such as col-laborative partnership between consumers and suppli-ers, and the semi-monopolized market nature. Conven-tional negotiation protocols usually only consider partof the factors and most of the conventional protocolsconsider price as an important factor.

Fourth, the main difference of the proposed negoti-ation mechanism from conventional negotiation proto-cols is that the final agreement on the bids is decidedby both consumers and suppliers, and demands driveand control the negotiation rounds.


8. Conclusions

This paper has demonstrated that MAS is an effec-tive methodology for SCM problems, called MA-SCM.The main benefit of MA-SCM is dynamic supply chainformation via multi-agent negotiation. While conven-tional negotiation protocols are based on the assump-tion of a pure market, this paper has proposed a newnegotiation mechanism to solve dynamic supply chainformation in a semi-monopolized market. The futureworks of this research can be undertaken in the follow-ing directions.

(1) To quantitatively evaluate the proposed negoti-ation mechanism, particularly the effectivenessof the negotiation, the adaptation to emergencyevents, and the supply chain optimization;

(2) To expand the multi-agent simulation scheme toother semi-monopolized markets, and to com-pare the simulation results with real-world sup-ply chain data; and,

(3) To develop a commercialized agent softwareplatform for dynamic supply chain formationsuitable for the petroleum industry.

Acknowledgements

Professor Xin Yao would like to acknowledge thesupports from Chinese Academy of Sciences (GrantNo. 2F03B01) and from the National 973 Program ofChina (Grant No. 2004CB318103).

References

[1] H.J. Ahn, H. Lee and S.J. Park, A flexible agent system forchange adaptation in supply chains, Expert Systems with Ap-plications 25 (2003), 603–618.

[2] A.M. Brandenburger and B.J. Nalebuff, Coopetition, NewYork, 1996.

[3] P. Buxmann and W. Konig, Inter-organizational Cooperationwith SAP Systems: Perspectives on Logistics and Service Man-agement, Springer-Verlag Berlin and Heidelberg, June 2000.

[4] Y. Chen, Y. Peng, Y. Labrou, S. Cost, B. Chu, J. Yao, R.Sun and B. Willhelm, A negotiation-based multi-agent systemfor supply chain management, Working Notes of Workshopon Agent Based Decision-Support for Managing the Internet-Enabled Supply-Chain, at the 3rd International Conference onAutonomous Agents (Agents’99), Seattle, Washington, 1–5May 1999, 15–20.

[5] J.M. Eilenberg, Dispute Resolution in Construction Manage-ment, University of New South Wales Press, Sydney, Austrilia,2003.

[6] A. Gerber, C. Russ and M. Klusch, Supply web coordinationby an agent-based trading network with integrated logisticsservices, Electronic Commerce Research and Applications 2(2003), 133–146.

[7] J. Himoff, Multi-agent technology: a revolution in supplychain planning and execution, Supply Chain Practice 6 (2004),50–59.

[8] C.M. Jonker and J. Treur, An agent architecture for multi-attribute negotiation, Proceedings of the 17th InternationalJoint Conference on Artificial Intelligence (IJCAI’01), Wash-ington, USA, 2–10 August 2001, 1195–1201.

[9] R. Keeny and H. Raiffa, Decisions with Multiple Objectives:Preferences and Value Tradeoffs, John Wiley Sons, 1976.

[10] M. Kelly and M. Thorne, C. An approach to multi-attributeutility analysis under parametric uncertainty, Annals of Nu-clear Energy 28 (2001), 875–893.

[11] J.K. Kim and S.H. Choi, A utility range-based interactivegroup support system for multi-attribute decision-making,Computer and Operations Research 28 (2001), 485–503.

[12] K. Kim, B.C. Jr. Paulson and C.J. Jr. Petrie, Agent-basedelectronic markets for project supply chain coordination, Pro-ceedings of Workshop on Knowledge-Based Electronic Mar-kets, at the 17th National Conference on Artificial Intelligence(AAAI’00), Texas USA, 30 July – 3 August 2000, 37–41.

[13] D. Kjenstad, Coordinated supply chain scheduling, PhD The-sis and NTNU Report 1998–1924, Department of Productionand Quality Engineering, Norwegian University of Scienceand Technology (NTNU), Trondheim, November 1998.

[14] K.J. Lee, Y.S. Chang and J.K. Lee, Time-bound negotiationframework for electronic commerce agents, Decision Support-ing Systems 28 (2000), 319–331.

[15] P.B. Luh, M. Ni and H. Chen, A price-based multi-agent ap-proach for supply chain coordination, Proceedings of First In-ternational Symposium on Multi-Agents and Mobile Agentsin Virtual Organisations and E-Commerce (MAMA’00), Wol-long, Australia, 11–13 December 2000.

[16] J. Moore, Implementation, contracts, and renegotiation in en-vironments with complete information, Advances in EconomicTheory I (1992), 182–282.

[17] F. Pena-Mora and C.Y. Wang, Computer-supported collabora-tive negotiation methodology, Journal of Computing in CivilEngineering 12 (1998), 64–81.

[18] D. Pruitt, Negotiation Behavior, Academic Press, London,1981.

[19] J. Rosenschein and G. Zlotkin, Rules of Encounter, The MITPress, Cambridge, USA, 1994.

[20] N.M. Sadeh, D.W. Hildum, D. Kjenstad and A. Tseng, MAS-COT: An agent based architecture for dynamic supply chaincreation and coordination in the internet economy, Journal ofProduction Planning and Control 12 (2001), 211–223.

[21] J. Sauer and H.-J. Appelrath, Scheduling the supply chain byteams of agents, Proceedings of the 36th Hawaii InternationalConference on System Sciences (HICSS’03), 6–9 January2003.

[22] R.G. Smith, The contract net protocol: high-level commu-nication and control in the distributed problem solver, IEEETransactions on Computers C-29 (1980), 1104–1113.

[23] L. Ta, Y. Chai and Y. Liu, Multi-agent-based architecture andmechanism for coordination and execution in agile supplychain operational management, Proceedings of the 2005 IEEEInternational Conference on e-Technology, e-Commerce ande-Service (EEE’05), Hong Kong, China, 29 March – 1 April2005.


[24] M. Ulieru and M. Cobzaru, Building holonic supply chainmanagement systems: an e-logistics application for the tele-phone manufacturing industry, IEEE Transactions on Indus-trial Informatics 1 (2005), 18–30.

[25] G. Vierke, TELETRUCK – A holonic multi-agent systemfor telematics, PhD Thesis, Universitat des Saarlandes,Saarbrucken, 2000.

[26] N. Vulkan and N.R. Jennings, Efficient mechanisms forthe supply of services in multi-agent environments, FirstInternational Conference on Information and ComputationEconomies (ICE’98), Charleston, SC, USA, 25–28 October1998.

[27] W.E. Walsh and M.P. Wellman, Modeling supply chain for-mation in multiagent systems, Proceedings of Workshop onAgent Mediated Electronic Commerce, at the 16th Interna-tional Joint Conference on Artificial Intelligence (IJCAI’99),Stockholm, Sweden, 31 July – 6 August 1999, 94–101.

[28] W.E. Walsh, M.P. Wellman and F. Ygge, Combinatorial auc-tions for supply chain formation, Proceedings of the Sec-ond ACM Conference on Electronic Commerce, Minneapolis,Minnesota, USA, 17–20 October 2000, 260–269.

[29] E. Wolfstetter, Auctions: an Introduction, Economic Surveys10 (1996), 367–420.

[30] X. Xue, X. Li, Q. Shen and Y. Wang, An agent-based frame-work for supply chain coordination in construction, Automa-tion in Construction 14 (2005), 413–430.

[31] Z. Yang, D. Zhang and J. Xu, The simulation of service supplychain formation based on mobile agent’s searching, Proceed-ings of IEEE International Conference on E-Commerce Tech-nology for Dynamic E-Business (CEC-East 2004), Beijing,China, 13–15 September 2004, 175–178.

[32] A.D. Youngblood and T.R. Collins, Addressing balancedscorecard trade-off issues between performance metrics usingmulti-attribute utility theory, Engineering Management Jour-nal 15 (2003), 11–17.

[33] Y. Yuan, T.P. Liang and J.J. Zhang, Using agent technologyto support supply chain management: potentials and chal-lenges, First Annual Symposium on Supply Chain Manage-ment, Toronto, Canada, 17 November, 2003.

[34] http://more.btexact.com/projects/agents/zeus/.[35] http://www.cnooc.com.cn/.[36] http://www.cnpc.com.cn/.[37] http://www.sinopec.com/jndex.jsp.

Authors’ Bios

Jiang Tian was a PhD research student in Glas-gow Caledonian University from September 2002 toDecember 2005. He obtained his Higher Diploma inPetroleum Engineering from Chongqing Higher Col-lege of Petroleum Technology in 1989, Bachelor De-gree in Industrial Economy from Southwest Univer-sity of Finance and Economics in 1995, and Bachelor(Honors) Degree in Accounting from Southwest Uni-versity in 1999, and Postgraduate Diploma in Technol-ogy Economy and Management from Southeast Uni-versity in 2002. Prior to his PhD research, he hadworked twelve years in Chinese petroleum industry.His research areas include Supply Chain Management,

Multi-Agent Systems, Virtual Organizations, and E-Business. So far he has published about ten researchpapers in journals and conference proceedings.

Richard Foley obtained his BSc (Honors) from Uni-versity of Strathclyde, Glasgow in 1980, and PhD fromGlasgow Caledonian University in 1994. Since 1984he has been a lecturer at Glasgow Caledonian Univer-sity. His original PhD study was in the use of Non-Monotonic logic in the provision of Knowledge BaseSystems. However, since then his interests have been inthe application of AI systems to different areas of Soft-ware Engineering, such as Software Quality Planningand supporting Virtual Software Corporations.

Xin Yao obtained his BSc from the University ofScience and Technology of China (USTC) in Hefei in1982, MSc from the North China Institute of Com-puting Technology in Beijing in 1985, and PhD fromUSTC in Hefei in 1990. Xin Yao was an associatelecturer and lecturer between 1985 and 1990 at USTCwhile working on his PhD. His PhD work on simulatedannealing and evolutionary algorithms was awarded thePresident’s Award for Outstanding Thesis by the Chi-nese Academy of Sciences. He took up a postdoc-toral fellowship in the Computer Sciences Laboratory,headed by Professor Richard Brent, at the AustralianNational University (ANU) in Canberra in 1990, andcontinued his work on simulated annealing and evolu-tionary algorithms. He joined the Knowledge-BasedSystems Group, led by Dr Ron Sharpe, at CSIRO Di-vision of Building, Construction and Engineering inMelbourne in 1991, working primarily on an industrialproject on automatic inspection of sewage pipes. Hereturned to Canberra in 1992 to take up a lectureshipin the School of Computer Science, University Col-lege, the University of New South Wales (UNSW), theAustralian Defence Force Academy (ADFA), where hewas later promoted to a senior lecturer and associateprofessor. Attracted by the English weather, he movedto the University of Birmingham, England, as a pro-fessor of computer science in 1999. Currently he isthe Director of CERCIA (the Centre of Excellence forResearch in Computational Intelligence and Applica-tions), a Distinguished Visiting Professor of the Uni-versity of Science and Technology of China in Hefei,and a visiting professor of three other universities. Heis an IEEE Fellow. He has more than 200 publicationsand won the 2001 IEEE Donald G. Fink Prize PaperAward for his work on evolutionary artificial neuralnetworks. In his spare time, he does the voluntary workas the editor-in-chief of IEEE Transactions on Evolu-tionary Computation, an associate editor or editorial


board member of several other journals, and the edi-tor of the World Scientific book series on “Advancesin Natural Computation”. He has given more than 30invited keynote and plenary speeches at conferencesand workshops world-wide. His research has been sup-ported by many funding bodies (more than 4M poundsin the last four years). His major research interests in-clude evolutionary artificial neural networks, automaticmodularization of machine learning systems, evolu-

tionary optimization, constraint handling techniques,computational time complexity of evolutionary algo-rithms, co-evolution, iterated prisoner’s dilemma, datamining, and real-world applications.

Huaglory Tianfield

See Volume 1 Number 2.

Date post:	20-Dec-2021
Category:	Documents
Upload:	others
View:	0 times
Download:	0 times

An extended contract net mechanism for dynamic supply ...

Documents