Date post: | 02-May-2018 |
Category: |
Documents |
Upload: | trannguyet |
View: | 216 times |
Download: | 3 times |
Best Practices in the Use of Micro Simulation Models
Prepared for:
American Association of State Highway and Transportation Officials (AASHTO)
Standing Committee on Planning
Prepared by:
Hayssam Sbayti and David Roden
3101 Wilson Boulevard, Arlington, VA 22201
March 2010
The information contained in this report was prepared as part of NCHRP Project 8-36, Task 90, National Cooperative Highway Research Program, Transportation Research Board.
SPECIAL NOTE: This report IS NOT an official publication of the National Cooperative Highway Research
Program, Transportation Research Board, National Research Council, or The National Academies.
Best Practices in the Use of Micro Simulation Models
ii
Acknowledgments
This study was conducted for the American Association of State Highway and Transportation Officials (AASHTO), with funding provided through the National Cooperative Highway Research Program (NCHRP) Project 8-36. The NCHRP is supported by annual voluntary contributions from the state Departments of Transportation. Project 8-36 is intended to fund quick response studies on behalf of the AASHTO Standing Committee on Planning. The report was prepared by Hayssam Sbayti and David Roden from AECOM. The work was guided by a task group co-chaired by Timothy A. Henkel and Jonette R. Kreideweis and that included Charles E. Howard, Susan A. Gorski, Jim Henricksen, Susan P. Mortel, Ram M. Pendyala, and Brian Gardner. Disclaimer
The opinions and conclusions expressed or implied are those of the research agency that performed the research and are not necessarily those of the Transportation Research Board or its sponsoring agencies. This report has not been reviewed or accepted by the Transportation Research Board Executive Committee or the Governing Board of the National Research Council.
NOTE: The Transportation Research Board of the National Academies, the National Research Council,
the Federal Highway Administration, the American Association of State Highway and Transportation
Officials, and the individual states participating in the National Cooperative Highway Research Program
do not endorse products or manufacturers. Trade or manufacturers’ names appear herein solely
because they are considered essential to the object of this report
Best Practices in the Use of Micro Simulation Models
iii
Table of Contents 1 Executive Summary .................................................................................................................................. 1
1.1 Under what circumstances have micro simulation modeling efforts been found to be warranted
and cost-effective?........................................................................................................................... 2
1.2 What are the benefits of microscopic simulation models compared to the currently available
macroscopic models? ...................................................................................................................... 3
1.3 In what ways have investing and conducting micro simulation led to better or more effective
designs and investment decisions? ................................................................................................. 3
1.4 How have the data needed for micro simulation and time-dependent models been developed? 4
1.5 What steps are being taken to attract and retain staff with the necessary expertise and
knowledge to perform and oversee micro simulation and dynamic traffic assignment modeling
efforts? ............................................................................................................................................. 5
1.6 What organizational or institutional arrangements have been useful in addressing any or all of
the issues identified above? ............................................................................................................ 6
2 Introduction ............................................................................................................................................. 8
2.1 Study Context ................................................................................................................................... 8
2.2 Research Approach ........................................................................................................................ 10
2.2.1 Literature Review .................................................................................................................... 10
2.2.2 Web-Based Survey................................................................................................................... 11
2.2.3 Peer Exchange Meeting ........................................................................................................... 11
2.2.4 Case Study Interviews ............................................................................................................. 11
3 Literature Review ................................................................................................................................... 12
4 Web-Based Survey ................................................................................................................................. 16
4.1 Use of Simulation Tools in the Industry ......................................................................................... 16
4.2 Linking Simulation with Regional Model ....................................................................................... 19
4.3 Application Experience .................................................................................................................. 21
4.3.1 Project Scope and Objectives .................................................................................................. 22
4.3.2 Data Preparation and Conversion ........................................................................................... 24
4.3.3 Model Implementation and Calibration .................................................................................. 25
4.3.4 Model Application ................................................................................................................... 28
4.3.5 Model Visualization and Presentation .................................................................................... 29
4.3.6 Data Management and Maintenance ..................................................................................... 29
4.4 Clients vs. Contractors ................................................................................................................... 29
Best Practices in the Use of Micro Simulation Models
iv
4.5 Lessons Learned ............................................................................................................................. 30
5 Peer Exchange Meeting ......................................................................................................................... 32
5.1 Words of Advice ............................................................................................................................. 33
5.2 Fundamental Considerations ......................................................................................................... 36
5.3 Recommendations ......................................................................................................................... 37
6 Case Studies ........................................................................................................................................... 38
7 Synthesis of Findings and Lessons Learned ........................................................................................... 45
7.1 Under what circumstances have micro simulation modeling efforts been found to be warranted
and cost-effective?......................................................................................................................... 45
7.2 What are the benefits of microscopic simulation models compared to the currently available
macroscopic models? .................................................................................................................... 47
7.3 In what ways have investing and conducting micro simulation led to better or more effective
designs and investment decisions? ............................................................................................... 48
7.4 How have the data needed for micro simulation and time-dependent models been developed?
49
7.5 What steps are being taken to attract and retain staff with the necessary expertise and
knowledge to perform and oversee micro simulation and dynamic traffic assignment modeling
efforts? ........................................................................................................................................... 50
7.6 What organizational or institutional arrangements have been useful in addressing any or all of
the issues identified above? .......................................................................................................... 51
8 Best Practices in Micro Simulation ......................................................................................................... 53
8.1 Data Quality and Consistency ........................................................................................................ 53
8.2 Network Coding and Error Checking .............................................................................................. 54
8.3 Origin-Destination Demand Preparation ....................................................................................... 55
8.4 Calibration and Validation ............................................................................................................. 56
8.5 Modeling Difficulties ...................................................................................................................... 57
9 Model Selection Guidelines ................................................................................................................... 58
9.1 Selecting a Modeling Approach ..................................................................................................... 58
9.2 Selecting Microscopic Simulation Software ................................................................................... 61
Appendix A: Case Study Summaries ........................................................................................................... 62
Caltrans Interstate 5 Traffic Study (I-5/91 to I-5/405) ........................................................................... 62
City of Moreno Valley, CA, TRANSIMS Implementation ........................................................................ 63
New York City Department of Transportation (NYCDOT) Broadway Boulevard Project in Midtown
Manhattan ..................................................................................................................................... 64
Caltrans I-5 North Coast Traffic Study .................................................................................................... 65
Best Practices in the Use of Micro Simulation Models
v
Sacramento Area Council of Governments (SACOG) Integration of an Activity-Based Model with
TRANSIMS ..................................................................................................................................... 66
Caltrans I-80 Integrated Corridor Management (ICM) project .............................................................. 67
Caltrans I-580 Corridor System Management Plan................................................................................ 68
MnDOT I-94 Managed Lane Operations Study ...................................................................................... 69
Maricopa Association of Governments (MAG) I-10 Integrated Corridor Management Analysis .......... 70
Colorado Department of Transportation (CDOT) Downtown Denver Multimodal Access Plan (DMAP)
....................................................................................................................................................... 71
Utah Department of Transportation (UDOT) NEPA Analysis ................................................................. 72
Arkansas State Highway and Transportation Department (AHTD) Hwy 112/Garland Ave ................... 73
Kansas Department of Transportation (KDOT) Johnson County Gateway I-435/I-35/K-10 .................. 74
San Diego Association of Governments (SANDAG) I-15 ICM Project ..................................................... 75
San Francisco County Transportation Authority (SFCTA) Doyle Drive Study ......................................... 76
List of Figures Figure 1: Types of Studies Where Micro Simulation Has Been Helpful ...................................................... 18
Figure 2: Reasons for Using Micro Simulation Models ............................................................................... 23
Figure 3: Temporal Dimensions of Micro Simulation Models .................................................................... 25
Figure 4: Level of Satisfaction with Calibration and Validation .................................................................. 26
Figure 5: Major Difficulties Implementing Micro Simulation Models ........................................................ 27
Figure 6: Benefits of Using Micro Simulation Models................................................................................. 28
Figure 7: Preparing Time-Dependent Demand ........................................................................................... 55
List of Tables Table 1: Project Summary—All Respondents ............................................................................................. 30
Table 2: Project Summary—Transportation Agency Respondents............................................................. 30
Table 3: Summary of 15 Case Study Projects .............................................................................................. 44
Table 4: General Applicability of Various Modeling Approaches ............................................................... 59
Best Practices in the Use of Micro Simulation Models
1
1 EXECUTIVE SUMMARY
Traffic congestion is a growing problem that is increasingly difficult for states and regional
transportation authorities to address. Congestion increases travel times, traveler stress, and accident
rates; reduces mobility, accessibility, and system reliability; and results in loss of productivity and
environmental degradation.
Transportation agencies are pursuing more efficient utilization of the existing transportation system as
one of the few available options to improve mobility. This includes strategies to refine facility operations
to improve performance and throughput, and strategies to manage or influence demand. Successful
management techniques can increase throughput and spread demand to other times of day, other
modes of transportation, or other routes.
The analytical tools required to support these efforts need to provide a detailed assessment of how
traffic operates and travelers respond to system impacts. Unfortunately, traditional macroscopic models
are generally ineffective in evaluating strategies designed to influence travel choices and optimize
system performance. In particular, traditional four-step models cannot capture traffic dynamics, queue
spill-backs, intersection control delays, vehicle–pedestrian interactions, geometric design impacts, or
traveler-specific response to congestion pricing and route-guidance programs.
As a result, simulation-based models are being recommended to aid transportation planners, designers,
and policy-makers in assessing future needs and mobility options. Microscopic and mesoscopic
simulation models overcome the inherent limitations of traditional four-step models through their
ability to model detailed system operations and management strategies.
While micro simulation models are definitely desirable, they require details about transportation
facilities and observed travel behavior at a granularity that is not typically available to Metropolitan
Planning Organizations (MPOs) and state agencies. This is further complicated by the fact that few
agencies have staff with sufficient expertise and knowledge to oversee micro simulation modeling
efforts.
This research focuses on an objective assessment of the state of the practice in the use of traffic micro
simulation tools by states and metropolitan areas for transportation planning studies. A major issue
faced by state agencies and MPOs is whether the extra overhead associated with developing, applying,
and maintaining micro simulation traffic models is warranted and justified. This study used a literature
review, web-based survey, peer exchange meeting, and case study interviews to determine if there is
consensus among transportation professionals with traffic simulation experience as to where, when,
and how micro simulation modeling can be best supported, justified, and cost-effective. The overall
objective of the research is to provide guidance to officials and policy-makers who are considering using
micro simulation models for future studies.
The following research questions were posed to establish a framework for identifying the pertinent
information that would best determine if a consensus exists regarding the appropriate use of micro
simulation modeling.
Best Practices in the Use of Micro Simulation Models
2
1.1 Under what circumstances have micro simulation modeling efforts been found to be warranted and cost-effective?
The level of detail and accuracy required to answer stakeholder questions is increasingly beyond the
capabilities of traditional models. As a general rule, microscopic simulation models have been shown to
be warranted in applications that are likely to induce a temporal shift of traffic among different roadway
facilities or experience capacity constraints that cause queues to influence the performance of other
facilities at a corridor or network-wide level.
Microscopic simulation models are warranted where the detailed interactions of vehicle movements are
the primary focus of the study. This includes the interactions of vehicles with pedestrians and bicyclists;
priority vehicle treatments such as high-occupancy vehicle (HOV) or bus lanes; and congested traffic
conditions, particularly if there are closely spaced intersections. Design considerations related to lane
changing, such as complex weaving sections and the length of pocket lanes, are also prime candidates
for micro simulation.
The research has also shown that micro simulation can be warranted for its visualization capabilities.
Visual animation of the traffic conditions can help to develop credibility for the modeling process and
demonstrate the benefits of the proposed solution in ways simple numbers cannot. This can be
important in communicating the results of the analysis to decision-
makers and the general public.
Despite all of these advantages, very few of the study participants
believed their micro simulation studies were cost-effective. They
almost always agreed that the tool answered the study questions
and influenced the design decisions, and they plan to do similar
studies in the future because they see no viable analysis
alternative. But from the perspective of cost-effectiveness, they
found the time and cost involved in developing and applying
microscopic simulation models to be excessive.
In most cases, the project schedule, budget, and analysis
expectations are set based on years of planning experience using
traditional models. This becomes the de facto benchmark by which cost-effectiveness is defined for
microscopic simulation projects. Modelers, managers, and decision-makers have not adjusted their
expectations to account for the cost and complexity of advanced modeling techniques.
Microscopic simulation models require significantly more data and analysis resources than macroscopic
models, but benefit planning by providing more accuracy and fidelity to actual traffic conditions.
Therefore, selecting a model type will always be a trade-off between the desired level of granularity and
confidence in the recommendations and the available resources to produce these more costly products.
From the perspective of
cost-effectiveness, most
study participants found
the time and cost involved
in developing and
applying microscopic
simulation models to be
excessive.
Best Practices in the Use of Micro Simulation Models
3
1.2 What are the benefits of microscopic simulation models compared to the currently available macroscopic models?
Macroscopic models were originally designed to estimate demand for major highway improvements
included in regional long-range plans. Travel times in these models are based on volume–capacity
relationships used in the assignment process to distribute the travel demand to a logical set of paths. In
this process, volumes can easily exceed capacities and loaded speeds have little relationship to reality.
The purpose of these macroscopic models is not so much to predict how a facility will actually operate,
but to estimate the demand for a given facility so that the design might accommodate the demand.
By contrast, microscopic simulation models are focused on the actual operations of the facilities and the
impacts of capacity constraints on upstream and downstream system performance by time of day. If the
purpose of the study is to address congestion problems with operational and management strategies,
microscopic models are much better positioned than macroscopic models to evaluate the effectiveness
of the alternatives.
Static assignments permit volume-to-capacity (v/c) ratios that are greater than one whereas microscopic
simulation models constrain the flow rate at bottleneck locations. The microscopic model will show
more congestion upstream of the bottleneck and less congestion downstream of the bottleneck than a
traditional model.
Microscopic simulation can also help to identify problems during moderate traffic conditions. In these
situations the v/c ratios are not high enough for a macroscopic model to predict any noticeable delay,
while the microscopic simulation can properly identify delays at off-ramps and propagating queues from
upstream intersections that warrant attention.
In general, the key advantages of micro simulation tools can be traced to their ability to:
Model different vehicle types and intersection controls
Consider temporal and spatial interactions
Consider traffic dynamics, queuing phenomena, and spill-backs
Model different behavioral assumptions and user classes
Visualize how proposed alternatives will operate
Despite these advantages and capabilities, there is still significant debate among transportation
professionals and decision-makers about the benefits of micro simulation modeling. Some decision-
makers are not convinced that micro simulation results are reliable or comprehensive enough for major
capital investment decisions. Others have yet to see effective returns from the investment in micro
simulation due to the amount of effort required to code, calibrate, and apply the model.
1.3 In what ways have investing and conducting micro simulation led to better or more effective designs and investment decisions?
The survey respondents reported numerous situations where micro simulation led to better investment
decisions and more effective designs. In general, a properly calibrated and validated microscopic
simulation model will more often than not lead to more effective designs and investment decisions
because it can more closely replicate what would likely occur in the real world.
Best Practices in the Use of Micro Simulation Models
4
Suggesting, however, that a design is better or more effective is either a subjective opinion or it requires
some basis of comparison. In most of the cases reported in the
web survey, the assessment was based on a comparison to prior
studies using macroscopic models or Highway Capacity Manual
(HCM) calculations. Micro simulation modeling proved particularly
useful in situations that are not well handled by macroscopic
modeling techniques such as complex intersection configurations
or heavily congested arterials.
Respondents to the web survey reported benefits of micro
simulation as improved decision-making that led to changing or
reversing an existing policy or design, less expensive capital
investments, and more efficient solutions.
1.4 How have the data needed for micro simulation and time-dependent models been developed?
Microscopic simulation models are significantly more complex and far more sensitive to minute details
about the transportation system and travel demand than traditional models. This is precisely why they
are better able to evaluate the subtle differences that result from demand management and operational
improvement strategies. This also means they are far more susceptible to small errors or inaccuracies in
the input data or model parameters that are multiplied by travel demand to create large errors in the
model results. To avoid or minimize these errors, considerable care should be taken in preparing the
network, travel demand, and model parameters.
Data quality and consistency are much more important for microscopic simulation models than they are
for macroscopic models. The modeler needs to know the pedigree of the data that are being used and
should never assume that the data are error-free or correct simply because they are provided by a
credible source. Data conversion tools can significantly reduce network coding times and sophisticated
editing tools can automatically add or synthesize many of the details. It is also beneficial to have staff
with a strong programming background create and automate some data entry and error-checking
processes.
The most common practice for preparing time-dependent origin–destination (O-D) data is to apply
diurnal distributions to trip tables extracted from regional travel demand models. It is preferable to
apply separate diurnal distributions by trip purpose and vehicle type. If possible, the distributions should
differ by general O-D pair. The distribution of demand by facility type should also be checked against
traffic counts to ensure that the overall shape of the curve is reasonable.
Time-dependent demand derived from regional planning models often includes travel patterns and
temporal distributions that differ from observed traffic counts. Micro simulation studies will generally
adjust the demand data using O-D estimation techniques to match time-of-day traffic counts on specific
facilities in the study area. These procedures are most effective if applied with 15-minute traffic counts
and where at least 60 percent of the trips travel through links with traffic counts.
Study participants
reported numerous cases
where micro simulation
led to changes in existing
policies or designs, less
expensive capital
investments, and more
efficient solutions.
Best Practices in the Use of Micro Simulation Models
5
Calibrating the simulation model parameters based on local conditions and traveler behavior is a key
component of the model development process. Visualization tools are helpful in comparing the results
to observed behavior and in identifying the source of simulation problems. The model should also be
validated against traffic counts, speeds, queue lengths, and other performance measures at a 15-minute
level. In addition to validating a specific point in time, it is also good practice to validate the model at a
different point in time when conditions are different. This is typically called “back-casting,” and it helps
build credibility for the model’s forecasts.
Gaining access to automated data collection systems is an important way to reduce costs and improve
model performance. The key is to ensure that the data collection and aggregation process provides
information that is useful to model development and operational planning. If at all possible,
arrangements should be established with the data collecting agency to summarize traffic volumes and
speeds simultaneously for relatively short time periods (e.g., 15 minutes). Quality control and data
cleaning procedures should also be included to minimize distortions in the results.
1.5 What steps are being taken to attract and retain staff with the necessary expertise and knowledge to perform and oversee micro simulation and dynamic traffic assignment modeling efforts?
As microscopic simulation models increasingly bridge the gap between planning and operations,
developing these models will require skills that are normally split between two distinct disciplines:
planners and traffic engineers. Simulation modelers with skill in both planning and traffic operations are
highly desirable. The majority of survey participants from public
agencies believe that they can offer significantly better benefits
and reasonable salaries compared to the private sector, and as
such they believe they can hire new talent or hold onto their
talented staff.
Public agencies compete for talent by emphasizing the quality of
the work environment for prospective employees. At public
agencies, retention is steady and the average tenure is measured
in years, which is attractive to those for whom job security is
paramount. They also cite more time off and less overtime as
reasons they are generally attractive places to work for modelers
starting families who are willing to trade greater time for familial obligations against the higher
monetary compensation available in the private sector. The economic recession seems to have made
the benefits of public employment comparatively greater, and as such it is hard to find any open
engineering positions of significance at public agencies.
However, public agencies face several long-term structural challenges when it comes to hiring talented
staff. For example, there is a ceiling that limits how much the agency can offer in terms of salary and
benefits, which are generally non-negotiable. There is also the challenge of a lack of upward mobility
within public agencies. They are also at a disadvantage with the private sector for recruiting prospective
employees needing a quick hiring decision. Public agencies have the administrative overhead of
Most public agency
participants believed they
can hire and hold onto
talented staff by offering
significantly better
benefits and reasonable
salaries compared to the
private sector.
Best Practices in the Use of Micro Simulation Models
6
developing, approving, and posting job announcements and following set hiring procedures. Consulting
firms can often hire staff based on recommendations and referrals rather than a formal job posting.
1.6 What organizational or institutional arrangements have been useful in addressing any or all of the issues identified above?
The web survey, peer exchange, and follow-up interviews identified several organizational and
institutional arrangements that are useful in developing cost-effective microscopic simulation models.
Perhaps the most important concept is to have an appropriate public agency be the “home” for a
particular model. This helps to ensure continued investment in the model’s capabilities and advocacy
and support for its application. It also provides good version control and an error tracking system for
networks and model parameters. It also helps to have progressive leadership that supports the reuse of
micro simulation data and sets the stage for strategic modeling initiatives.
Easy access to software support staff and technical manuals is also important. These services could be
provided by software vendors, but it is often better to have experienced software users providing this
assistance. Modelers are likely to hit obstacles as they calibrate or implement the model and need a
resource to which they can turn for advice. If the agency does not have much experience with the
software, it may be desirable to include the software vendor as a paid advisor on the project. This often
results in the vendor enhancing or fixing software capabilities sooner than if they were not involved.
In addition, coordinating with other agencies that routinely conduct microscopic simulation studies
provides less experienced agencies with the information to identify likely obstacles and how to address
them in each phase of model development. This can be achieved through user group meetings or peer
exchange sessions for strategic planning or a specific project.
Coordination should not, however, be limited to simulation modelers. The technical and planning staffs
from various stakeholders should be kept apprised during the model development process. Their input
will be useful in addressing specific questions and identifying or critiquing proposed alternatives. Being
an experienced simulation modeler is not enough; understanding the theory behind traffic operations is
even more important. Providing training in traffic engineering principles and involving traffic engineers
in the modeling process are important ingredients for success.
The peer exchange panel encouraged states and MPOs to begin the process of migrating from one-off
project models to integrated multi-scale modeling capabilities within a region. From a technical or
theoretical perspective the multi-scale approach recognizes the fact that modeling small areas with
microscopic detail fails to consider the likely impact of performance changes in that small area on the
larger corridor or regional travel behavior. The multi-scale approach was viewed as a practical and
computationally efficient way to consider the full range of issues, traveler responses, and performance
measures that influence investment decisions.
Panel participants recognized that the technical challenges of developing an integrated multi-scale
modeling capability are significant, but they did not view them as insurmountable. Solutions to the
administrative and organizational issues were more difficult to identify. The fact that different
government agencies have primary responsibilities for different types of projects, and funding and staff
Best Practices in the Use of Micro Simulation Models
7
resources are spread thinly among a large number of departments, programs, and projects, makes the
prospect of a coordinated, consolidation initiative difficult to visualize, let alone implement.
In the meantime, most modeling decisions will be based on the
opportunities and constraints of a specific project. In this case the
focus is on schedule and budget limitations, project purpose and
scope, available data and tools, staff resources and expertise, and
the need for visual aids to communicate the results to decision-
makers and the general public. The decision may also be
constrained by regulatory requirements, political sensitivities, and
attributes of the study area that require special attention. In other
words, the decision is made based on immediate project needs
rather than long-range goals of the public agency.
Given this environment, the peer exchange panel recommended the follow actions:
Consult more experienced agencies before selecting a model type
Develop guidelines for selecting and operating model types
Nurture an understanding that modeling costs increase with the level of detail required
Consider the consequences of selecting the wrong model type
Collect data now for the eventual adoption of a micro simulation model in the future
Pilot test micro simulation models before committing to a major project
Phase in the adoption of micro simulation modeling
Study the software capabilities and algorithms in detail
The study found that most
modeling tool decisions
are based on the
opportunities and
constraints of a specific
project.
Best Practices in the Use of Micro Simulation Models
8
2 INTRODUCTION
2.1 Study Context
Traffic congestion is a growing problem that is increasingly difficult for states and regional
transportation authorities to address. Congestion increases travel times, traveler stress, and accident
rates; reduces mobility, accessibility, and system reliability; and results in loss of productivity and
environmental degradation.
Traditional strategies for improving network performance through adding lanes or new facilities are
becoming largely infeasible. The space required to build new infrastructure is not available or the
impacts on the social or natural environment required to make the space are too onerous to consider.
Even in situations where space is not the issue, the cost of constructing and maintaining new facilities is
increasingly prohibitive.
Transportation agencies are pursuing more efficient utilization of the existing transportation system as
one of the few available options to improve mobility. This includes strategies to refine facility operations
to improve performance and throughput, and strategies to manage or influence demand. Successful
management techniques can increase throughput and spread demand to other times of day, other
modes of transportation, or other routes.
The analytical tools required to support these efforts need to provide a detailed assessment of how
traffic operates and travelers respond to system impacts. Unfortunately, traditional macroscopic models
are generally ineffective in evaluating strategies designed to influence travel choices and optimize
system performance. The likely result of using these models for the critical questions of today is either
no or possibly incorrect sensitivities to change in inputs and policies. In particular:
Traditional models do not consider the impacts of nonrecurring congestion, system
reliability, vehicle performance characteristics, random or differential behavior,
household travel constraints, available information, or weather conditions on travel
choices or the response of travelers to conditions that change during the course of their
trip.
Traditional models do not consider the interactions of vehicles and pedestrians with
transportation facilities and operational controls at the level of spatial or temporal
detail needed to evaluate improvements to traffic operations, transit access, geometric
design, and demand management strategies on system performance by time of day.
The simplifying assumptions that make the traditional modeling paradigm
computationally efficient for major facility forecasting make them ineffective from a
practical and theoretical perspective for predicting the effect of transportation
management strategies or analyzing the capacity-related policies that cities and states
are considering today.
Because traditional four-step models cannot capture traffic dynamics, queue spill-backs, intersection
control delays, vehicle–pedestrian interactions, geometric design impacts, or traveler-specific response
Best Practices in the Use of Micro Simulation Models
9
to congestion pricing and route-guidance programs, advanced traffic analysis tools are needed. To this
end, simulation-based models are being recommended to aid transportation planners, designers, and
policy-makers in assessing future needs and mobility options. Simulation models overcome the inherent
limitations in traditional four-step models and are the tool most suited for operational and policy
planning.
Simulation refers to the movement of people and/or vehicles on the network and is the method of
choice used to estimate the state of the system (link travel times, turn penalties, link flows) when pure
mathematical and analytical models are inadequate to do so. Traffic simulation is a technique for
describing the behavior of vehicles in a network and is generally used in conjunction with assignment
models. Simulation has the merit of closely approximating the behavior of individual drivers while easily
incorporating all kinds of control measures. Simulation implies keeping track of time and the
interactions of vehicles over time, and as such provides traffic realism that cannot be captured using
analytical models.
Traffic simulation tools are generally classified based on three levels of resolution, namely, microscopic,
mesoscopic, and macroscopic. Microscopic simulation tools, such as TRANSIMS, MITSIM, AIMSUN,
Paramics, TransModeler, and VISSIM, account for the movements of individual travelers dynamically and
stochastically in the network on a second-by-second basis using cellular automata or car-following
models. They require detailed geometric, control, and demand data and a large number of calibrated
parameters to accurately model driver behavior in the network. At the other end of the spectrum
macroscopic simulation models, such as FREQ and TRANSYT-7F, model the movement of packets of
vehicles over a fixed period of time based on the hydrodynamic theory of traffic flow. These models
require much less processing power and only a handful of relatively simple parameters in order to work
reasonably well. A third approach that is gaining popularity is mesoscopic simulation tools, such as
DynaMIT, DynusT, VISTA, and DYNASMART, where a macroscopic traffic flow model is used but vehicles
are tracked individually in the network to maintain a higher level of detail.
While micro simulation–based traffic models are definitely desirable, they require details about
transportation facilities and observed travel behavior at a granularity that is not typically available to
MPOs and state agencies. A number of simulation applications have attempted to address this issue by
synthesizing the detailed data needed for micro simulation from existing transportation modeling
networks, geographic information system (GIS) databases, and/or route guidance system street
networks. The details required for future planning studies may be synthesized based on basic traffic
engineering principles and warrants. These methods can produce a network quickly, but still require the
analyst to edit and validate the model results based on real-world traffic data and operational details
that are difficult and costly to collect.
This is further complicated by the fact that few agencies have staff with sufficient expertise and
knowledge to oversee micro simulation modeling efforts. In addition, simulation models are often
difficult to use for the types of analytical approaches that most planners and decision-makers are
accustomed to working with. For example, traditional approaches often include a Do Nothing alternative
from which the future build alternatives are compared. These alternatives tend to be grossly congested
and unrealistic, but this does not stop traditional models from generating traffic volumes and speeds for
the whole network. Because simulation models cannot over-assign vehicles to roadways, a Do Nothing
Best Practices in the Use of Micro Simulation Models
10
alternative may be infeasible to simulation or result in cascading queues that totally distort the
performance measures for a significant area. This issue can be addressed by restructuring the analysis to
talk about the maximum demand or growth that the Do Nothing alternative is able to accommodate or
by including a minimum list of improvements necessary to avoid system collapse.
2.2 Research Approach
This research focuses on an objective assessment of the state of the practice in the use of traffic micro
simulation tools by states and metropolitan areas for transportation planning studies. A major issue
faced by state agencies and MPOs is whether the extra overhead associated with developing, applying,
and maintaining micro simulation traffic models is warranted and justified. This study used a literature
review, web-based survey, peer exchange meeting, and case study interviews to determine if there is
consensus among transportation professionals with traffic simulation experience as to where, when,
and how micro simulation modeling can be best supported, justified, and cost-effective. The overall
objective of the research is to provide guidance to officials and policy-makers who are considering using
micro simulation models for future studies.
The following research questions were posed to establish a framework for identifying the pertinent
information that would best determine if a consensus exists regarding the appropriate use of micro
simulation modeling:
1. Under what circumstances have micro simulation modeling efforts been found to be warranted and cost-effective?
2. What are the benefits of microscopic simulation compared to the currently available macroscopic models?
3. In what ways have investing and conducting micro simulation led to better or more effective designs and investment decisions?
4. How have the data needed for micro simulation and time-dependent models been developed?
5. What steps are being taken to attract and retain staff with the necessary expertise and knowledge to perform and oversee micro simulation and dynamic traffic assignment modeling efforts?
6. What organizational or institutional arrangements have been useful in addressing any or all of the issues identified above?
The answers to these main research questions, based strictly on the responses obtained from
transportation officials with experience performing micro simulation modeling, are summarized in
Section 7, Synthesis of Findings and Lessons Learned.
2.2.1 Literature Review
A brief literature review of traffic simulation studies was conducted to help identify MPOs and state
agencies that would be good candidates for the online survey and case study interviews. The review also
helped to identify the state of the practice and key issues faced by micro simulation applications.
The results of this literature review are presented in Section 3 of this report.
Best Practices in the Use of Micro Simulation Models
11
2.2.2 Web-Based Survey
A web-based survey was developed to gather a variety of perspectives on the travel forecasting
practices employed by MPOs and state departments of transportation. Survey respondents were asked
to assess when and under what circumstances micro simulation had been beneficial to their planning
efforts.
Survey invitations were extended to members of the peer exchange panel, registrants of the 2009
Integrated Corridor Management workshop, modelers at 110 MPOs, and members of the Travel Model
Improvement Program (TMIP) list-serve. In total, 125 transportation professionals participated in the
survey.
The results of the web-based survey are presented in Section 4 of this report.
2.2.3 Peer Exchange Meeting
As part of a related task, the Transportation Research Board (TRB) invited 20 transportation
professionals with a variety of experience related to simulation models to a peer exchange meeting at
the Beckman Center in Irvine, CA, on September 12-13, 2009. The exchange consisted of presentations
and roundtable discussions about their experiences in using simulation models for transportation
planning and operations studies.
The underlying objective of the peer exchange meeting was to develop general guidelines for how to
identify or select the best modeling tool to use for a given type of study, with a focus on the network
simulation tools used for transportation planning or operational analysis.
2.2.4 Case Study Interviews
Based on the results of the literature review, web-based survey, and peer exchange meeting, a list of
case studies were identified for more in-depth analysis. Telephone or in-person interviews were
conducted with the key personnel responsible for each study. These interviews helped to clarify the
facts and opinions related to studies where traffic simulation models were found to be cost-effective or
not cost-effective.
Best Practices in the Use of Micro Simulation Models
12
3 LITERATURE REVIEW
A review of the literature documenting traffic simulation studies conducted by or for public agencies
identified a number of key issues faced by micro simulation applications. Other materials summarized
the state of the practice from a number of perspectives. The key documents reviewed by this project are
listed below:
1. Boyles, S., Ukkusuri, S., Waller, T., Kockelman, K. A Comparison of Static and Dynamic Traffic Assignment Under Tolls: A Study of the Dallas-Fort Worth Network. Transportation Research Board 85th annual meeting, 2006.
2. ITE. A report on the use of traffic simulation models in the San Diego region, 2004.
3. Austroads. The use and application of micro simulation traffic models, Report AP-R286, 2006.
4. Transport for London. Micro simulation modeling guidance notes for Transport for London, Transport for London, 2003.
5. FHWA. Traffic analysis toolbox volume III: guidelines for applying traffic micro simulation modeling software, FHWA-FRT-04-040, Federal Highway Administration (FHWA), Washington, DC, 2004.
6. Halcrow Group Ltd. The use and application of micro simulation traffic models, Halcrow Group, London, 2003.
7. Fox, K. Is micro-simulation a waste of time? Proceedings of the European Transport Conference 2008, Leeuwenhorst Conference Centre, The Netherlands, 6-8 October, 2008.
8. May, A. Traffic flow fundamentals, Prentice-Hall Inc., New Jersey, 1990.
9. The SMARTEST Project. Best Practice Manual. RO-97-SC.1059. 1999.
10. Caltrans. Guidelines for Applying Traffic Micro Simulation Modeling Software. 2002.
11. Hoogendoorn, S.P. and Bovy, P.H.L. State-of-the-art of vehicular traffic flow modeling, Journal of Systems and Control Engineering, 215, 283-303, 2001.
12. Liu, R. and Hyman G. Generic Guidance for Modeling Merges, Proceedings of the European Transport Conference 2008, Leeuwenhorst Conference Centre, The Netherlands, 6-8 October, 2008.
13. NCHRP. Highway Traffic Data for Urbanized Area Project Planning and Design. Washington, DC: Transportation Research Board, 1982.
14. TRB. A Primer for Dynamic Traffic Assignment, ADB30 Network Modeling Committee, Transportation Research Board, 2009.
The insights gleaned from these documents that are of particular interest to this study include those
highlighted below.
Austroads (2006) categorizes the benefits of micro simulation over traditional traffic analysis techniques
in three main areas: clarity, accuracy, and flexibility. These benefits are discussed below.
Best Practices in the Use of Micro Simulation Models
13
Clarity. A comprehensive real-time visual display and graphical user interface (GUI) illustrates traffic operations in a readily understandable manner. The animated outputs of micro simulation modeling are easy to understand and simplify checking whether the network is operating as expected and driver behavior is being modeled sensibly. With micro simulation, what you see is what you get. If a micro simulation model does not look right, then it probably is not right and vice versa.
Accuracy. By modeling individual vehicles through congested networks, the potential exists for more accurate modeling of traffic operations at complex and simple intersections or merges. Individual drivers of vehicles make their own decision on speed, lane changing, and route choice, which could better represent the real world than other modeling techniques. For example, analytical and mesoscopic simulation models often use fixed values of saturation flows and all vehicles are assumed to behave in the same manner. In contrast, micro simulation models represent individual vehicles and detailed networks. A parameter such as the saturation flow can actually be an output of the model.
Flexibility. A greater range of problems and solutions can be assessed than with conventional methods, e.g., vehicle-activated signals, demand-dependent pedestrian facilities, queue management, public transport priorities, incidents, toll booths, road works, signalized roundabouts, shock waves, incidents or flow breakdown, or slip road merges. The interaction between different vehicle types and with other modes (bus, tram, and light rail) can all be represented.
The SMARTEST project (1999), through case examples, showed that macroscopic simulation models can
be successfully applied to several Intelligent Transportation System (ITS) applications. However, the
study concluded that this requires a level of innovations that only users with complete understanding of
how each model works can possess.
Caltrans (2002) lists the following as study conditions where micro simulation models are desirable:
Conditions that violate one or more basic assumptions of independence required by HCM models Queues spill back from one intersection to another Queues overflow turn pockets Queues from city streets back up onto freeways Queues from ramp meters back up onto city streets
Conditions not covered well by available HCM models Queue spill-back Multi-lane with traffic signals or stop signs Truck climbing lanes Short through lane adds or drops at a signal Boundary points between different signal systems operating at different cycle
lengths Signal pre-emption (e.g., railroad crossings and fire stations) HOV lane entry options or design options for starting or ending an HOV lane Two-way left turn lanes (however, currently no commercially available micro
simulation software can model this) Roundabouts
Best Practices in the Use of Micro Simulation Models
14
Tight diamond interchanges Incident management options (Because HCM and macroscopic models assume a
steady-state condition within each analysis period, they are not well suited to accurately track the build-up and dissipation of congestion related to random transitory conditions caused by incidents.)
Choosing among alternatives, none of which eliminates congestion
Testing options that change vehicle characteristics and driver behavior
Transport of London (2003) lists the following issues as being suitable for microscopic simulation models:
Complex traffic operation schemes (e.g., bus priority, advanced signal control, incident management, different modes of toll collection)
Significant conflicts among different road users (e.g., pedestrians, cyclists, buses)
Major traffic movement restrictions (e.g., lane closures, one-way system, toll plazas)
Politically sensitive projects that could benefit from visualization
Planning and design of high-value projects with potential large savings if detailed microscopic simulation models are prepared
Emulation of the operation of a dynamic signal control system, with a simulated network driven directly by the control system and with significant saving in signal timing preparation and optimization
Town center studies
Tram and light rail operations
The literature also points out a number of limitations and concerns about using simulation models for
planning studies. These concerns typically focus on:
High costs
Data requirements
Lane changing phenomena
Results extraction
For example, Fox (2008) expressed the opinion that a pure microscopic simulation approach is totally
inappropriate and that multi-resolution (macro, meso, and micro) should be used instead. The basis for
his comment seems to stem from watching traffic animations and concluding that there are problems
with car-following, lane-changing, and gap-acceptance models, because animated vehicles were seen as
moving at inappropriate times. It also reflects the importance of modeling the impact of simulation
results on travel demand in the overall area.
Hoogendoorn and Bovy (2001) claim that car-following rules in virtually all microscopic simulation
models cannot properly reproduce the congested traffic flow phenomenon. Liu and Hyman (2008) have
shown that the gap-acceptance models tend to underestimate ramp capacities and thus overestimate
delays due to merging and weaving.
Finally, it is always beneficial to restate Dr. May’s observations regarding the use of micro simulation
(May 1990):
1. There may be easier ways to solve the problem; consider all possible alternatives.
Best Practices in the Use of Micro Simulation Models
15
2. Micro simulation can be time-consuming and expensive; do not underestimate time and cost.
3. Micro simulation packages require considerable input characteristics and data, which may be difficult or impossible to obtain.
4. Micro simulation applications or models require calibration, validation and verification, or auditing, which if overlooked could make the model useless.
5. Development of simulation models requires knowledge in a variety of disciplines, including traffic flow theory, computer programming and operation, probability, decision-making, and statistical analysis.
6. Micro simulation is difficult unless the model developer fully understands the software platform.
7. The micro simulation package may be difficult for non-developers to use because of lack of documentation or unique computer facilities.
8. Some users may apply micro simulation packages and treat them as black boxes and really do not understand what they represent.
9. Some users may apply simulation models and not know or appreciate model limitations and assumptions.
Best Practices in the Use of Micro Simulation Models
16
4 WEB-BASED SURVEY
A web-based survey was developed to gather a variety of perspectives on the travel forecasting
practices employed by MPOs and state departments of transportation. In particular, survey respondents
were asked to assess when and under what circumstances micro simulation had been beneficial to their
planning efforts. Information derived from the survey is descriptive of the methodologies used and
many of the application details. While this information documents the state of the practice, it does not
reveal whether the models produced accurate forecasts. Respondents were guaranteed confidentiality
for the web-based survey and the follow-up interviews with selected agencies or personnel.
The survey included 76 questions organized into three main sections: general questions about micro
simulation applicability, experience using simulation tools on planning studies, and general context
questions. The 15 general questions were designed to obtain descriptions of the state of the practice in
the use of micro simulation tools. The 58 project-related questions were divided into seven sections,
covering the model development and application process, in the following order:
1. Project Scope and Objective
2. Data Preparation and Conversion
3. Model Implementation and Calibration
4. Model Application
5. Model Visualization and Presentation
6. Data Management and Maintenance
7. Conclusions
The final section contained three questions about the size and type of the respondent and the size of
the region served.
Survey invitations were extended to the members of the peer exchange panel, registrants of the 2009
Integrated Corridor Management workshop, modelers at 110 MPOs, and members of the TMIP list-
serve. In total, 125 transportation professionals participated in the survey and 41 responded to the full
set of questions. The range of responses is summarized below.
4.1 Use of Simulation Tools in the Industry
Most respondents (89 percent) have conducted simulation studies in some way or another. The lack of
experienced and trained staff is the top reason why the remaining respondents (11 percent) did not
pursue a simulation study. Reasons to forego simulation, in order of importance, include the following:
1. Lack of experienced and trained staff
2. Lack of understanding of advanced models and their capabilities
3. Data requirements or collection cost
4. Software expense
5. Model implementation costs
6. Model calibration/validation costs
7. Model maintenance and upkeep costs
Best Practices in the Use of Micro Simulation Models
17
8. Software immaturity or lack of a proven track record
9. Software inability to generate the required outputs
10. Output data post-processing difficulties
One respondent commented:
“Management does not see the benefits of conducting microscopic simulation studies. Most of our [MPO] studies are regional or subarea level, which is too complicated for microscopic simulation. ITS, signal control, or other studies are generally conducted by cities or counties or their consultants.”
None of the respondents have rejected a given simulation software tool for use in planning studies. On
the other hand, no public agency respondents had experience with MATsim, Paramics [SIAS], or
WATsim. The software packages most commonly used by the majority of respondents include VISSIM,
Synchro, SimTraffic, and CORSIM.
Most respondents identified Synchro as the software of choice for signal coordination and optimization
and traffic operations analysis. There was no consensus, however, about the software packages that are
best suited for the following types of applications:
Congestion pricing
Highway improvement/new infrastructure
Construction and work zone activities
Integrated corridor management
Regional modeling/planning studies
Incident management
Transit operations
Multimodal planning
This was exemplified by the range of responses provided for each application. In fact, the plurality of
respondents expressed a “no preference” on the choice of software.
There is a strong correlation between ownership of a software package and the agency’s tacit
preference. In the majority of cases, software ownership dictates or limits the choice of software for a
given study.
During the past 5 years, most respondents have used micro simulation tools for basic traffic studies such
as traffic operations analysis and signal coordination and optimization. Relatively few respondents have
used micro simulation tools for congestion pricing, incident management, transit operations,
multimodal regional planning, and impact assessment studies.
Simulation tools were found to be helpful in a broad range of study types, including 91 percent of traffic
operations studies, 83 percent of facility design studies, 82 percent of planning studies, 68 percent of
demand management studies, and 60 percent of investment decision and policy studies (Figure 1).
Best Practices in the Use of Micro Simulation Models
18
Figure 1: Types of Studies Where Micro Simulation Has Been Helpful
Based on the comments submitted as part of the questionnaire, it appears that the simulation tools
have been most helpful in providing policy-level officials and the public with information, through
animation, that is readily understood and appreciated. A few representative comments include:
“One of the best tools for effective communication of analysis results.”
“Many people with non-engineering background could possibly have a hard time differentiating the capabilities of demand models vs. simulation models. Therefore, providing visual tools such as a 3D animation of the project in public meetings has shown to be extremely beneficial.”
Another aspect in which simulation tools have been helpful is their ability to answer questions
concerning the impacts of a traveler’s behavior or choices on system performance that are not captured
by traditional models. This is accomplished by the simulation tool’s ability to extract performance
measures, such as person throughput, person delay, and fuel consumption, calculated from individual
vehicle behavior. These measures can be used to assess traffic flow improvements that would not be
captured in vehicle miles traveled (VMT)-based statistics.
A particularly ardent supporter of microscopic simulation tools provided the following comment:
“Microscopic simulation is the next best thing to actually building the project to understand operations.”
Other analysis tasks where microscopic simulation tools were found to be helpful include:
Providing reliable data to assist planning and design
Providing evidence-based policy formulation and development
29%
24%
27%
30%
44%
53%
62%
7%
26%
34%
38%
39%
30%
29%
21%
32%
27%
23%
10%
10%
5%
7%
11%
7%
4%
5%
36%
7%
5%
5%
0% 20% 40% 60% 80% 100%
Other Studies
Policy Analysis
Investment Decisions
Demand Management
Planning Studies
Facility Design
Traffic Operations
Percent of Respondents
For what types of studies has micro simulation been helpful?
Very Helpful Helpful Neutral Not Helpful Not Applicable
Best Practices in the Use of Micro Simulation Models
19
Assessing efficacy of potential improvements
Addressing issues that cannot be studied using traditional planning models
Evaluating weave/merge movements at interchange ramps
Identifying potential design problems early in the process
Evaluating factors such as platoon effects, length of pocket lanes, impact of long cycle lengths on queuing, driveway impacts, and raised medians
Setting the policy for the given project
On the other hand, some respondents point to the huge amount of time, cost, and upfront data
required to develop and run microscopic simulation models as a significant roadblock. Others expressed
disappointment due to the lack of documentation regarding the model’s procedures and features and
the incongruence between the software’s default parameters and the agency’s acceptable range of
values. Respondents also expressed evident disappointment with the lack of guidance by software
vendors with respect to model calibration and validation.
Also, some respondents were skeptical about the ability of microscopic simulation tools to accurately
model over-saturated facilities. Others expressed concern about the appropriateness of using
microscopic simulation models given their limited geographical scale and computational capacity when a
particular project’s impacts are expected to be regional.
The following comment characterizes the sentiment against microscopic simulation models:
“Microscopic simulation tools do not sufficiently integrate into the full systems analysis to withstand any methodological scrutiny.”
4.2 Linking Simulation with Regional Model
Recently, there has been significant interest in having versatile and integrated tools that can be used for
planning and operational analysis. The industry sees these tools as a means to coordinate the efforts of
planners and operators to help ensure that regional transportation investment decisions consider a wide
variety of strategies in order to meet regional goals and objectives. In short, these multi-purpose tools
are vital to improving transportation decision-making and the overall effectiveness of transportation
systems. One way to develop such a tool is by linking the regional travel demand forecasting model with
one or more network simulation tools. For the purpose of this study, three types of linkages were
investigated:
Type 1 – Planning Model Network Simulation Tool. In this case, demand data or trip tables are fed into the simulator to perform time-dependent traffic assignments.
Type 2 – Planning Model Network Simulation Tool. This approach includes an iterative feedback loop between the micro simulator and demand models to generate an equilibrated result.
Type 3 – Simulation Model Planning Model. This approach uses the detailed performance measures generated by the simulation software to refine the network speeds or enhance the travel times used in demand modeling.
The majority of respondents (57 percent) indicated that they currently have a simulation tool linked to
their planning model in some fashion. Furthermore, of those without a linkage, 36 percent plan to
Best Practices in the Use of Micro Simulation Models
20
integrate such tools in the near future. Transportation agencies, on the other hand, appear to be well
aware of the importance of linking a simulation tool with their regional planning model. Of the agencies
that currently do not have a link between their planning model and their simulation tool, 50 percent
plan to have one in the future.
Most respondents (86 percent) with a simulation tool linked to their planning model employ a Type 1
linkage. Few (11 percent) employ a Type 2 linkage, and almost none (3 percent) employ a Type 3 linkage.
The majority of respondents (57 percent) use TransCAD as their planning model followed by CUBE
(19 percent) and EMME2 (11 percent).
The survey responses and comments revealed that there are a number of platforms that offer a linkage
between macroscopic and microscopic models. Examples of these (obtained from the survey) include:
TransCAD TransModeler
VISUM VISSIM
DynusT VISSIM-VISUM
CUBE Dynasim
CUBE CUBE Avenue
Other linkages between macroscopic and microscopic tools have been typically developed by software
developers or users. Some examples (obtained from the survey) include:
EMME/2 TRANSIMS
CUBE TRANSIMS
TransCAD TRANSIMS
TransCAD VISTA
Respondents who have no plans to link their planning model to a simulation tool are mostly discouraged
by the difficulties in designing the data exchange mechanism. The following comment illustrates the
complexity and the effort involved in linking a planning model to a microscopic simulation model:
“In most cases, we link the [macroscopic] travel demand model to a mesoscopic scale network assignment model such as VISUM. VISUM is used for network processing and trip table estimation. VISUM is seamlessly linked to VISSIM as well as Synchro. The Synchro model is used as a database warehouse for the entire network and all traffic control information plus it is used to provide optimized signal timings. O-D trip tables and paths are exported from VISUM to the simulation model and an iterative process is set up to refine the trip tables and paths based on how well the simulation model validation matches peak 15-minute, peak hour, and peak period traffic volumes, queues, and travel times. Only by accurately matching all three parameters can we be assured that the trip table we estimated is the correct one. This process starts with a seed O-D matrix from the macro level model but we may iterate this process to feed back speeds to the macro level model from the simulation model when we find glaring mismatches.”
The following two comments raise several practical and theoretical issues that need to be addressed
when attempting to integrate microscopic simulation models with planning models.
Best Practices in the Use of Micro Simulation Models
21
“30-year forecasts generate demands that make alternatives analysis with microscopic simulation models meaningless.”
“I generally don't feed back microscopic simulation travel times to the demand model because the demand model was not originally calibrated for this kind of input.”
The biggest challenge in linking the planning model to the microscopic simulation model is fitting or
estimating O-D demand to link counts and carrying these adjustments into estimates for future years.
Another challenge to integration is the difference in the way traffic is loaded onto the network. For
instance, planning models typically use zone connectors to load traffic onto a collector and above
network. Microscopic simulation applications typically model a subarea within this network, but need all
of the actual entry points from the local streets and parking lots to accurately simulate travel conditions.
Moreover, the chances of performing a successful linkage are greatly hampered by a lack of good
understanding of the regional model, the nature of dynamics between the macro-level and micro-level
models, their assumptions, and a technology plan to deal with possible computing issues. Another
challenge stems from the tendency to select software based on independent capabilities rather than its
interoperability or compatibility with other software packages. For example, many of the respondents
use TransCAD as their planning model and VISSIM as their simulation tool. Respondents felt that
linkages between TransCAD and VISSIM are significantly more difficult than linkages between TransCAD
and TransModeler or VISUM and VISSIM.
Another issue is the lack of prospective projects that require the use of such linked models. It is difficult
to justify the effort spent on such a linkage when relatively few projects exist to sustain a combined
model. The following comment illustrates this point:
“The biggest challenge is staff training and continued use of the software. ‘Use it or lose it’ is relevant—one must make sure there is a real project to do before training staff in the software.”
Other difficulties include:
Inadequate funding and time issues
Lack of evidence that a meaningful translation can be generated for congested networks
Lack of confidence or faith that the results would represent reality
Fear that simulation models would contradict or magnify the results of the macro model to the point that project justification may no longer hold
Computing power/hardware needs
Network scale
Availability of data to support calibration
4.3 Application Experience
Respondents were asked in this section of the survey to select the most complex simulation study they
were recently involved in and answer questions related to it. The major findings are presented below.
Best Practices in the Use of Micro Simulation Models
22
4.3.1 Project Scope and Objectives
The majority of respondents (65 percent) had the chance to serve as a modeling lead, 50 percent as a
consultant or advisor, and 43 percent as a stakeholder or decision-maker on one or more planning
studies that used micro simulation software.
Based on the types of projects undertaken, the web survey revealed that respondents were using
simulation methods to conduct complex studies. Moreover, several respondents were seriously
evaluating the applicability and usefulness of dynamic traffic assignment (DTA) and simulation models
for proof-of-concept studies. The reported experience can be broadly grouped into three major
categories: design and operational improvements, regional impact analysis, and proof-of-concept
studies. Below is a list of the types of simulation projects described by the web survey participants.
Design and Operational Improvements
Create a new roadway design that will function well into the future
Analyze the present and future operations of a complex section of congested interchanges where two interstates and another major highway come together
Test the effectiveness of different operational strategies to reduce the congestion caused by a construction project
Develop a remedy for a traffic bottleneck
Analyze different cross-section alternatives for a highway improvement project including motor vehicle interactions with bus transit, bicyclists, and pedestrians
Assess traffic operational measures around a work zone
Assess the impact of managed lanes on level of service
Assess the impact of route choice behavior due to a major freeway widening project
Assess the effectiveness of real-time ITS traffic management strategies in reducing congestion
Maximize person-throughput for recurrent and non-recurrent congested conditions
Determine the optimum operational strategy for ramp metering, active traffic management, and variable speed limits to improve mobility
Regional Impact Analysis
Visually demonstrate the impact of recent military base re-alignments and closures on the surrounding transportation system
Assess the capacity and reliability of the transportation system to accommodate projected demand
Assess the impact of removing one of the major arteries on the area street network
Proof-of-Concept Studies
Assess the impact of significant route and mode choice changes on travel demand estimated by the long-range planning model
Assess the applicability of using planning-level trip tables as input to microscopic simulation models
Best Practices in the Use of Micro Simulation Models
23
Assess the ability of DTA models to represent route, mode, and departure time choices realistically
Use micro simulation to evaluate the performance of a planning-level traffic assignment model
Determine if improvements can be made to the traffic assignment and volume-delay functions in a planning model to generate more accurate trip tables for microscopic simulation
Assess whether DTA models can be applied effectively at a regional scale
Forty-four percent of studies described above had a project budget of less than $250,000, 20 percent
had a budget between $250,000 and $500,000, 18 percent had a budget between $500,000 and
$1 million, and 16 percent had a budget greater than $1 million. Half of the simulation studies are still
ongoing, with the remaining projects usually requiring 15 months on average to complete.
The majority (64 percent) of the simulation studies were conducted at the corridor or subarea level. This
could be influenced in part by the fact that the survey was distributed to the Integrated Corridor
Management workshop registrants, but it also highlights the difficulties in applying a microscopic
simulation tool at a regional level. In this regard, only 27 percent of the projects were conducted at the
regional level.
Sixty percent of respondents claim that agency preference is the reason why microscopic simulation was
used (Figure 2). In addition, 59 percent of the respondents reported that traditional modeling tools did
not provide the necessary level of policy or analysis sensitivity, and 48 percent reported that visualizing
traffic operations for public outreach and decision-making were key reasons for choosing a micro
simulation approach. Other reasons include contractor preference (21 percent) and funding incentives
(8 percent). In Washington and California, Interchange Justification Report Guidelines and Corridor
System Management Plans (CSMPs) mandate the use of simulation tools.
Figure 2: Reasons for Using Micro Simulation Models
0% 10% 20% 30% 40% 50% 60% 70%
Funding incentives
Other
Contractor preference
Visualizing traffic operations was important for …
Traditional modeling tools did not provide …
Agency preference
Percentage of Studies
Why was simulation software used for this study?
Best Practices in the Use of Micro Simulation Models
24
Most of the simulation applications were relatively complex. More than half of the studies accounted for
items not normally present in traditional models such as pre-timed and actuated signals, merging and
weaving, time-of-day differences, and transit interactions. A considerable number of studies
(40 percent) included HOV lanes, stop signs, dual-ring signals, signal progression, turn prohibitions,
and/or incidents in their modeling process. Only a few studies considered signal preemption, high-
occupancy toll (HOT) lanes, pedestrians and bicycles, bus lanes, traveler value of time, lane closures,
transit-only lanes, pre-trip information, variable message signs, reversible lanes, work zones, and
roundabouts.
4.3.2 Data Preparation and Conversion
Most respondents (86 percent) created the simulation network as part of the study. In other words, a
pre-existing simulation model did not exist. This can be a sign either that this is the first simulation
model to be developed or that simulation models are generally not sustainable or maintained such that
each new study must start from scratch.
TransCAD was the primary source of network data in 34 percent of the cases, followed by CUBE
(24 percent) and EMME2 (21 percent). In addition, GIS street layers; Google, Yahoo, and MSN maps;
aerial photos; and Synchro networks were used to supplement the traditional network data. Basic
network geometry and signal control checks were conducted in most cases. In particular, facility type
and the number of lane checks were almost always conducted. Most respondents (75 percent) also
checked basic link and node attributes including signal timing plans and progression offsets, saturation
flow rates, turn prohibitions, and allowed movements.
The majority of respondents (72 percent) did not have the demand data available from an earlier study
(one-off study); however, most respondents (85 percent) had regional trip tables available. For these
studies, a two-step approach was typically used to generate the time-dependent O-D demand. A time-
dependent seed matrix is first synthesized from the planning-level trip tables. Then an O-D estimation
procedure is used to adjust the seed matrix to match observed link traffic counts. Time-dependent
demand data was synthesized from hourly planning-level trip tables in 68 percent of the studies. In most
cases (77 percent), the synthesized demand data was later adjusted using O-D estimation techniques.
In a few cases, other information was used to develop or refine the demand data. This included:
Detector measurements
Peak spreading models
DTA/mesoscopic assignment results
Signal optimization plans
Activity-based demand model output
Travel diaries
Referenced global positioning system (GPS) information
Most respondents (83 percent) used either PM or AM peak periods in their modeling process with more
than half using only the PM/AM peak hour (Figure 3). Few (17 percent) used 24-hour or off-peak
Best Practices in the Use of Micro Simulation Models
25
periods, which could be attributed to memory or software limitations that prevent modeling for
extended time periods. It was interesting to note that very few studies modeled weekend conditions.
Figure 3: Temporal Dimensions of Micro Simulation Models
On average, 68 percent of the data preparation and conversion effort was performed by contractors. In
51 percent of the studies the task was shared by clients and contractors, and the client was solely
responsible for the effort in 14 percent of the studies. The data preparation and conversion costs ranged
from 5 percent to 60 percent of the total project budget, with an average of 27 percent. The time
required to finish this task varies from 1 week to more than 1 year, with an average of 13 weeks.
4.3.3 Model Implementation and Calibration
There was no consensus among respondents regarding what calibration and validation criteria should be
adopted and the appropriate level of rigor. For example, some respondents explicitly calibrate for queue
lengths whereas others may only visually inspect resulting queues up to the point of their analyst’s
satisfaction. Some agencies used linear regression statistics such as the R2, while others used the
Geoffrey E. Havers (GEH) statistic. However, most agencies appear to have an adopted a set of
calibration and validation guidelines.
The majority of respondents (56 percent) used link counts available at 15-minute resolutions in the
calibration and validation process. Link counts at higher resolutions (less than 15 minutes) and turning
movement counts were also used by a considerable number of respondents. Few respondents had O-D
vehicle counts and an even smaller number had subarea O-D matrices from GPS-based trip information.
The majority of respondents (70 percent) reported that the calibration results were at least satisfactory for (see Figure 4):
Link capacities, volumes, delays, speed, and travel time
Turning movement fractions
Path travel times and distances
Transit ridership
0% 10% 20% 30% 40% 50% 60% 70% 80%
Weekend
Off-Peak Period
Daily
Weekday
AM Peak Hour
PM Peak Hour
AM Peak Period
PM Peak Period
Percentage of Studies
What were the simulation periods considered for this study?
Best Practices in the Use of Micro Simulation Models
26
Bike/pedestrian movements
Figure 4: Level of Satisfaction with Calibration and Validation
However, this is not to say that implementing the simulation model was an easy task. The majority of
respondents (65 percent) reported that accurately modeling traveler behavior and achieving reasonable
calibration were the major difficulties encountered while implementing the simulation model (Figure 5).
Furthermore, 48 percent of respondents reported difficulties in their ability to model a large urban area
network, and 28 percent of respondents had difficulty achieving convergence or stabilization. Eighteen
percent of the respondents reported difficulties in obtaining HCM-consistent results and in developing
visualization results for stakeholders.
15%
21%
23%
27%
31%
32%
33%
33%
36%
41%
44%
44%
60%
54%
50%
54%
57%
59%
46%
48%
52%
59%
38%
48%
50%
20%
15%
14%
15%
17%
7%
14%
11%
11%
14%
9%
20%
15%
14%
8%
7%
7%
5%
7%
0% 20% 40% 60% 80% 100%
Transit Ridership
Bike/Pedestrian …
Truck/Bus Operations
Travel Distances
Link Capacities [Flow …
Travel Times
Turning Movements
Speed Distribution
Link Delays
Link Volumes (15 min)
Queue Lengths
Screenline Volumes
Link Volumes (Hourly)
Percentage of Studies
How satisfactory were the calibration results for the following?
Very Satisfactory Satisfactory Somewhat Satisfactory Not Satisfactory
Best Practices in the Use of Micro Simulation Models
27
Figure 5: Major Difficulties Implementing Micro Simulation Models
Some of these difficulties were related to computing power (an insufficient number of available
machines) and memory resources. The inability to model large networks was a direct consequence of
these shortfalls. For these cases, a 64-bit operating system with more than 3 gigabytes of memory was
required. Other difficulties pertain to the software capabilities. Certain simulation tools are unable to
achieve convergent and stable solutions because of the way they handle minor flows between O-D pairs.
Other difficulties pertain to the lack of guidelines regarding queue lengths and bottlenecks. The FHWA
Traffic Analysis Toolbox Volume 3 indicates that bottleneck capacity calibration is to be done visually
and to the analyst’s satisfaction. The subjectivity of this guidance causes modelers to perform the
bottleneck calibration almost as an afterthought. Similarly, there are no guidelines on how to deal with
congestion points and hot spots at network boundaries and loading points, which can impact the
modeled results greatly. Without concrete directives, modelers find themselves modifying the proposed
strategy in order to improve the model’s results.
On average, 72 percent of the model calibration and implementation effort was performed by
contractors. In 52 percent of the studies the task was shared by clients and contractors, and the client
was solely responsible for the effort in 14 percent of the studies. Model calibration and implementation
costs ranged from 5 percent to 70 percent of the total project budget, with an average of 38 percent.
The time required to finish this task varied from 2 weeks to more than 1 year, with an average of
17 weeks.
0% 10% 20% 30% 40% 50% 60% 70% 80%
Obtaining HCM-consistent results
Developing visualization results for stakeholders
Other
Achieving convergence or stabilization
Ability to model a large urban area network
Accurately modeling traveler behavior
Achieving reasonable calibration/validation
Percentage of Studies
What were the major difficulties faced in implementing the model?
Best Practices in the Use of Micro Simulation Models
28
4.3.4 Model Application
On average, the reported studies evaluated six or more scenarios or alternatives using the simulation
model. Most respondents (78 percent) modeled operational and capacity improvement scenarios. A
considerable portion of the respondents (36 percent) also modeled demand management scenarios. On
the other hand, it was interesting to note that only a few respondents (11 percent) used simulation to
analyze what they would characterize as land-use or policy impact scenarios. This is a bit surprising given
that proponents of simulation models tout their unique ability to analyze the impact of policy changes
that cannot be modeled using traditional planning models.
Most respondents reported time-dependent link travel times, speeds, and volumes as the primary
performance measures. The majority also reviewed network-wide average speeds, time-dependent link
densities, and path travel times. A considerable number of respondents reported network-wide
statistics such as VMT, vehicle hours traveled (VHT), and total delay. Few considered queue lengths,
formation/dispersion rates for bottlenecks, and cycle failures.
All respondents expressed their satisfaction with simulation results, with 32 percent being very satisfied.
Most importantly, almost all respondents (94 percent) believed that the simulation tool answered the
study questions adequately. In comparison to traditional planning models, most respondents
(74 percent) believed that the simulation results produced more useful and accurate answers (Figure 6).
The majority of the respondents (54 percent) believed that simulation results led to better capital
investment decisions. A considerable number of respondents (34 percent) believed that simulation
results led to changing or reversing an existing design or policy.
Figure 6: Benefits of Using Micro Simulation Models
Other advantages over traditional planning models reported by the respondents include:
Buy-in by non-technical stakeholders
Better understanding of upstream–downstream impacts of capacity change
0% 10% 20% 30% 40% 50% 60% 70% 80%
Other
Changing or reversing an existing design/policy
Better capital investment decisions
More effective or accurate answers
Percentage of Studies
Have the simulation results led to:
Best Practices in the Use of Micro Simulation Models
29
More useful and helpful performance evaluation measures
Recognition that congested networks must be evaluated on a system-wide basis
Better integration of microscopic simulation throughout the capital outlay process
On average, 75 percent of the model application effort was performed by contractors. In 55 percent of
the studies the task was shared by clients and contractors, and the client was solely responsible for the
effort in 13 percent of the studies. Model application costs ranged from 5 percent to 75 percent of the
total project budget, with an average of 20 percent. The time required to finish this task varied from
3 weeks to more than 1 year, with an average of 24 weeks.
4.3.5 Model Visualization and Presentation
Visual animations were found to be more effective at communicating the results than charts or maps.
Most respondents (87 percent) used animation to present the simulation results to stakeholders. The
following comment illustrates the power of animations:
“Stakeholders were allowed to review the animation and ask for any view or time period to confirm their observations of traffic conditions. This created a high level of confidence with stakeholders.”
Moreover, most respondents (86 percent) believed that the visualization capabilities were at least useful
in presenting the simulation results, with a slight majority (53 percent) believing that they were very
useful. The majority of respondents (76 percent) also used charts, tables, screenshots, snapshots, and
GIS maps to supplement the animations.
On average, 77 percent of the visualization and presentation effort was performed by contractors. In
40 percent of the studies the task was shared by clients and contractors, and the client was solely
responsible for the effort in 14 percent of the studies. Visualization and presentation costs ranged from
5 percent to 40 percent of the total project budget, with an average of 9 percent. The time required to
finish this task varied from 1 week to 3 months, with an average of 4 weeks.
4.3.6 Data Management and Maintenance
Most of respondents (78 percent) maintained their simulation model after the study was completed.
Typically, all aspects of the model were maintained, including input data, calibration data, and scenarios.
Model maintenance costs were projected to range from $2,000 to $100,000 per year.
Lack of funding and staff time was the primary reason given by those who did not maintain their model.
Because most models are tailored to a specific project, there are relatively few situations where a given
simulation model can be developed to serve the needs of multiple projects or studies. In other words,
most simulation models were developed for a project and not as a strategic-level investment.
4.4 Clients vs. Contractors
The survey revealed that clients and contractors view each other’s effort differently. This is not
surprising because few clients understand what contractors are actually doing and contractors are not
privy to the number of hours spent by clients on their studies. Table 1 presents the cost, time, and level
Best Practices in the Use of Micro Simulation Models
30
of effort distributions for all respondents. Table 2 presents the same information filtered by respondents
from MPO and state transportation agencies.
In general, the distributions of cost and time durations are similar between all respondents and agency
respondents. Agencies perceive the cost and time for calibration and application to be slightly more and
data preparation and conversion to be slightly less than the group as a whole. The major differences are
in the level of effort estimates. Transportation agencies (i.e., clients) believe that they are much more
involved in all aspects of a project than the overall response suggests. In all cases, the percentage of
effort attributed to “only contractor” is three times greater for all respondents than it is for
transportation agency respondents. This suggests a significant misunderstanding or under-appreciation
by contractors of the contribution made by public agency clients related to the technical aspects of their
projects. Transportation agencies tend to view the work more as a shared effort than do contractors.
Table 1: Project Summary—All Respondents
Project Phase Cost (percent) Duration (weeks) Effort (percent)
Range Avg. Range Avg. Client Only
Contractor Only
Overall Split Client-Contractor
Data Preparation and Conversion 5-60 27 1-48 13 14 35 32-68 Model Implementation and Calibration 5-70 38 2-48 17 14 34 28-72
Model Application 5-75 20 3-48 24 13 33 25-75 Model Visualization and Presentation 5-40 9 1-8 4 14 46 23-77 Data Management and Maintenance N/A N/A N/A N/A N/A N/A 68-32
Table 2: Project Summary—Transportation Agency Respondents
Project Phase Cost (percent) Duration (weeks) Effort (percent)
Range Avg. Range Avg. Client Only
Contractor Only
Overall Split Client-Contractor
Data Preparation and Conversion 5-50 23 1-48 11 13 13 42-58 Model Implementation and Calibration 5-70 41 2-48 20 13 13 34-66
Model Application 5-75 21 3-48 40 8 8 32-68 Model Visualization and Presentation 5-40 7 1-8 5 8 17 27-73 Data Management and Maintenance N/A N/A N/A N/A N/A N/A 68-32
4.5 Lessons Learned
Most respondents (70 percent) that were involved in simulation studies believed that the simulation tools highlighted inadequacies in traditional modeling.
The primary concerns are that traditional models:
Best Practices in the Use of Micro Simulation Models
31
Are unable to adequately capture queues, bottlenecks, delays, and weaving or merging effects
Generally under-represent travel times
Lack animation or visualization tools
The following two comments reflect these concerns.
“Isolated location analysis using HCS did not show the effects of opening a bottleneck and the downstream traffic effects.”
“Truck-specific impacts could not have been analyzed at a detailed level without the simulation capability.”
There also was an insightful comment regarding the cost-effectiveness of microscopic simulations:
“Results highlighted the great dichotomy between micro simulation development costs and their usefulness in planning studies.”
Most respondents (94 percent) believe that a similar simulation study in the future will not be more
expensive than the current one, with 38 percent believing that the future study will be less expensive
primarily due to staff being more familiar with the software capabilities. This in turn could either reduce
the time required to perform the various tasks or increase their involvement, and hence reduce the total
project costs.
On the other hand, the reported studies highlight the need for better data. This is an issue that will likely
cause an increase in project costs in the future, which may or may not be offset by the potential cost
savings due to an increased familiarity with the software capabilities. Few respondents indicated that
project costs were kept at a minimum due to tight budgets and time deadlines, but that the next study,
without a tight deadline, will naturally be more expensive than the preceding one.
Most transportation agencies (75 percent) believe they now have the capability, staff level, and
expertise to conduct a similar simulation study on their own in the future. A few agencies, however,
believe that they have the necessary expertise, but not the resources; therefore, they may still require
help from a contractor, especially to make the initial network conversion. Very few agencies flatly admit
that they do not have the time or staff resources to conduct a simulation study of a similar scale.
Most respondents (81 percent) believe that the simulation study was cost-effective. Also, the majority
(73 percent) are actively retaining, training, and attracting staff to conduct or oversee micro simulation
studies in the future. Some respondents indicated that they have policies that discourage the retaining
of trained staff or believe that they do not have the funding to retain talented staff.
Moreover, the majority of respondents (68 percent) are willing to undertake more studies of this nature
in the future. However, a considerable number of respondents (21 percent) are likely to modify their
procedures, budgets, or schedules before the next study. Few respondents (5 percent) were not entirely
satisfied with their choice of software and are likely to consider another tool for their next study. Only
2 percent of respondents stated that they will be reluctant to use microscopic simulation software again
for a similar study.
Best Practices in the Use of Micro Simulation Models
32
5 PEER EXCHANGE MEETING
The overall aim of the peer exchange meeting was to develop a guide that will help practitioners select
the most appropriate modeling tool for a given type of project, with an emphasis on when it would be
best to implement network simulation tools.
The peer exchange participants are listed in alphabetical order below:
Behruz Paschai, North Central Texas Council of Governments, [email protected]
Brad Winkler, Michigan Department of Transportation, [email protected]
Brian Gardner, Federal Highway Administration, [email protected]
Brian Isaacson, Minnesota Department of Transportation, [email protected]
Chi Ping Lam, Houston-Galveston Area Council, [email protected]
Craig Hellman, Washington State Department of Transportation, [email protected]
Doug MacIvor, Caltrans, [email protected]
Eric Pihl, Federal Highway Administration, [email protected]
Guy Rousseau, Atlanta Regional Commission, [email protected]
John Kerenyi, City of Moreno Valley, CA, [email protected]
Karl Wunderlich, Noblis, [email protected]
Maren Outwater, Puget Sound Regional Council, [email protected]
Michael Mahut, INRO, [email protected]
Naim Rasheed, New York City Department of Transportation, [email protected]
Natarajan “Jana” Janarthanan, Fehr & Peers Associates, [email protected]
Ron Milam, Fehr & Peers Associates, [email protected] Scott Higgins, Portland Metro, [email protected]
V. Thera Black, Thurston Regional Planning Council, [email protected]
Yi-Chiang Chiu, University of Arizona, [email protected]
The meeting facilitators included:
Kimberly Fisher, National Academy of Sciences, [email protected]
Nanda Srinivasan, National Academy of Sciences, [email protected]
David Roden, AECOM, [email protected]
Hayssam Sbayti, AECOM, [email protected]
To prepare for the peer exchange, each participant was emailed the following:
1. Traffic Analysis Toolbox Volume I: Traffic Analysis Tools Primer. Publication No. FHWA-HRT-04-038, FHWA, 2004.
2. TRB. A Primer for Dynamic Traffic Assignment, ADB30 Network Modeling Committee, Transportation Research Board, 2009.
The stated objective for the meeting was:
The peer exchange will bring together agency technical staff and decision makers to
outline a model structure decision-making process given a wide variety of
considerations. The elements of a decision-making process may include a list of the
critical issues facing the agency, a summary of the available data, an appraisal of the
staff expertise available to build, run and maintain the future model, and consideration
Best Practices in the Use of Micro Simulation Models
33
of the types of models that would fit a given set of conditions. The objective of the peer
exchange is to assist transportation planners and decision makers select the best tool for
their needs.
The meeting was held on September 12-13, 2009, at the Beckman Center of the National Research
Council in Irvine, CA. The first day was spent reviewing the results of the web-based survey and sharing
the experiences of the peer exchange participants on studies involving simulation models. The second
day was organized around four general discussion questions:
1. What should be considered when choosing a model type?
2. What is the range of potential model types given their strengths and weaknesses?
3. What are the steps to making a model decision? Can we draw it?
4. What would you advise for each step of the decision process?
5.1 Words of Advice
The discussion was mostly free-flowing, with participants sharing their experiences managing the
modeling phase of past projects. However, due to the complexity of the issues and the many caveats
that impact a decision, the group had difficulty developing a consensus about the best approach for
making a model decision. Notwithstanding, the themes that emerged repeatedly from participants’
accounts are discussed below.
Consult more experienced agencies before selecting a model type. Public agencies
could consult with fellow agencies, experts, and reputable analysts to select the most
appropriate approach or software package before modeling a project’s likely impact,
especially given budget and time constraints.
Develop guidelines for selecting and operating model types. The public agency could
take the opportunity to develop a criteria list for adopting one model type over another
based on its findings from consulting with more experienced agencies during the pre-
selection phase. In addition to adhering to the list during selection, agencies should
make a list of best practices in micro simulation and follow them during implementation
in order to realize any meaningful return on investment and ensure studies are
conducted in an efficient manner.
Continue interagency cooperation during initial phase. Extensive peer review is still
needed to assess the soundness of the modeling methodology after selecting the model
type and during the development process. It is also good practice to receive outside
feedback while testing the model. Practitioners at public agencies raised the point that
this cooperation gives them a better understanding of when to apply certain software
programs for modeling various types of projects and consequently reduces the
likelihood that they could be misinformed by software developers and consultants.
Cost increases geometrically with the level of detail required. Agencies need to be
aware of the full cost (the cost of acquiring and operating the software program over a
full cycle) of developing mesoscopic and microscopic simulation models. On a per-link
Best Practices in the Use of Micro Simulation Models
34
cost, mesoscopic simulation models tend to cost an order of magnitude (i.e., ten times)
more to develop than macroscopic models. On a similar scale, microscopic simulation
models tend to cost an additional order of magnitude more to develop than mesoscopic
simulation models on a per-link basis.
Consequences of selecting the wrong model type. Selecting a model with less than the
level of fidelity required to satisfactorily conduct the analysis, particularly to adhere to
budget and time constraints, could be detrimental in the long run because it could lead
to legislators and decision-makers making misplaced investment decisions based on an
inappropriate modeling approach. If a more detailed level of modeling is required that is
beyond available resources, agencies should discuss the tradeoff between increasing
their budget limits and revising cost estimates now versus the risk of building
inadequately scaled infrastructure later.
Size of the area to model will constrain the model type. Micro simulation is helpful in
modeling travel in corridors, but less so for regional studies. The modeler’s desired size
of the network, temporal scale, and travel demand load determine to a large extent the
class of simulation models that can produce adequate results. Network size is based on
the number of links, nodes, and O-D pairings. Mesoscopic and DTA models are better
equipped to handle large-scale projects, generally those with 15,000 links, 5,000 nodes,
1,000 O-D pairs, and 1 million vehicles. Conversely, it is recommended that microscopic
simulation models, to be cost-effective, be confined to an area significantly less than
regional in size. This is generally on a scale of 50 to 200 nodes and tens of thousands of
vehicles, although multi-threading and parallel computing can stretch the simulation
model’s area of analysis much larger. The level of effort and data requirements for
accurately modeling a large area using microscopic simulation should also be
considered.
Specificity of the research topic and planning horizon will impact the decision. The
divergence in scale is because microscopic models are more appropriate for answering
questions pertaining to a specific issue relevant to a limited area whereas macroscopic
models should be used to determine a general measure, such as the network’s average
performance. If accuracy and specificity are required, then microscopic simulation
models are more suitable than macroscopic models. Applications that fall in-between
these requirements are probably better suited for mesoscopic models. In addition,
macro scale models were deemed far more appropriate for long-range plans, usually in
the 30-year range, due to the fact that they require far fewer assumptions to produce
adequate results as opposed to micro simulation models. Micro simulation models were
found to be useful in planning for short time periods, such as rush hour, in the near
future.
Collect data now for the eventual adoption of a micro simulation model in the future.
Agencies should start developing and standardizing procedures for collecting and
estimating time-dependent trip tables along with those used to estimate trip tables
used at a planning level. The standardization of the data should ease the transition from
Best Practices in the Use of Micro Simulation Models
35
macroscopic to mesoscopic and finally microscopic level modeling. Traffic data should
be collected during the scoping stage of engineering projects. Specific line-item budgets
should be included in the project plan to fund data collection. In this way, whenever a
more comprehensive study is needed, good quality data are already available. Also,
automated data collection techniques should be used whenever possible. The extra
costs associated with using these resources will be easily offset by the quality and
consistency of the collected data, which should help lower the cost and hours required
for calibrating and validating afterwards.
Pilot test micro simulation models. Agencies should adopt a “proof of work”
philosophy, which means testing and evaluating micro simulation models in short
phases, possibly on smaller pilot projects, before deciding to adopt a tool for a major
project or as a requirement for all future projects. For example, an agency could use a
simulation tool for a 3-month case study on a previously evaluated corridor, and if
successful, add more corridors until a regional model can be developed.
Phase in the adoption of micro simulation modeling. Agencies considering using
microscopic simulation approaches should do so in stages. It is advisable to do
incremental improvements to existing models, with major improvements every 1 or
2 years. A starting point would be to use DTA instead of static assignment in variants of
four-step models. This could later be expanded to include increasing levels of detail and
fidelity. There are several benefits of implementing micro simulation in phases. It gives
modelers time to gain confidence and comfort in the concepts and assumptions in the
model, learn what data are needed and how to collect the data, and develop effective
methods for explaining and demonstrating the results of the model to non-users. For
politically sensitive projects that require public meetings, it is advisable that modelers
take advantage of the animation capabilities in micro simulation models to demonstrate
the potential change in traffic operations to decision-makers and members of the
general public.
Modelers must learn software capabilities and algorithms. Microscopic simulation
model development should be undertaken only by analysts with the required expertise
and theoretical appreciation of the software solutions. This means understanding how
the algorithms differ in terms of capabilities and limitations in the various model types,
and the importance of knowing the nuances in the algorithms used by different
software programs available for use on the same model type.
Perform back-casting exercises to validate modeling approach. Practitioners should
use the microscopic simulation models to estimate link speeds and volumes for prior
years and validate their accuracy against actual measured speeds and volumes. If the
estimates closely match the actual measures, then it would give credibility to the
modelers and their selected approach or methodology. FHWA Traffic Analysis Toolbox
Volume III and NCHRP Report 255 should be used in the absence of agency-specific
calibration or validation guidelines. A clear validation framework for microscopic
Best Practices in the Use of Micro Simulation Models
36
simulation models could be established and adopted. This will standardize the quality
and confidence in microscopic simulation results.
Use specialized utilities to convert network and flow data between model types.
Manual network coding should be avoided at all times. It is suggested that conversion
and editing utilities be used to convert the network from macroscopic or mesoscopic
scale models to the microscopic level. It is advisable that specialized utilities be
developed for situations not covered by existing conversion and network editing
utilities. Most importantly, these tools will be available when another microscopic
simulation project starts. It may be advantageous to use a mesoscopic model to prepare
the demands (or path assignments) for a microscopic model. Mesoscopic models will
limit the demand to the network capacity and provide a more reasonable starting point
for microscopic simulation than capacity restrained planning models. This helps to
minimize simulation problems caused by link demands that significantly exceed
reasonable capacity estimates.
5.2 Fundamental Considerations
Even though the peer exchange participants could not identify a specific set of rules that could be used
by others to select a model type, they were able to identify several fundamental questions that should
be considered in making a decision. The first question is relatively simple, but has a huge impact on
subsequent considerations:
Is the agency choosing a model type for a specific project or designing a modeling
capability for a region, agency, or series of projects?
If it is a project-level decision, the opportunities and constraints of the project dominate the decision
process. In this case the focus is on schedule and budget limitations, project purpose and scope,
available data and tools, staff resources and expertise, and the need for visual aids to communicate the
results to decision-makers and the general public. The decision may also be constrained by regulatory
requirements, political sensitivities, and attributes of the study area that require special attention. In
other words, the decision is made based on immediate project needs rather than long-range goals.
If the agency is making a strategic modeling decision, the issues and choices should be viewed in a very
different way. The implementation should be incremental and staged over multiple years to garner
support and build credibility given limited funding for long-range programs. Relatively small
demonstrations of increased capabilities on an annual basis are likely to be far more successful in
achieving the ultimate goals than large, complex development programs that require many years to
generate anything useful. This also provides time to develop the data resources and staff expertise
required to support the program.
The peer exchange participants from regional planning organizations viewed the ultimate goal as a suite
of integrated modeling tools that support regulatory requirements and studies representing a broad
spectrum of spatial and temporal fidelity. This suite should include macroscopic, mesoscopic, and
microscopic components that share common data and work together to support the analytical needs
within the region. The design also should consider interfaces to land-use models and air quality models.
Best Practices in the Use of Micro Simulation Models
37
Funding such an effort was a major concern. Because funding is primarily attached to individual projects
or studies, methods of tapping into these funds to develop capabilities for the common good were
discussed. One strategy was to charge a data services fee to each project that requests network, traffic,
and travel demand data from the regional repository. The region might also require each project to send
back its enhanced networks, traffic counts, and time-dependent trips to the regional agency for
inclusion in the data repository. Alternatively, the region could refund a portion of the initial data
service fee to those projects that return enhanced data.
5.3 Recommendations
The peer exchange participants recognized that decisions regarding the selection of a modeling tool are
very complex and often agency-specific. General guidelines and rules of thumb could be developed, but
developing a formalized expert system would be difficult and convoluted at best. A more helpful
approach might be to organize a group of experts that could advise agencies on a case-by-case basis
about the best way to proceed. The groups might be organized based on major categories of studies
such as traffic operations analysis, corridor planning studies, or regional strategic plans. The TRB and the
FHWA have begun an effort to develop a knowledge and information sharing website that may help
provide some of this advice and guidance through chat rooms, webinars, and state-of-the-practice
summaries.
The other major recommendation of the peer exchange meeting was to encourage states and MPOs to
begin the process of migrating from one-off project models to integrated multi-scale modeling
capabilities within a region. From a technical or theoretical perspective the multi-scale approach
recognizes the fact that modeling small areas with microscopic detail fails to consider the likely impact
of performance changes in that small area on the larger corridor or regional travel behavior. The multi-
scale approach was viewed as a practical and computationally efficient way to consider the full range of
issues, traveler responses, and performance measures that influence investment decisions.
The group recognized that the technical challenges of developing an integrated multi-scale modeling
capability are significant, but they did not view the challenges as insurmountable. Solutions to the
administrative and organizational issues were more difficult to identify. The fact that different
government agencies have primary responsibilities for different types of projects, and funding and staff
resources are spread thin among a large number of departments, programs, and projects, makes the
prospect of a coordinated, consolidation initiative difficult to visualize, let alone implement.
Best Practices in the Use of Micro Simulation Models
38
6 CASE STUDIES
Often the decision to use a traditional planning model versus a simulation tool is made based on the
unique circumstances of the project, the objective of the study, or the nature of the proposed measures.
This section summarizes 15 case studies where the decision was made to use a micro simulation model.
It highlights some of the reasons for that choice and some of the benefits and challenges that resulted.
Additional details about each case study are available in Appendix A.
Caltrans I-5 TDM/TSM Traffic Study. Caltrans used the VISSIM micro simulation
software to evaluate various demand management efforts, capacity improvements, and
enhancements to operations on the I-5 corridor in the current year and for a long-range
planning horizon. Specifically, it employed a signal optimization tool (Synchro) and a
microscopic simulation model (VISSIM) to conduct sensitivity analysis on roadway
projects designed to improve travel conditions. One of the questions was whether the
investments made by the state legislature produced the intended result. In addition, the
agency used the software to perform cost-benefit analysis of projects within the
corridor and determine the impacts of Transportation System Management
(TSM)/Transportation Demand Management (TDM) programs. Caltrans reported
difficulties with the micro simulation approach related to modeling traveler behavior,
sufficiently calibrating and validating the model, and even collecting the right form of
data. These challenges made it difficult to replicate the observed bottlenecks. Despite
the technical difficulties, the agency believed the simulation had answered the intended
questions for the study. Through the exercise, Caltrans acquired enough experience
using the software to be comfortable pursuing the next study without outside help and
thereby reduce project costs.
City of Moreno Valley, CA, TRANSIMS Implementation Study. The City used the
TRANSIMS microscopic simulation model to analyze the impact on the network,
especially with respect to commuting patterns and truck traffic, from converting the
zoning of more than 4,700 acres from residential and light industrial classifications to
warehousing and distribution centers. The analysis identified links and intersections that
would need to be improved to accommodate the proposed zoning changes, determined
which arterial and streets would need their “Traffic Index” raised due to increased truck
traffic, and helped decide whether the mass zoning change would alter commuting
patterns to the extent that it required major geometric changes to planned interchange
improvements. In addition, the study illustrated how truck routes would align
themselves to the existing and proposed locations of industrial zones and whether new
route alignments would detrimentally impact traffic in existing and future non-industrial
areas along their path. Difficulties encountered by the City included accurately modeling
traveler behavior, achieving convergence/stabilization, achieving a reasonable
calibration/validation, and allocating the computing time required to gauge adjustments
to input data. Nonetheless, the City reported that results were satisfactory, anticipated
benefits had been reported, and better capital investment decisions were made as a
result of the exercise. In the end, the City concluded that the simulation was warranted
Best Practices in the Use of Micro Simulation Models
39
and resulted in data that would not have otherwise been available, but was not cost-
effective.
New York City Department of Transportation (NYCDOT) Broadway Boulevard Project
in Midtown Manhattan Study. The Department used the AIMSUN simulation software
to study the effect of removing a major arterial on mobility and the safety of motorists,
bicyclists, and pedestrians. In addition, it aimed to determine the impact on air quality
of a large-scale diversion of traffic. The simulation focused on 4 to 5 hours during the
PM peak period. Difficulties arose with modeling a large urban network, accurately
modeling traveler behavior, and achieving reasonable calibration/validation. The model
started with a network from a previous study, but many real-life signal timing plans and
turn movement restrictions needed to be added to increase realism. NYCDOT found that
the exercise answered the study’s questions and led to a pilot version of the project
being implemented. It also helped in refining the proposed improvement measures.
Although the simulation helped the City answer these questions, the effort required a
large amount of resources and the project was deemed to be not cost-effective.
Caltrans I-5 North Coast Traffic Study. Caltrans combined a signal optimization tool
(Synchro) and a meso/microscopic simulation model (TransModeler) to study multiple
aspects of peak period traffic. Simulation was chosen because traditional travel demand
models did not offer a sufficient level of sensitivity to the variety of scenarios. The
objective was to understand how each project alternative would impact traffic
operations and travel behavior, provide standardized performance measures to judge
each alternative objectively, and define the mobility benefits from each proposed
construction phase. The study was also designed to develop visual animations to show
the results of building project alternatives for various audiences, and to use the
simulation work as a management tool for corridor operations. Difficulties mentioned
included the ability to model a large urban area network, accurately modeling traveler
behavior, achieving convergence or stabilization, and achieving reasonable
calibration/validation. However, Caltrans deemed the results satisfactory. The results
improved the capital investment decisions and decision-makers found visuals helpful.
However, the Department also expressed concern that the next simulation would be as
expensive and that it still did not have the time or qualified staff to do the exercise
completely in-house.
Sacramento Area Council of Governments (SACOG) Integration of Activity-Based
Model with TRANSIMS. The Sacramento MPO studied the benefits and challenges of
integrating its activity-based demand model (DaySim) with the TRANSIMS time-
dependent router. The objective was to more accurately model capacity improvements
at the regional level and provide better travel time information to the activity-based
demand model. The TRANSIMS router generated zone-to-zone travel times for every 15
minutes of the day. This data was then used to generate travel tours between activity
parcels. A user equilibrium process was used within the TRANSIMS environment to
assign the trips to the network. Developing the model interface was the primary
Best Practices in the Use of Micro Simulation Models
40
challenge. File size, computer processing time, and appropriate convergence algorithms
were major concerns. Additional tasks have been identified to address these concerns.
Caltrans I-80 Integrated Corridor Management (ICM) Study. Caltrans modeled the
impacts of ramp metering, active traffic management, and variable speed limits on
traffic flow and to determine what level of development would help create optimum
traffic flow on one of California’s most heavily used corridors. The study combined the
application of a signal optimization tool (Synchro) and a microscopic simulation model
(Paramics [Quadstone]) to evaluate demand management scenarios and operational
improvements. Modeling traveler behavior and achieving reasonable calibration and
validation were reported as the primary difficulties encountered. Caltrans was satisfied
that the results answered the study’s intended questions adequately. However, Caltrans
believed the simulation was not cost-effective and the agency anticipates the next
exercise would be even more expensive. The tool’s advanced visualization techniques
did not improve the decision-making process over what could have been illustrated
from a traditional planning model. As often is the case, good quality data are the key to
the results. Despite the limitations, the agency plans to continue simulation work, but
will likely consider a different simulation tool for the next study.
Caltrans I-580 Corridor System Management Plan (CSMP) Study. I-580 serves as the
major east-to-west transportation corridor between the Bay Area and the Central Valley
to the east. It experiences the highest levels of congestion-related delays in Northern
California. Caltrans used a microscopic simulation model (VISSIM/Paramics [Quadstone])
to identify current and future freeway bottlenecks, develop strategies for mitigating
bottlenecks, prioritize the mitigation strategies, and evaluate the impact of the
mitigation strategies on the performance of the freeway and adjacent arterials. The
model featured truck traffic, results by time of day, and traffic flow calculations related
to signals, ramps, and link saturation. The agency was satisfied with the simulation
results, and believed that it had led to better capital investment decisions and that the
results gave credibility to answers that were generated using other means. Nonetheless,
Caltrans believed that micro simulation was an expensive way to conduct what was
essentially a planning study and that the results highlighted the dichotomy between
micro simulation development costs and its usefulness in planning studies. For example,
the length of simulation periods (i.e., a 4-hour AM peak period and a 5-hour PM peak
period) and the length of the study area (30 miles) precluded meaningful visualization of
the results. Caltrans expressed concern that future projects would overrun schedules
and budgets if tight deadlines were not enforced. Moreover, the data in the model used
for this exercise are not being maintained due to the lack of a successor project and
management’s decision not to approve staff-hours for model maintenance.
Minnesota Department of Transportation (MnDOT) I-94 Managed Lane Operations
Study. MnDOT is conducting a high-priority project to study the feasibility of introducing
a Priced Managed Lane to the I-94 freeway. This is part of a region-wide effort to
evaluate design features and mobility options in Minneapolis after the collapse of the
Best Practices in the Use of Micro Simulation Models
41
I-35W Bridge in August 2007. The Department applied a mesoscopic simulation model
(CORSIM) to determine whether a managed lane would degrade conditions on the
general purpose lanes. Specifically, the software was used to model the likely impact on
network conditions resulting from the implementation of initiatives to increase vehicle
capacity; improve mobility for transit, pedestrians, and bicyclists; and improve
operations. In addition to making a decision about placing a managed lane on the
freeway, the purpose of the study was to evaluate the performance of the four concepts
that were developed. Overall, the client was satisfied with the simulation results
because it answered the study questions, provided visualization features that were
useful for presenting results to decision-makers, and led to changing or reversing an
existing design or operational policy. In particular, the simulation exercise demonstrated
that direct access ramps were needed to make the design function as expected.
Maricopa Association of Governments (MAG) I-10 Integrated Corridor Management
(ICM) Study. The MPO used mesoscopic simulation software (DYNASMART-P/DynusT) to
perform an elaborate study of a proposed major widening of I-10 in Phoenix, AZ. The
model was used to test six interrelated and successive policy questions. The first was to
determine the impact on traffic (e.g., route choice, travel time) that would result from a
freeway widening project, and then to test the effectiveness of various operational
strategies to reduce the congestion caused by construction. In general, planners were to
evaluate the realism of the software’s dynamic traffic assignment procedure in terms of
route choice, travel time, departure time, and the practicality of DTA for modeling
transit. Finally, tests were made to determine how the resulting changes to route choice
and mode choice could be fed back into the long-range planning model to estimate the
demand impacts. The primary difficulty was the need to upgrade computer systems in
order to have enough memory to run the model. MAG was inclined to do additional
simulation studies in the future and believed it had the necessary experience to perform
future studies without outside consultants.
Colorado Department of Transportation (CDOT) Downtown Denver Multimodal Access
Plan (DMAP) Study. The study modeled the impact of capacity and operational
improvements aimed at reducing congestion from excessive vehicle merging between
I-25 and SR 56 in Denver, CO. In addition, the improvements had to minimize the
potential for future regional traffic to cut through local streets to connect between the
two major roadway channels. Mesoscopic and microscopic simulation (VISUM/VISSIM)
software was used to specifically quantify the impact that direct freeway connectors
would have on peak hour traffic operations, which was evaluated according to HCM
consistent performance measures. The model was also used to determine the change in
“cut-through” traffic on local streets between I-25 and SR 56 resulting from each
improvement scenario. It also calculated network-wide performance measures and the
percentage of peak hour demand that would benefit from each alternative. There were
no problems reported with the micro simulation development effort or the results.
Best Practices in the Use of Micro Simulation Models
42
Utah Department of Transportation (UDOT) NEPA Study. The Department turned to a
microscopic simulation model (VISSIM) to quantify the amount of environmental
pollutants in the region induced by traffic and to model the likely results of mitigation
efforts as part of a National Environmental Policy Act (NEPA) analysis. The study
evaluated scenarios involving demand management policies, capacity improvements,
measures to improve alternative travel modes (transit, bicycle, and pedestrian), and the
application of operational improvements. The agency reported that traditional modeling
tools lacked sufficient robustness or sensitivity to have been used to answer these
questions. The major hindrances encountered from micro simulation were obtaining
HCM-consistent results, modeling a large urban network, depicting the traveler’s
behavior, and reaching convergence or stabilization with the model.
Arkansas State Highway and Transportation Department (AHTD) Hwy 112/Garland
Avenue Study. Garland Avenue (Hwy 112) in Fayetteville, AR, is a major congested
thoroughfare providing access to the University of Arkansas. The Department used a
signal optimization tool (Synchro) and a microscopic simulation model (VISSIM) to
evaluate several cross-section alternatives for a project to improve and widen the
highway. Due to the project’s location, the Department exercised special care to
measure likely interactions between motor vehicles and transit buses, bicyclists, and
pedestrians. The Department selected an advanced modeling tool (VISSIM) in order to
have the right level of sensitivity to the detailed analysis requirements. The Department
had difficulty generating HCM-consistent results, modeling traveler behavior, and
calibrating and validating the model to a reasonable level. Nonetheless, it believed that
the exercise had provided satisfactory answers to the intended questions of the study,
led to making better decisions regarding capital investment, and changed an existing
design/policy for the better. Visuals were also helpful for explaining results to policy-
makers. Most importantly, the simulation results highlighted the shortcomings of
performing similar studies using traditional planning models. The Department plans to
use simulation tools in future studies.
Kansas Department of Transportation (KDOT) Johnson County Gateway (I-435/I-35/K-
10 Interchanges) Study. KDOT is redesigning a highly congested interchange that
connects two interstates and a limited access state route in Kansas City, KS. The design
is extremely complicated from an operations standpoint. The purpose of the study is to
model the present and future operations given various design proposals in order to
select the option that will eliminate the bottleneck in the long run. The Department had
difficulty modeling a large urban network, accurately representing traveler behavior,
and achieving a realistic calibration and validation. It was, however, satisfied with the
application results given the policy choices presented under the various scenarios.
Overall, the application led to better capital investment decisions and policy-makers
found the visual aids helpful for understanding the results. Despite concluding that the
next simulation-based study would be as expensive as this one, the Department found
the simulation modeling to be cost-effective and was in favor of using simulation in
additional studies.
Best Practices in the Use of Micro Simulation Models
43
San Diego Association of Governments (SANDAG) I-15 Integrated Corridor
Management (ICM) Study. I-15 is the primary artery that commuters and commercial
vehicles use to travel from northern San Diego County to downtown San Diego. It has
heavy congestion that needs to monitored and managed by various state and local
transportation agencies on a daily basis. The MPO leveraged simulation modeling
capabilities to estimate detailed traffic flow for making decisions in its traffic
management center. The objective was to find approaches that would maximize person-
throughput in the face of recurrent and non-recurrent congestion. The agency intended
to use the application to prepare messages for the variable message signs on the
corridor that advise drivers of an accident’s occurrence and to provide temporal and
modal options (route-paths) for avoiding the resulting congestion. The agency used a
simulation model capable of providing analysis at the macro, meso, and microscopic
levels (TransModeler). The modeling scenarios contained proposed demand
management initiatives, improvement measures for transit and pedestrian movement,
and modifications to advance vehicle-centered operations. The agency believed that the
animation tool was more effective in helping to communicate the results than
comparable maps, tables, and reports.
San Francisco County Transportation Authority (SFCTA) Doyle Drive Project. The SFCTA
faces the challenge of redesigning a 60-year-old thoroughfare that provides direct
access to the Golden Gate Bridge. The current design is inadequate to handle the
exponential growth in volume on the facility. The purpose of the study was to provide a
short-range forecast of route diversion options during the reconstruction phase and
prepare the authority for the potential bottlenecks and queues that could result. To that
end, it applied a mesoscopic level DTA tool with simulation capability (DynamEQ) to
evaluate possible solutions to these issues. The Authority encountered difficulty with
the tight project deadline versus the amount of time required to perform the necessary
analysis. Nevertheless, the exercise led to more effective or accurate answers to the
main policy dilemmas surrounding the construction project. Visuals were deemed useful
to decision-makers. The agency believed that similar studies could be less expensive, but
that it had been cost-effective thanks to very tight deadlines and the significant
involvement by the software developer. The authority is inclined to use micro
simulation tools for future studies.
Table 3 provides a quick summary of the 15 case study projects. More details are provided in
Appendix A.
Best Practices in the Use of Micro Simulation Models
44
Table 3: Summary of 15 Case Study Projects
Project Model Type Software Scale Budget Duration Cost-Effective1
Caltrans I-5 TDM/TSM Micro VISSIM Corridor > $1M > 18 Months Yes
Moreno Valley TRANSIMS Implementation
Macro/Micro TRANSIMS Regional, Subarea $250k - $500k > 12 Months No
NYCDOT Midtown Manhattan Micro AIMSUN Subarea $500k - $1M 12 Months No
Caltrans I-5 North Coast Micro TransModeler Corridor $500k - $1M N/A Yes
SACOG ABM-TRANSIMS Micro TRANSIMS Regional < $250k > 12 Months No
Caltrans I-80 ICM Micro Paramics, SimTraffic,
Synchro Corridor, Subarea > $1M > 18 Months No
Caltrans I-580 CSMP Micro Paramics, SimTraffic,
Synchro Corridor, Subarea $250k - $500k 11 Months No
MnDOT I-94 Micro CORSIM Corridor $250k - $500k 5 Months Yes
MAG I-10 ICM Meso DYNASMART-P/DynusT Regional, Corridor < $250k N/A N/A
CDOT DMAP Micro VISSIM Subarea < $250k N/A Yes
UDOT NEPA Analysis Micro VISSIM N/A < $250k > 18 Months Yes
AHTD Hwy 112/Garland Ave Micro Synchro, VISSIM Corridor < $250k > 12 Months Yes
KDOT Johnson County Gateway Micro VISSIM Corridor > $1M > 12 Months Yes
SANDAG I-15 ICM Micro TransModeler Corridor $500k - $1M > 12 Months No
SFCTA Doyle Drive Micro DynamEQ Subarea < $250k 2 Months Yes
1 Cost-effectiveness is based on the subjective assessment by the survey participant.
Best Practices in the Use of Micro Simulation Models
45
7 SYNTHESIS OF FINDINGS AND LESSONS LEARNED
This section uses the lessons learned from the web survey, peer review exchange, and case studies to
address the research questions posed by the AASHTO Standing Committee on Planning:
1. Under what circumstances have micro simulation modeling efforts been found to be warranted and cost-effective?
2. What are the benefits of microscopic simulation compared to the currently available macroscopic models?
3. In what ways have investing and conducting micro simulation led to better or more effective designs and investment decisions?
4. How have the data needed for micro simulation and dynamic traffic assignment models been developed?
5. What steps are being taken to attract and retain staff with the necessary expertise and knowledge to perform and oversee micro simulation and dynamic traffic assignment modeling efforts?
6. What organizational or institutional arrangements have been useful in addressing any or all of the issues identified above?
The information gathered from these sources is synthesized and summarized below.
7.1 Under what circumstances have micro simulation modeling efforts been found to be warranted and cost-effective?
There are many operational questions that cannot be answered through the use of a traditional demand
model. The level of detail of stakeholder questions and the information required to answer these
questions is increasingly beyond the capabilities of traditional models. Instances where microscopic
simulation models have been found to be warranted are reasonably well documented in the literature,
and the results of the web survey suggest that most respondents are well aware of the limitations of
traditional models. As a general rule, dynamic (mesoscopic and microscopic) models have been shown
to be warranted in applications that are likely to induce a temporal or spatial pattern shift of traffic
among different roadway facilities at a corridor or network-wide level.
Microscopic simulation models are nearly always warranted where the detailed interactions of vehicle
movements are required to replicate real-world conditions. Prime examples include traffic signal
systems and freeway weaving and merging sections. Microscopic simulation is also important for
measuring and evaluating the interactions of vehicles with pedestrians and bicyclists, time-of-day
parking restrictions and lane closures, and priority vehicle treatments such as high-occupancy vehicle or
bus lanes.
In addition, microscopic simulation is beneficial in analyzing and understanding vehicular movements in
congested, urban conditions, particularly if there are closely spaced intersections and access points. For
example, the micro simulation analysis applied to downtown Salt Lake City had a major influence on the
design because it predicted queuing on median ramps and closely spaced intersections that the
Best Practices in the Use of Micro Simulation Models
46
deterministic models did not identify. In other words, micro simulation is important in situations where
traffic queues interact with each other and cascade into gridlock or stop-and-go traffic conditions.
In situations where the projected demand is greater than network capacity, micro simulation analysis
will predict a very different result from static assignment models. Static assignments permit traffic
volume to exceed capacity whereas microscopic simulation models constrain the flow rate at bottleneck
locations. The microscopic model will show more congestion upstream of the bottleneck and less
congestion downstream of the bottleneck than a traditional model.
Microscopic simulation can also help to identify problems during moderate traffic conditions. In these
situations the volume-to-capacity ratios are not high enough for a macroscopic model to predict any
noticeable delay, but the microscopic simulation can identify delays at off-ramps and propagating
queues from upstream intersections that warrant attention.
Another example of a suitable application for micro simulation is the analysis of the effects of truck lane
restrictions on a freeway. Macroscopic and mesoscopic simulations are useless in this situation because
they do not model individual lanes on a given link. Micro simulation, on the other hand, is able to show
the full effects of truck restrictions on traffic flow. They can also quantify the impacts of transit bus
routes on traffic flow. Microscopic simulations can estimate the impact of buses on cars and vice versa.
Similarly, design considerations related to lane changing can be modeled only with micro simulation. A
prime example is the location of merge and diverge areas for left-side HOV lanes. The proximity of these
access areas to the on- and off-ramps of the interchanges are integrally related to the level of traffic
congestion and the driver lane changing behavior. If the distances between the ramps and the access
areas are not adequate, aggressive drivers will significantly increase lane weaving, which produces a
ripple effect upstream that deteriorates traffic flow.
If the concept of “warranted” is expanded to include successfully communicating the results of the
analysis to decision-makers and the general public, several survey participants readily claim that
microscopic simulation was warranted for its visualization capabilities alone. Visual animation of the
traffic conditions can help to develop credibility for the modeling process and demonstrate the benefits
of the proposed solution in ways simple numbers cannot.
Despite all of these advantages, very few of the study participants believed their micro simulation
studies were cost-effective. They almost always agreed that the tool answered the study questions and
influenced the design decisions, and they plan to do similar studies in the future because they see no
viable analysis alternative. But from the perspective of cost-effectiveness, they found the time and cost
of developing and applying microscopic simulation models to be excessive.
Microscopic simulation models require significantly more data and staff resources than macroscopic or
mesoscopic simulation models, but benefit planning by providing more accuracy and fidelity to actual
traffic conditions on the roadway network. Therefore, selecting a model type will always be a trade-off
between the desired level of granularity and confidence in the recommendations and the available
resources to produce these more costly simulations.
Best Practices in the Use of Micro Simulation Models
47
Time and budget limitations appear to be the most significant factor in choosing a modeling approach.
Tighter deadlines and budgets have forced some modelers to choose lower fidelity models to alleviate
the risk of cost overruns and to finish on time because they are in a production setting where they have
decision deadlines to meet.
In most cases, the project schedule and budget is set by years of planning studies using traditional
models. This becomes the de facto benchmark by which cost-effectiveness is defined for microscopic
simulation projects. Modelers, managers, and decision-makers have not adjusted their expectations to
account for the cost and complexity of advanced modeling techniques. Because micro simulation is a
relatively new tool for transportation planning, there may be opportunities to gain efficiencies and
increase productivity in the years to come. Plans to develop data resources and reusable networks will
likely pay dividends in the future. In the near term, however, the survey respondents did not see many
opportunities to save costs on their next project.
The benefit-to-cost tradeoff should command attention, but it is also important to place the cost of
developing a model, be it macroscopic, mesoscopic, or microscopic, in the context of the project’s
overall price tag. Modeling is typically a fraction of the design costs, and even smaller compared to the
construction costs, associated with any selected alternative. Using the results from the wrong modeling
approach, regardless of its cost-effectiveness in the short run, could prove to be more costly in the long
run if the wrong solution is selected. Many of the survey participants that considered their simulation
efforts to be cost-effective discovered through the modeling process a less-costly solution than they
originally anticipated.
7.2 What are the benefits of microscopic simulation models compared to the currently available macroscopic models?
Macroscopic models were originally designed to estimate demand for major highway improvements
included in regional long-range plans. In most cases they are calibrated based on daily volumes and
counts. Travel times based on volume–capacity relationships are used in the assignment process to
distribute the travel demand to a logical set of paths. In this process, volumes can easily exceed
capacities and loaded speeds have little relationship to reality. The purpose of these models is not so
much to predict how a facility will actually operate, but to estimate the demand for a given facility so
that the design might accommodate the demand.
Microscopic (and mesoscopic) simulation models are focused on the actual operations of the facilities
and the impacts of true capacity constraints on upstream and downstream system performance over
time. The systematic comparison of static traffic assignments to a DTA model (VISTA) performed by
Boyles in 2005 demonstrated that traditional models significantly underestimate network congestion
levels. If the purpose of the study is to address congestion problems with operational and management
strategies, mesoscopic and microscopic models are much better positioned than macroscopic models to
evaluate the effectiveness of the alternatives.
In general, the key advantages of micro simulation tools can be traced to their ability to:
Model different vehicle types and intersection controls
Consider temporal and spatial interactions
Best Practices in the Use of Micro Simulation Models
48
Consider traffic dynamics, queuing phenomenon, and spill-backs
Model different behavioral assumptions and user classes
Visualize how the proposal alternative will operate
Despite these advantages and capabilities, there is still significant debate among transportation
professionals and decision-makers about the benefits of micro simulation modeling. While microscopic
simulation tools have been used successfully to ease the overall decision-making process and improve
facility designs, some decision-makers are convinced that micro simulation results are not reliable
enough for agencies to stake decisions on major capital investments. In addition, many have yet to see
effective returns from the investment in micro simulation due to the amount of effort required to code,
calibrate, and apply the simulation model.
7.3 In what ways have investing and conducting micro simulation led to better or more effective designs and investment decisions?
Survey respondents reported numerous situations where micro simulation led to better investment
decisions and more effective designs. There are likely to be situations where simulation models provided
faulty predictions, but these projects were not included in the web survey. It is probably safe to assume,
however, that a properly calibrated and validated microscopic simulation model will more often than
not lead to more effective designs and investment decisions because it can more closely replicate what
is likely to occur in the real world.
Suggesting that a design is better or more effective is either a subjective opinion or it requires some
basis of comparison. In most of the cases reported in the web survey, the assessment was based on a
comparison to prior studies using traditional models or HCM calculations. For example, an intersection
designed as an all-way stop using traditional traffic engineering calculations did not perform as expected
in the real world. A microscopic simulation was then used to confirm the observed behavior and develop
a more effective design for the intersection. The evaluation led to the decision that traffic would be
better served if the intersection was configured as a round-about.
Micro simulation modeling has also proved useful in situations that are outside the bounds of traditional
techniques. These can include odd or complex intersection configurations or heavily congested arterials.
For example, a heavily congested arterial with two cross streets 120 feet apart could not be designed
using traditional stop sign or signalization calculations. A round-about option was modeled using micro
simulation with the software providing guidance on the appropriate diameter for the round-about,
whether it required a single or dual lane circle, and how queuing on minor approaches would be
eliminated. In addition, the model was able to show that driveways for existing businesses around the
proposed round-about would be too close to the traffic circle. The tool provided the data needed to
relocate the driveways a safe distance from the round-about.
Several of the studies reported as part of the web survey highlighted the benefits of micro simulation for
improved decision-making. A number of these observations are repeated below:
Simulation model results led to changing or reversing an existing policy or design
A DOT wanted to add tolls to a proposed bypass around a city. A consultant performed a toll revenue analysis using a static travel demand forecasting model. A subsequent study
Best Practices in the Use of Micro Simulation Models
49
using a DTA model showed significantly less volume and revenue on the proposed facility. As a result the DOT decided not to pursue the project.
A DTA model was used to analyze a heavily congested intersection that led to recommending that the existing round-about be replaced by a signal.
The DOT had a preferred alternative from a traditional planning study, but simulation showed better results (metrics) from a different alternative, which ended up being adopted.
Simulation results led to better capital investments
The micro simulation projected higher traffic flows than those predicted by a travel demand model, and as such justified the HOT lane investment.
Classical weaving analysis predicted that a key weaving section would not work, but micro simulation predicted it would. The additional detail of the micro simulation allowed a very expensive design to be deferred indefinitely, thereby putting capital funds to better use.
Simulation results led to more effective answers
A local agency needed to find a way to improve traffic conditions quickly and cost-efficiently. The corridor was already synchronized with a 100-second cycle length. Although a longer cycle length was needed to prevent cycle failures, the agency was reluctant to implement the improvements because of concern about access to a regional mall. Synchro/SimTraffic was used to demonstrate that a 120-second cycle length would not adversely impact queuing on the mall’s access roadway or anywhere else. But more importantly, the longer cycle length had an immense positive benefit on the main street’s operation. The scope of the project was changed from presenting alternatives in a public meeting to signal timing implementation. The new timing plans had the predicted effect, and the team was able to take credit for quickly, dramatically, and cost-effectively improving traffic on an important, highly visible facility without adversely impacting cross streets.
Simulation results highlighted a particular inadequacy in similar studies conducted using traditional planning practices
When converting from a static model to a micro model, entry flow rates needed to be adjusted to reflect realistic trip distributions by time of day. The static model overestimated demand at the start and end of simulation and underestimated demand during the peak period. A multi-resolution modeling process was employed to evaluate the peak-hour conditions. The travel demand (macro) model was first converted to a DTA (meso) model. The DTA model generated the time-dependent equilibrium conditions, which in turn were fed into a microscopic simulation model to accurately assess the actual dynamics of the system.
7.4 How have the data needed for micro simulation and time-dependent models been developed?
Microscopic and dynamic models require substantially more data beyond what are commonly needed
for traditional models. Savings in the time or cost of obtaining these data will directly translate into
budget savings that can be invested in other ways. Developing methods for converting or synthesizing
Best Practices in the Use of Micro Simulation Models
50
data from traditional models and other existing data sources is an effective way of reducing data needs
and costs. Data conversion tools can significantly reduce network coding times, and sophisticated
editing tools can automatically add many of the details. It is also beneficial to have staff with a strong
programming background to create and automate some data entry and error-checking processes.
Multi-resolution modeling (macro, meso, and micro) is the future of current modeling practice. For this approach to be successful, the integration process needs to transfer data seamlessly between software platforms. The data required for all levels of modeling would be stored in a common data repository for extraction and processing by a given model. The results would be returned to the repository for use by the next model in the overall process.
For standalone microscopic simulation models, data for networks, travel demand, and model calibration
and validation need to be collected as part of the project or converted from existing sources. The key
inputs include detailed traffic counts and speed data for relatively small time periods (e.g., 15 minutes).
Data collection techniques that measure volume and speed simultaneously are important for calibrating
the simulation parameters. These data are also useful for O-D estimation procedures that increase or
decrease demand between each origin and destination pair based on observed counts and travel
speeds. Loop detectors or counters that record vehicle occupancy by lane at 30-second intervals are also
useful for model development.
Calibrating and validating the simulation result against field data is critical for demonstrating the level of
accuracy of the model and the reliability of the forecasts. In practice, validation is done mostly to ensure
that base-year conditions were simulated closely. Post-construction conditions are rarely investigated or
confirmed. Back-casting applications, however, are frequently performed. The major drawback to back-
casting is that there is rarely sufficient historical data to adequately validate the model estimates.
7.5 What steps are being taken to attract and retain staff with the necessary expertise and knowledge to perform and oversee micro simulation and dynamic traffic assignment modeling efforts?
As microscopic simulation models increasingly bridge the gap between planning and operations,
developing these models will require skills that are normally split between two distinct disciplines:
planners and traffic engineers. Simulation modelers with skill in both planning and traffic operations are
highly desirable. Several of the survey respondents, however, were of the opinion that only a handful of
agencies and institutions perform “leading edge” transportation modeling, and therefore attracting and
retaining qualified staff is not an issue for most agencies at this point in time. The few agencies that
conduct a significant amount of state-of-the-art modeling do so for research purposes, generally in
conjunction with university research centers where a pool of students can do the agency’s micro
simulation modeling work as part of their thesis and dissertation research at a lower cost than hiring
additional staff.
The majority of public agencies that participated in the survey believe that they can offer significantly
better benefits and reasonable salaries compared to the private sector, and as such they believe they
can hire new talent or hold onto their talented staff. However, they plan to continue having relatively
small simulation modeling teams and anticipate continuing to leverage consultant services to complete
a large portion of their modeling tasks. That being said, public agencies often try to stay up to date by
Best Practices in the Use of Micro Simulation Models
51
reading papers, attending conferences, and having conversations with researchers and software
developers. The survey results indicate that agencies strive to provide training and skills development
for their staff, but funding remains a major obstacle.
The participating agencies were not concerned about competing for talent with the private sector
because of the different work environment each sector offers to the prospective employee. At public
agencies retention is steady and the average tenure is measured in years, which is attractive to those for
whom job security is paramount. The predominant hiring philosophy is that individuals are the party
seeking the employment opportunity and not vice versa. In general, agencies cite greater work security,
more time off, and less overtime as reasons they do not have difficulties attracting talent. They are also
generally attractive places to work for modelers starting families who are willing to trade greater time
for familial obligations against the higher monetary compensation available in the private sector. The
economic recession seems to have made the benefits of public employment comparatively greater, and
as such it is hard to find any open engineering positions of significance at public agencies.
However, public agencies face several long-term structural challenges when it comes to hiring talented
staff. For example, there is a ceiling that limits how much the agency can offer in terms of salary and
benefits, which are generally non-negotiable. There is also the challenge of a lack of upward mobility
within public agencies. They are also at a disadvantage with the private sector for recruiting prospective
employees needing a quick hiring decision. Public agencies have the administrative overhead of
developing, approving, and posting job announcements and following set hiring procedures. Consulting
firms can often hire staff based on recommendations and referrals rather than a formal job posting.
7.6 What organizational or institutional arrangements have been useful in addressing any or all of the issues identified above?
The web survey, peer exchange, and follow-up interviews identified several organizational and
institutional arrangements that are useful in developing cost-effective microscopic simulation models.
Perhaps the most important concept is to have an appropriate public agency be the “home” for a
particular model. This helps to ensure continued investment in the model’s capabilities and advocacy
and support for its application. At the beginning of a project the model is “checked out” by the
consultant or public agency and at the conclusion of the study the model is “checked back in.” This gives
the model a “home base” with good version control and an error-tracking system for networks and
model parameters. It also helps to have progressive leadership that supports the reuse of micro
simulation data and sets the stage for strategic modeling initiatives.
Easy access to software support staff and technical manuals is also important. These services could be
provided by software vendors, but it is often better to have experienced software users providing this
assistance. Most microscopic simulation software is relatively new and constantly being improved. As a
result, the tools are generally not well documented. Modelers are likely to hit obstacles as they calibrate
or implement the model and need a resource to which they can turn for advice. If the agency does not
have much experience with the software, it may be desirable to include the software vendor as a paid
advisor on the project. This often results in the vendor enhancing or fixing software capabilities sooner
than if it were not involved. A win-win situation develops as the software vendor expands its user base
by promoting the new tool and the agency can use the tool for the project.
Best Practices in the Use of Micro Simulation Models
52
The use of automated data collection systems is an important way to reduce costs and improve
performance. The key is to ensure that the data collection and aggregation process provides information
that is useful to model development and operational planning. If at all possible, arrangements should be
established with the data collecting agency to summarize traffic volumes and speeds simultaneously for
relatively short time periods (e.g., 15 minutes). Quality control and data cleaning procedures should also
be included to minimize distortions in the results. These data provide an excellent source of longitudinal
and temporal information for model calibration and validation that is not available from traditional
modeling datasets.
Most existing automated data collection systems are located on key freeway sections within the region.
If there is any opportunity to expand this capability to include arterials, it should be pursued. Centralized
traffic signal management systems typically include a mechanism for monitoring traffic volumes and
speeds at signal detectors. These are often real-time systems that do not include data aggregation and
storage capabilities. If arrangements can be made to add data collection procedures to a traffic signal
system, the information would be extremely useful for simulation modeling.
Finally, coordinating with other agencies that routinely conduct microscopic simulation studies provides
less experienced agencies with the information to identify likely obstacles and how to address them in
each phase of model development. This can be achieved through user group meetings or peer exchange
sessions for strategic planning or a specific project. Moreover, coordinating with neighboring agencies
and MPOs offers the added benefit of correctly quantifying the flows between cities that are external to
the regional model.
Coordination should not, however, be limited to simulation modelers. The technical and planning staffs
from various stakeholders should be kept apprised during the model development process. Their input
will be useful in addressing specific questions and identifying or critiquing proposed alternatives.
Stakeholder input and feedback is also invaluable for evaluating the plausibility of the modeled results.
Being an experienced simulation modeler is not enough; understanding the theory behind traffic
operations is even more important. Providing training in traffic engineering principles and involving
traffic engineers in the modeling process are important ingredients for success.
Best Practices in the Use of Micro Simulation Models
53
8 BEST PRACTICES IN MICRO SIMULATION
Microscopic simulation models are significantly more complex and far more sensitive to minute details
about the transportation system and travel demand than traditional models. This is precisely why they
are better able to evaluate the subtle differences that result from demand management and operational
improvement strategies. This also means they are far more susceptible to what is commonly referred to
as amplification errors. Amplification errors are small errors or inaccuracies in the input data or model
parameters that are multiplied by travel demand to create large errors in the model results. To avoid or
minimize these errors, considerable care should be taken in preparing the network, travel demand
inputs, and model parameters. This section highlights some of the steps that are typically employed to
develop a reliable modeling process.
8.1 Data Quality and Consistency
Data quality and consistency are much more important for microscopic simulation models than they are
for traditional models. The modeler needs to know the pedigree of the data that are being used and
should never assume that the data are error-free or correct simply because they are provided by a
credible government agency or contractor. This is particularly true of data extracted from traditional
planning models. These models can include a significant number of inaccuracies that ultimately have no
noticeable impact on the macroscopic model results. Factors to consider include those discussed in the
following paragraphs.
Network Consistency. If the traditional planning model network is being converted to a
mesoscopic and/or a microscopic simulation network, it is desirable to ensure that the
lane configurations are consistent in all models. This means coding the correct number
of lanes, lane connectivity, and lane geometry at the macroscopic level before
converting the network. GIS files can considerably simplify this step because their data
on centerlines and other geometry is typically more accurate. Additional work may be
required to define all allowed and prohibited turning movements. Online maps and
aerial photographs can be invaluable sources of data at this stage, especially for
validating lane connections.
Zone Connectors. When moving from static to dynamic models, it is important to note
that the geometry and flow characteristics of zone connectors have physical significance
in dynamic models. The connectors should therefore be modeled as physical roadways.
This implies that zone connectors should not attach to major intersections, but rather
be moved to mid-block locations or distributed on the link based on trip origin and
destination locations.
Intra-Zonal Trips. Macroscopic models do not load intra-zonal trips to the network. As a
result, it is common practice in regional modeling to “dump” all of the intra-zonal trips
estimated by the trip distribution model into the auto trip table and skip all mode choice
calculations. When these trips are converted to a mesoscopic or microscopic model they
often are given intra-zonal origins and destinations in order to model traffic on a more
detailed network. The analyst needs to decide if all of the intra-zonal trips should be
Best Practices in the Use of Micro Simulation Models
54
converted to auto trips or if some of the trips should be deleted because they should
really be walk trips. This can be a major issue in downtown areas where the zone sizes
are typically quite small and parking costs are significant.
Data Consistency. The data used for model calibration and validation need to be
internally consistent and as closely correlated with the modeling scenario as possible.
For example, it is important that link speeds and volumes be measured simultaneously
because link speed is highly correlated with volume. When link speeds and volumes are
collected on different days or different time periods, it is extremely difficult to reconcile
differences in speed or volume predicted by the simulation model. In addition, modelers
need to know the sources of data used to limit any inconsistencies that could arise from
combining data from different years, seasons, days of the week, or times of day. Finally,
modelers should confirm that the data are consistent with modeled conditions. For
example, signal timing databases do not always correspond to the timing plans
implemented in the real world, or timing plans may have changed in the real world after
the traffic counts were collected. Understanding the full range of data issues is
important for properly calibrating a microscopic simulation model.
8.2 Network Coding and Error Checking
Microscopic network coding is a time consuming task. Geometric and control attributes are both
required. Geometric attributes include link lengths and widths, link geometry (shape points), number of
lanes, lane connectivity, pocket lanes, lane use and restrictions, and junction layouts. Control attributes
include control type (pre-timed/actuated signal, stop/yield signs), phasing, and timing plans. Common
practice is to use existing conversion utilities wherever possible to convert the existing regional travel
demand network or GIS street layers to a skeletal network (initial network) for construction of a
microscopic simulation model. Traffic control data will typically need to be obtained from a signal
database such as Synchro or from the traffic control department records.
Some conversion utilities work with software developed by the same vendor, such as TransCAD-
TransModeler and VISSIM-VISUM. Others work with different sets of software such as DynusT-VISSIM,
CUBE-TRANSIMS, and EMME/2-TRANSIMS. The latter are usually developed as part of previous studies
and have been later made available to the user base. Some conversion utilities are flexible in that they
allow the user to specify the network data in a given format using Excel or text files before conversion.
In that case, it may be advisable to have a utility that reads existing input files and outputs them in the
required format.
A few software vendors provide conversion tools to import control settings from traffic models such as
Synchro or TRANSYT-7F. Some conversion utilities also allow the user to enter default signal timing and
phasing plans for use when detailed signal data are not available or are too costly to collect. Some
microscopic simulation tools even allow for signal optimization and progression. Actual signal timing
information is important for model calibration and validation, but it is less important for model
application. Because the purpose of most microscopic simulation applications is to test alternatives that
change the network attributes or travel demand, it is helpful to code generic actuated signals in future
or alternative networks and use the simulation results to identify signal failure locations. These signals
Best Practices in the Use of Micro Simulation Models
55
can then be adjusted based on demand or delay data, or imported into a signal optimization model to
determine optimal timings.
Furthermore, some software packages provide conversion tools that allow a subarea network to be
extracted from the traditional planning model network, along with the corresponding demand matrix or
path assignments. If extraction is necessary, it is best practice to run the regional model to user
equilibrium conditions before a subarea is extracted. If the simulated alternatives are likely to have
regional impacts, the alternative can be coded into the regional model, the model can be re-run, and the
subarea trips can be extracted for detailed analysis by the simulation model.
Most microscopic simulation software is bundled with a graphical user interface (GUI) for editing
network and displaying model results. Using a GUI to perform error-checks on the network is highly
advisable. Test simulations with minor levels of demand are a useful place to start. This will quickly
expose major connectivity or coding errors. This can be followed with more demanding stress tests to
identify false bottlenecks, lane connectivity and restriction issues, and missing traffic controls.
8.3 Origin-Destination Demand Preparation
The most common practice for preparing time-dependent O-D data is to apply diurnal distributions to
trip tables extracted from regional travel demand models. Diurnal distributions may be derived from
household travel surveys or traffic counts. It is preferable to apply separate diurnal distributions by trip
purpose and vehicle type. If possible, the distributions also should differ by general O-D pairs. In
addition, the distribution of demand by facility type should be checked against traffic counts to ensure
the overall shape of the curve is reasonable.
Figure 7 depicts the importance of using accurate diurnal distribution data to convert static model
demand to the time-dependent demand required by microscopic simulation models.
Figure 7: Preparing Time-Dependent Demand
The figure also shows the importance of including a reasonable start buffer (warm-up) period before
collecting network performance statistics. It is also desirable to extend the simulation time beyond the
period of interest to enable all of the traffic loaded within the study period to reach their destination. As
a general rule, the start and end buffer periods should be as long as it takes for the longest trip to
traverse the network under the prevailing traffic conditions.
Best Practices in the Use of Micro Simulation Models
56
Time-dependent demand derived from regional planning models often includes travel patterns and
temporal distributions that differ from observed traffic counts. A detailed micro simulation will generally
adjust the demand data using O-D estimation techniques to match time-of-day traffic counts on specific
facilities in the study area. O-D estimation uses optimization techniques such as linear and nonlinear
programming to minimize the variances between simulated link volumes and observed link counts. The
time-dependent trip matrix from the regional planning model is adjusted until the distribution of traffic
by time of day replicates the traffic counts. These procedures are most effective if applied with
15-minute traffic counts and where at least 60 percent of the trips travel through links with traffic
counts.
The result of the O-D estimation process can be divided by or subtracted from the original time-
dependent demand matrix to create a set of correction factors that are applied to future year travel
demand data. The choice between multiplicative and fixed adjustments typically relates to the size of
the correction and the nature of the alternatives to be tested. If the multiplicative factors are huge, it is
better to use fixed adjustments. If the travel patterns are expected to change significantly or the future
forecast is much higher than today, multiplicative factors can be more responsive to the changing
dynamic. To temper these effects, some applications apply both methods and average the result.
8.4 Calibration and Validation
The web survey revealed that calibration and validation are the most challenging aspects of developing
a microscopic simulation model. They are also perhaps the most important. To give an illustration, one
respondent commented that:
“Micro simulation is a useful tool if calibrated and validated, and a garbage generator otherwise.”
Calibration and validation, however, cannot be done blindly or mechanically. They require a substantial
appreciation for the rationale behind each adjustment. This is illustrated by another comment:
“A calibrated and validated model is useless without understanding the software algorithms and its limitations.”
The calibration process is typically an iterative approach that requires experience and theoretical
prowess to adjust supply, demand, and behavior parameters based on realizations of existing solutions.
It is helpful if analysts first develop an understanding of the sensitivity of each model parameter before
attempting calibration. Software expertise and knowledge of existing field conditions is required to
adjust traffic control information, demand data, and network attributes to reflect operating
characteristics.
Calibrating the simulation model parameters based on local conditions and traveler behavior is a key
component of the model development process. Most software provides parameters to adjust vehicle
compositions, car following behavior, maximum acceleration/deceleration rates, minimum headway,
and weight and power distributions. The flexibility to set these parameters should not be mistaken as a
free pass to tweak the parameters indiscriminately. Parameter changes should be undertaken by
knowledgeable and experienced modelers. In situations where time constraints are tight or where the
budget is limited, it may be acceptable to use calibrated parameters from similar studies conducted with
Best Practices in the Use of Micro Simulation Models
57
the same software package. The analyst should not attempt to use default or calibrated parameter
values from other software packages because each package implements algorithms somewhat
differently.
The model should be validated against traffic counts, speeds, queue lengths, and other performance
measures at an hourly or, preferably, a 15-minute level. Statistical standards by volume levels, facility
types, screenlines, and other summary classifications should be set and achieved for the large majority
of network links. Visualization tools are helpful in comparing the results to observed behavior and
identifying the source of simulation problems. In addition to validating a specific point in time, it is good
practice to also validate the model at a different point in time when conditions are different. This is
typically called “back-casting,” and it helps build credibility for the model’s forecasts.
8.5 Modeling Difficulties
The web survey indicated that (1) accurately modeling traveler behavior, (2) achieving reasonable calibration and validation, and (3) achieving convergence or stabilization are the primary difficulties in developing microscopic simulation models. The follow-up interviews provided an illustration of each challenge:
Accurately modeling traveler behavior
“A difficulty has been achieving the right balance between freeway and surface-street flows. Because run times are so long for the regional model, we have not been able to keep fine-tuning the parameters and instead had to declare ‘good enough’ and move on. The solution for future studies is to manage run times by not modeling such a large region, providing more computing capacity, or both.”
Achieving reasonable calibration and validation
“Calibration and validation are never easy. In addition to balancing flows between facilities, another issue has been overly congested interchanges resulting from the router diverting more traffic to local streets than the micro simulator can handle. Once ramp queues impact the freeway mainline, the micro simulator is no longer capable of feeding back the delays correctly. This must be manually managed by implementing time penalties for movements that are to be discouraged. This represents an academic, as opposed to practical, limitation to the micro simulator stabilization process because the end result is still acceptable.”
Achieving convergence or stabilization
“The difficulty relates to the fact that the base network is congested, in some cases severely. The congestion impacts all aspects of validation but more so convergence and stabilization because re-routed drivers can oscillate between parallel paths. Orderly convergence was attained by damping the system’s response to changes between iterations. “
Best Practices in the Use of Micro Simulation Models
58
9 MODEL SELECTION GUIDELINES
The goal of the peer exchange meeting was to design a decision-making process for selecting the best
type of model for different types of applications. The group ultimately concluded that the process was
too complex and agency specific to lend itself to a formal set of rules. There were, however, a number of
general guidelines identified that may prove helpful in considering micro simulation for a given study.
The insights are summarized below.
9.1 Selecting a Modeling Approach
In many cases, the model selection process is dictated by study requirements that have very little to do
with the technical aspects of the project. Answering the following basic questions will often determine if
microscopic simulation is a viable option for the problem at hand.
Study scope. Are the study impacts expected to be localized to a small subarea, along a corridor, or
regional?
Funding. Is available funding sufficient for the modeling approach under consideration?
Scheduling. How much time is available to conduct the study? Can the model be developed and applied
within the schedule requirements given the number of alternatives to be evaluated?
Expertise. Does the agency have the expertise required by the modeling approach? Does the agency
have quick access to expert consultants?
Staffing. Does the agency have sufficient staff resources to meet the schedule requirements? Does the
agency have quick access to contractors?
Model suitability. Will the modeling approach satisfy the study requirements? Will it be able to produce
the necessary performance measures at the required fidelity? Is lane-based analysis required?
Data quality and availability. Are data available at the required resolution? Are the data consistent?
Does the data need to be collected?
Visualization capabilities. Is visualization important to decision-makers and public outreach?
Legal mandates. Is there a legal mandate for using a specific modeling approach?
In developing any traffic model, it is essential that modelers reflect on whether the model’s capabilities
fit the purpose of the analysis. Also, practitioners should keep in mind that the accuracy of the model’s
results is heavily dependent on the quality of the input data. Thus, modelers should determine
beforehand what data are required for the modeling approach, what data are available or need to be
collected, and how the data will be assembled to create an internally consistent representation of a
specific point in time.
Choosing a modeling approach is always a trade-off between accuracy and cost. It is important to
recognize upfront that simulation modeling by its very nature requires more time and cost to develop
Best Practices in the Use of Micro Simulation Models
59
and apply than traditional modeling approaches. One rule of thumb is that mesoscopic models cost
about an order of magnitude (i.e., ten times) more than macroscopic models per link or intersection,
and microscopic models cost ten times more than mesoscopic models. This has obvious implications for
the number of links or intersections that can be modeled using mesoscopic or microscopic simulations
given a fixed schedule and budget.
Even if the time and budget is not an issue, the size or nature of the problem can make simulation less
desirable. Studies that impact large areas or regions are difficult to model using microscopic techniques
and existing software. On the other hand, if the impact area is small and the issues involve operations,
microscopic simulation i the logical choice. It is worth noting that mesoscopic models can only
approximately represent signal controls, and as such their estimates of intersection capacity and delay
are at most simplistic. If the application at hand requires a detailed representation of signal timing and
coordination and vehicle interactions at intersections or ramps, microscopic simulation is needed.
Similarly, if the application requires designing signal timing and progression, a signal optimization tool
such as Synchro is better suited for the application.
For studies that combine microscopic detail with regional impacts, a mixed or nested modeling approach
is desirable. The macro-level model considers regional demand impacts and major route choice changes.
The meso-level model evaluates the time-dependent implications of these changes and prepares the
detailed demand data for a subarea micro-level simulation. The performance data from the more
detailed model are used to update and refine the performance measures and assumptions used in the
more aggregate model. A convergence process is used to equilibrate the whole system.
Table 4 provides a synthesis of the data collected from the web survey and case study interviews,
comments made by the peer exchange panel, and information gleaned from the literature to estimate
the potential applicability of a given modeling approach to various study considerations.
Table 4: General Applicability of Various Modeling Approaches
Applicability of the Modeling Approach
Decision Criteria Criteria
Value Options Macroscopic Simulation
Mesoscopic Simulation
Microscopic Simulation
Study Characteristics
Scope Regional Yes Maybe Rarely
Corridor Yes Yes Maybe
Subarea Maybe Yes Yes
Network Size Large (> 10K Links, > 3K Nodes, > 1,000 Zones)
Yes Maybe Rarely
Medium Yes Yes Maybe
Small (< 1K Links, < 400
Nodes, < 100 Zones) Yes Yes Yes
Time Periods 24 Hours Yes Maybe Rarely
6 Hours Yes Yes Maybe
Peak Period Yes Yes Yes
Peak Hour Maybe Yes Yes
Best Practices in the Use of Micro Simulation Models
60
Applicability of the Modeling Approach
Decision Criteria Criteria
Value Options Macroscopic Simulation
Mesoscopic Simulation
Microscopic Simulation
Demand Large (> 1 M vehicles) Yes Maybe Rarely
Intermediate Yes Yes Maybe
Small (< 200k vehicles) Yes Yes Yes
Agency Resources
Modeling Expertise Microscopic Maybe Yes Yes
Mesoscopic Maybe Yes Maybe
Macroscopic Yes Maybe Maybe
Staffing > 2 Yes Yes Yes
< 2 Yes Yes Maybe
Funding > $1M Yes Yes Yes
$250K-$1M Yes Yes Yes
<$250K Yes Yes Maybe
Time Deadlines > 12 Months Yes Yes Yes
4-12 Months Yes Yes Maybe
< 4 Months Yes Maybe Rarely
Data Characteristics
Quality Consistent
and Balanced Yes Yes Yes
In Between Yes Yes Maybe
Needs Reconciling
and Balancing Yes Maybe Rarely
Fidelity < 15 Minutes Rarely Yes Yes
15 Minutes – 1 Hour Maybe Yes Maybe
> 1 Hour Yes Maybe Rarely
Performance Measures
Required Accuracy < 15 Minutes Rarely Yes Yes
15 Minutes – 1 Hour Maybe Yes Maybe
> 1 Hour Yes Maybe Rarely
Analysis Dimension Vehicle/Person-Based Rarely Yes Yes
Link-Based Yes Yes Yes
Path-Based Rarely Yes Yes
Network-Based Yes Yes Yes
Desired Functionalities
Animation No Yes Yes
Weaving/Merging No No Yes
Queuing/Shock Waves No Somewhat Yes
Link-Based Flow Model Yes Yes Yes
Lane-Based Flow Model No No Yes
Signals No Somewhat Yes
Best Practices in the Use of Micro Simulation Models
61
Applicability of the Modeling Approach
Decision Criteria Criteria
Value Options Macroscopic Simulation
Mesoscopic Simulation
Microscopic Simulation
Sign Control No Somewhat Yes
Other
Legal Mandate Use Dynamic Model No Yes Yes
Use Static Model Yes No No
None Yes Yes Yes
9.2 Selecting Microscopic Simulation Software
It is rare that a single software package will fit all types of simulation needs. The following are some factors considered by study participants when choosing a particular software package for a given project.
Non-technical factors
Level of expertise within the project team and the client agency
Level of support from the software supplier or vendor
The quality of the software documentation
Training required to develop a model using the tool
Level of transparency of the package structure and outputs
Prior applications and available data
Cross-sectional success stories
Technical factors
Network size limitations
Parallel computing and multi-threading capabilities
Software maturity and track record
Experience in applying a package for different network sizes, i.e., the scale of application
Suitability of the package to simulate the phenomenon that an agency wishes to investigate, e.g., pedestrian movements
Sensitivity of the model parameters to the specific features of the proposed scenarios
Accuracy of vehicle movement logic such as gap acceptance, lane changing, and car-following maneuvers
Driver behavior characteristics
Default validation and calibration parameters
Visualization capabilities and input data formats
User interface control and flexibility
Compatibility and integration with other traffic modeling tools
Availability of conversion and network editing utilities
Short-term vs. long-term modeling capabilities
Best Practices in the Use of Micro Simulation Models
62
APPENDIX A: CASE STUDY SUMMARIES
Caltrans Interstate 5 Traffic Study (I-5/91 to I-5/405)
Reporting agency Caltrans
Study scope Corridor Simulation software VISSIM (micro simulation)
Why was simulation used? Traditional modeling tools did not provide sufficient policy/analysis sensitivities
Mandated by legislature for CSMP
Study objective(s) Determine whether the improvements the legislature funded actually worked
Perform an effective cost-benefit analysis of projects
Determine what the maximum inputs of TSM/TDM were for the corridor Time periods modeled Daily, AM peak hour, AM peak period, PM peak hour, PM peak period, off-peak
period, weekday
Scenarios modeled Demand management, capacity improvements, operational improvements
Number scenarios More than nine Simulation time Minimum: 1-2 hours; maximum: 3-4 hours; average: 3-4 hours
Planning horizons modeled Current year, long-range forecast
Budget > $1M
Duration Ongoing since 02/2008
Budget breakdown by task and duration
Data preparation: 50 percent [7 months]
Model calibration and implementation: 20 percent [6 months]
Model application: 5 percent [12 months]
Results and visualization: N/A [2 months] Number of simulation staff No data was available
Network data sources TransCAD, GIS Street Layers, NAVTEQTIGER Files, Caltrans As-Builts
Network Size No data was available.
Model features Trucks, signal progression, traveler value of time, incidents, merging/weaving, time-of-day differences
Demand preparation Estimated from traffic counts
Synthesized from entry rates and turning fractions
Synthesized from planning-level trip tables (daily and peak period)
Referenced back to GPS information Performance measures Network-wide: average speed, VMT, VHT, Delay
Link: (time-dependent) speeds, volumes, densities, travel times
Path: travel times
Other: formation/dispersion rate of bottlenecks Calibration results Satisfactory: link capacities, link volumes (15 min), screenline volumes, link speeds,
link travel times, link delays, turning movements, queue lengths, speed distribution
Model Maintenance All aspects of the model are being maintained.
Major difficulties Accurately modeling traveler behavior
Achieving reasonable calibration/validation
Collecting data and making the model accurately replicate observed bottlenecks Conclusions The simulation answered the intended questions for this study.
Simulation results led to better integration of micro simulation throughout the capital outlay process and more effective or accurate answers.
Simulation results gave further clarity to information that is displayed in tables.
Next (similar) simulation study is likely to be less expensive.
The Department is now capable of doing a simulation study on its own.
The Department is likely to consider another simulation tool for the next study.
Best Practices in the Use of Micro Simulation Models
63
City of Moreno Valley, CA, TRANSIMS Implementation
Reporting agency City of Moreno Valley
Study scope Combination of regional (macroscopic) and subarea (microscopic)
Simulation software TRANSIMS (micro simulation)
Why was simulation used? Funding incentives Study objective(s) What link and intersection improvements are required to accommodate the
proposed zoning changes while maintaining a given level of service standard?
Will the additional truck traffic associated with the proposed zoning changes result in the need to increase the Traffic Index (and resulting structural cross-sections) of the impacted arterials and streets?
Will commute patterns be altered so significantly as to require major geometric changes to planned interchange improvements?
To what extent would the City’s existing and proposed future industrial areas interact, and absent mitigation, how would trucks route themselves?
Would these trucks impact existing or future non-industrial land uses in between? Time periods modeled Daily, AM peak period, PM peak period
Scenarios modeled Capacity Improvements, land-use/policy impacts, operational improvements Number scenarios More than three
Simulation time Average: > 24 hrs
Planning horizons modeled Current year, long-range forecast
Budget $250K-$500K
Duration Ongoing since 10/2008 Budget breakdown by task and duration
Data preparation: 25 percent [3 months]
Model calibration and implementation: 20 percent [3 months]
Model application: 30 percent [6 months]
Results and visualization: 15 percent [1.5 months] Number of simulation staff Five
Network data sources TransCAD
Network Size Links, nodes, zones, signals
Model features Trucks, dual-ring signal, HOV Lanes Demand preparation Synthesized from travel diaries
Performance measures Still in development as project is still in progress
Calibration results Satisfactory: link capacities, link volumes (hourly), screenline volumes, link speeds, link travel times, link delays, turning movements, queue lengths, speed distribution
Model Maintenance All aspects will be maintained and reused for future analyses.
Major difficulties Accurately modeling traveler behavior
Achieving convergence or stabilization
Achieving reasonable calibration/validation
Compute time to gauge adjustments to input data Conclusions Simulation results are satisfactory, anticipated benefits are reported.
Simulation results led to better capital investment decisions and more effective or accurate answers.
Not cost-effective, but simulation was definitely warranted.
Visualization was of neutral impact to decision-makers.
Truck impacts could not be analyzed at a detailed level without simulation.
Next (similar) simulation study is likely to be less expensive.
The City is now capable of doing a simulation study on its own.
The City will be inclined to do more studies of this nature in the future.
Best Practices in the Use of Micro Simulation Models
64
New York City Department of Transportation (NYCDOT) Broadway Boulevard Project in Midtown Manhattan
Reporting agency New York City Department of Transportation (NYCDOT)
Study scope Subarea
Simulation software AIMSUN (micro simulation)
Why was simulation used? Traditional modeling tools did not provide sufficient policy/analysis sensitivities.
Visualizing traffic was important for public outreach and decision-making.
Agency preference. Study objective(s) Determine the effects of removal of one major artery on mobility and safety of
vehicular, pedestrian, and bicycle traffic
Determine air quality impacts due to diversion of traffic Time periods modeled PM peak hour, PM peak period
Scenarios modeled Capacity improvements, operational improvements
Number scenarios Five
Simulation time Minimum: 3-4 hours; maximum: 6-7 hours; average: 4-5 hours Planning horizons modeled Short-range forecast (2008)
Budget $500K-$1M
Duration 1 year
Budget breakdown by task and duration
Data preparation: 10 percent [2 months]
Model calibration and implementation: 50 percent [6 months]
Model application: 10 percent [12 months]
Results and visualization: 5 percent [1 month]
Number of simulation staff Major difficulties Ability to model a large urban area network
Accurately modeling traveler behavior
Achieving reasonable calibration/validation Network data sources Network was fully functional from a previous study.
Network Size 2,000 links, 500 nodes, 500 signals, 60,000 trips, 3 highway miles, 45 arterial miles
Model features Pre-timed signals, actuated signals, signal progression, signal preemption, turn prohibitions, pre-trip information, lane closures
Demand preparation Synthesized from planning-level trip tables (peak period)
Performance measures Network-wide: average speed, VMT, VHT, stopped/delay time
Link: (time-dependent) volumes, travel times Model Maintenance Model is maintained
Calibration results Satisfactory: screenline volumes
Somewhat satisfactory: link capacities, link volumes (hourly), turning movements, speed distribution, travel times
Not satisfactory: queue length, link delays Conclusions Not cost-effective.
Simulation results were satisfactory.
Simulation results answered the intended questions for this study.
Simulation results led to implementation of the project as a pilot to verify the model results and fine-tune the proposed improvement measures.
Visualization was very useful to decision-makers.
Next (similar) simulation study is likely to be as expensive.
The Department is now capable of doing a simulation study on its own.
The Department will be inclined to do more studies of this nature in the future.
Best Practices in the Use of Micro Simulation Models
65
Caltrans I-5 North Coast Traffic Study
Reporting agency Caltrans, District 11
Study scope Corridor Simulation software FREQ, Synchro, TransModeler
Why was simulation used? Traditional modeling tools did not provide sufficient policy/analysis sensitivities.
Visualizing traffic was important for public outreach and decision-making.
Agency preference. Study objective(s) Understand traffic operations and behavior of project alternatives
Provide performance measures basis for the alternatives
Evaluate proposed construction phasing and define mobility benefits of each
Develop visual animations of project alternatives for various audiences
Utilize the simulation work as a corridor operations management tool
Time periods modeled AM peak hour, AM peak period, PM peak hour, PM peak period Scenarios modeled Demand management, capacity improvements, operational improvements
Number scenarios Eight
Simulation time Minimum: 5-15 minutes; maximum: 2-3 hours; average: 1-2 hours
Planning horizons modeled Current year, short-range forecast, long-range forecast
Budget $500K-$1M Duration
Budget breakdown by task and duration
Data preparation: 35 percent [6 months]
Model calibration and implementation: 35 percent [5 months]
Model application: 15 percent [3 months]
Results and visualization: 10 percent [2 months] Number of simulation staff
Network data sources TransCAD
Network Size Freeway miles: 26 Model features Trucks, pre-timed signals, HOT lanes, HOV lanes, merging/weaving
Demand preparation Estimated from traffic counts
Synthesized from planning-level trip tables
Performance measures Network-wide: average speed, VMT
Link: (time-dependent) speeds, volumes, travel times
Path: travel times Calibration results Very satisfactory: screenline volumes, speed distribution
Satisfactory: link capacities, link volumes (hourly), turning movements, queue lengths, travel times, link delays
Model Maintenance Model is being maintained for future use; maintenance costs: $35,000 per year.
Major difficulties Developing visualization results for stakeholders
Ability to model a large urban area network
Accurately modeling traveler behavior
Achieving convergence or stabilization
Achieving reasonable calibration/validation Conclusions Simulation results were satisfactory and led to better capital investment decisions
and more effective or accurate answers.
Visualization was very useful to decision-makers.
Next (similar) simulation study is likely to be as expensive.
The firm is not capable of doing a simulation study on its own because of the lack of time and staff resources.
Cost-effective.
The Department is inclined to modify the procedures, budget, and schedule before the next study.
Best Practices in the Use of Micro Simulation Models
66
Sacramento Area Council of Governments (SACOG) Integration of an Activity-Based Model with TRANSIMS
Reporting agency Resource Systems Group Inc. (RSG)
Study scope Regional
Simulation software TRANSIMS
Why was simulation used? Funding incentives Study objective(s) Integration of the activity-based demand model used by the region for planning
purposes (DaySim) with TRANSIMS
Time periods modeled Daily, AM peak period, PM peak period, Off-peak period
Scenarios modeled Capacity improvements Number scenarios Two
Simulation time Average: 18-24 hours
Planning horizons modeled Current year
Budget < $250K
Duration Ongoing since 10/2008
Budget breakdown by task and duration
Data preparation: 30 percent [2 months]
Model calibration and implementation: 50 percent [8 months]
Model application: 20 percent [2 months]
Results and visualization: N/A [2 months] Number of simulation staff 15
Network data sources CUBE, TP+
Network Size 7,000 links, 4,500 nodes, 2,600 signals, 1,000 stop/yield signs, 5M daily trips
Model features Trucks, dual-ring signals, turn prohibitions, HOT lanes, HOV lanes, merging/weaving, time-of-day differences
Demand preparation Activity-based demand model output
Performance measures Link: (time-dependent) speeds, volumes, travel times Calibration results
Model Maintenance Model is being maintained for future by FHWA.
Major difficulties Ability to model a large urban area network
Accurately modeling traveler behavior
Achieving convergence or stabilization
Achieving reasonable calibration/validation
Conclusions Simulation results were satisfactory and did answer the intended questions.
Next simulation study is expected to be as expensive.
RSG is capable of doing a similar study on its own.
Study was not cost-effective.
RSG will likely modify the procedures, budget, and schedule before the next study.
Best Practices in the Use of Micro Simulation Models
67
Caltrans I-80 Integrated Corridor Management (ICM) project
Reporting agency Caltrans
Study scope Corridor, subarea
Simulation software Paramics (Quadstone), SimTraffic, Synchro Why was simulation used? Traditional modeling tools did not provide sufficient policy/analysis sensitivities Study objective(s) Evaluate the impacts of ramp metering, active traffic management, variable speed
limits
Determine the optimum operational strategy for ramp metering, active traffic management, variable speed limits
Time periods modeled AM peak period, PM peak period
Scenarios modeled Demand management, operational improvements
Number scenarios Six Simulation run time Average: 1-2 hours
Planning horizons modeled Current year, short-range forecast
Budget > $1M
Duration Ongoing since 03/2008
Budget breakdown by task and duration
Data preparation: 25 percent [6 weeks]
Model calibration and implementation: 50 percent [6 months]
Model application: 5 percent [12 months]
Results and visualization: 5 percent [2 weeks] Number of simulation staff 40
Network data sources No data was available
Network Size No data was available
Model features Buses, trucks, pre-timed signals, turn prohibitions, bus lanes, HOV lanes, incidents, time-of-day differences
Demand preparation Synthesized from planning-level trip tables
Estimated from traffic counts Performance measures Network-wide: average speed
Link: (time-dependent) speeds, volumes, travel times
Path: travel times Calibration results Satisfactory: link capacities, link volumes (hourly), screenline
volumes, queue lengths, speed distribution, travel times, link delays
Model Maintenance Model is being maintained for future years by Caltrans.
Major difficulties Accurately modeling traveler behavior
Achieving reasonable calibration/validation Conclusions Simulation results adequately answered the intended questions.
Simulation results were satisfactory and led to more effective or accurate answers.
Visualization was of neutral impact to decision-makers.
Good quality data are always needed.
Next simulation study is expected to be as expensive.
Agency is now capable of doing a similar simulation study on its own.
Not cost-effective.
Agency likely to consider another simulation tool next time.
Best Practices in the Use of Micro Simulation Models
68
Caltrans I-580 Corridor System Management Plan
Reporting agency Dowling Associates
Study scope Corridor, Subarea
Simulation software Paramics (Quadstone), SimTraffic, Synchro
Why was simulation used? Agency preference
Mandated by Caltrans Corridor System Management Plans (CSMP) Study objective(s) Identify current and future freeway bottlenecks
Develop and prioritize mitigations for bottlenecks
Evaluate performance of freeway and arterials with recommended mitigations Time periods modeled AM peak period, PM peak period
Scenarios modeled Capacity improvements, operational improvements
Number scenarios Four
Simulation time Average: 1-2 hours
Planning horizons modeled Current year, short-range forecast, long-range forecast
Budget $250K-$500K Duration 11 months
Budget breakdown by task and duration
Data preparation: 30 percent [7 weeks]
Model calibration and implementation: 55 percent [7 weeks]
Model application: 15 percent [3 weeks]
Results and visualization: N/A [2 months] Number of simulation staff Six
Network data sources Synchro for signals, aerials Network Size 30 freeway miles, 60 arterial miles, 75 signals
Model features Trucks, dual-ring signals, HOT lanes, HOV lanes, merging/weaving, time-of-day differences, signal timing and progression data, ramp metering data, saturation and service flow rates
Demand preparation Estimated from traffic counts
Synthesized from planning-level trip tables Performance measures Network-wide: average speed, VMT, VHT, stopped/delay time Calibration results Very satisfactory: link volumes
Satisfactory: queue lengths, travel times
Model Maintenance Model not maintained.
Maintenance without an identifiable upcoming project difficult to get management signoff for staff-hours.
Major difficulties Time constraint dictated by funding deadline.
Conclusions Micro simulation was an expensive way to conduct essentially a planning study.
Simulation results were satisfactory and led to better capital investment decisions.
Simulation was used to give “authority” to answers generated by other means.
Length of simulation periods (4 hours AM, 5 hours PM) and 30-mile length of study precluded meaningful visualization of the results.
The simulation was too microscopic for the duration and extent of the study area.
Results highlighted the great dichotomy between micro simulation development costs and their usefulness in planning studies.
We were very efficient and focused—time deadline facilitated this.
Next study, without a tight focus or deadline, will naturally exceed the cost.
The firm always has been capable of doing micro simulation studies on its own
Not cost-effective.
The firm would be reluctant in using simulation software again for a similar study.
Best Practices in the Use of Micro Simulation Models
69
MnDOT I-94 Managed Lane Operations Study
Reporting agency FHWA
Study scope Corridor
Simulation software CORSIM
Why was simulation used? Agency preference Study objective(s) Determine whether or not to add a managed lane to an existing freeway
Determine the performance of the four concepts that were developed
Time periods modeled AM peak period, PM peak period, weekday
Scenarios modeled Capacity improvements; transit, pedestrian, and bicycle improvements; operational improvements
Number scenarios Four
Simulation time Average: 3-4 hours
Planning horizons modeled Current year, short-range forecast, long-range forecast Budget $250K-$500K
Duration 5 months
Budget breakdown by task and duration
Data preparation: 10 percent [1 month]
Model calibration and implementation: 30 percent [3 months]
Model application: 40 percent [4 months]
Results and visualization: 5 percent [1 week] Number of simulation staff
Network data sources Available from previous application. Network Size 120 links, 100 nodes, 20 freeway miles, 3 arterial miles, 20 signals, 120,000 trips
Model features Buses, pre-timed signals, bus lanes, HOT lanes, transit-only lanes, merging/weaving, time-of-day differences
Demand preparation Synthesized from freeway management system, lane-by-lane detection, on- and off-ramp detectors and stop line detectors
Performance measures Link: (time-dependent) speeds, volumes, densities, travel times
Calibration results Very satisfactory: link capacities, link volumes (15 minute and hourly), screenline volumes, turning movements, queue lengths, speed distribution, travel times, link delays, travel distances
Satisfactory: transit ridership, truck/bus operations
Somewhat satisfactory: bike/pedestrian movements Model Maintenance Model is being maintained.
Major difficulties Congestion at the ends of the model or just beyond the end of the model—these congestion points impacted the model and the results. This was accommodated in a modification to proposed modeling strategy.
Conclusions Simulation results were very satisfactory and answered the intended questions for this study.
Simulation results led to changing or reversing an existing design/policy and more effective or accurate answers.
Visualization was very useful to decision-makers.
It was clear from the study that direct access ramps were needed to make the design work.
The process was standardized about 6 years ago and modeling costs have been reduced by 75 percent.
A similar study will be as expensive.
Cost-effective.
The agency is inclined to do more studies of this nature in the future.
Best Practices in the Use of Micro Simulation Models
70
Maricopa Association of Governments (MAG) I-10 Integrated Corridor Management Analysis
Reporting agency Maricopa Association of Governments (MAG)
Study scope Regional, corridor
Simulation software DYNASMART-P/DynusT
Why was simulation used? Traditional modeling tools did not provide sufficient policy/analysis sensitivities. Study objective(s) Study the impact on traffic (e.g., route choice, travel time) due to major freeway
widening project.
Test the effectiveness of different operational strategies to reduce the congestion caused by the construction.
How realistically does the dynamic traffic assignment represent the actual conditions in terms of route choice, travel time, and departure time? How does it model transit?
Where are we in terms of providing feedback to the long-term planning model for significant route choice and mode choice changes?
How bad it will get without any changes on the operations side for I-10 compared to being without a work zone?
How much improvement can be achieved with different or combinations of strategies?
Time periods modeled Daily, AM peak period, PM peak period
Scenarios modeled Not available
Number scenarios Not available
Simulation time Not available
Planning horizons modeled Current year, short-range forecast Budget < $250K
Duration Ongoing
Budget breakdown by task and duration
Data preparation: 10 percent [1 week]
Model calibration and implementation: 30 percent [3 months]
Model application: Not available
Results and visualization: Not available Number of simulation staff 1
Network data sources Network was available from previous application. Network Size 13,693 links, 9,963 nodes, 840 freeway miles, 4,200 arterial miles, 4,000 signals,
14.5M vehicle trips/day
Model features Buses, trucks, actuated signals, turn prohibitions, HOV lanes, traveler value of time, pre-trip information, variable message signs, work zones, incidents lane closures, time-of-day differences
Demand preparation Demand data was available from previous application Performance measures Link: (time-dependent) speeds, volumes, densities, travel times
Path: vehicle trajectories, travel times Calibration results Not completed yet
Model Maintenance Simulation model is being maintained by MAG.
Major difficulties Computing power; 64-bit system looks like a must to utilize more memory Conclusions The agency is capable of doing a similar study on its own.
The agency is inclined to do more studies of this nature in the future.
Best Practices in the Use of Micro Simulation Models
71
Colorado Department of Transportation (CDOT) Downtown Denver Multimodal Access Plan (DMAP)
Reporting agency Fehr and Peers
Study scope Subarea
Simulation software VISSIM
Why was simulation used? Traditional modeling tools did not provide sufficient policy/analysis sensitivities.
Visualizing traffic was important for public outreach and decision-making. Study objective(s) Reduce congestion for existing freeway-to-freeway movements (I-5 and SR 56)
while minimizing the potential for future regional traffic to cut through on local streets.
How would the direct freeway connectors affect peak hour traffic operations based on HCM consistent performance measures?
How does network-wide performance change under each alternative?
What percent of peak hour demand served is accommodated by each alternative? Time periods modeled AM peak hour, AM peak period, PM peak hour, PM peak period, weekday
Scenarios modeled Capacity improvements, operational improvements Number scenarios Four
Simulation time Average: 30-60 minutes
Planning horizons modeled Current year, long-range forecast
Budget < $250K
Duration Ongoing since 05/2003 Budget breakdown by task and duration
Data preparation: 20 percent [4 months]
Model calibration and implementation: 15 percent [4 months]
Model application: 20 percent [3 months]
Results and visualization: 10 percent [1 month] Number of simulation staff 100
Network data sources TRANPLAN, Synchro network, GIS street layers, aerials
Network Size 10 freeway miles, 25 arterial miles, 75 signals, 25 stop/yield signs
Model features Buses, pedestrians/bicycles, trucks, pre-timed signals, actuated signals, signal progression, turn prohibitions, bus lanes, HOV lanes, truck-only lanes, transit-only lanes, stop signs, merging/weaving, time-of-day differences
Demand preparation Estimated from O-D vehicle counts and traffic counts
Synthesized from entry rates, turning fractions, and planning-level trip tables Performance measures Network-wide: average speed, VMT. VHT. stopped/delay time
Link: (time-dependent) speeds, volumes, densities, travel times
Path: vehicle trajectories, travel times
Other: percent demand served Calibration results Very satisfactory: link capacities, link volumes (15 minute and hourly),
screenline volumes, turning movements, queue lengths, speed distribution, travel times, link delays, travel distances, truck/bus operations
Satisfactory: transit ridership, bike/pedestrian movements
Model Maintenance The model is being maintained for future years by Fehr and Peers. Major difficulties Not available
Conclusions Simulation results were very satisfactory.
Simulation results led to better capital investment decisions, changing or reversing an existing design/policy, and more effective or accurate answers.
Congested networks must be evaluated on a system-wide basis.
Visualization was very useful.
Best Practices in the Use of Micro Simulation Models
72
Utah Department of Transportation (UDOT) NEPA Analysis
Reporting agency Utah DOT
Study scope Not available
Simulation software VISSIM
Why was simulation used? Traditional modeling tools did not provide sufficient policy/analysis sensitivities.
Agency preference. Study objective(s) Conduct a NEPA analysis
Time periods modeled Daily, AM peak hour, AM peak period, PM peak hour, PM peak period, off-peak period, weekend, weekday
Scenarios modeled Demand management; capacity improvements; transit, pedestrian, and bicycle Improvements; operational improvements
Number scenarios Nine
Simulation time Minimum: 5-15 minutes; maximum: 6-7 hours; average: 1-2 hours Planning horizons modeled Current year, long-range forecast
Budget < $250K
Duration Ongoing
Budget breakdown by task and duration
Data preparation: 5 percent [7 months]
Model calibration and implementation: 5 percent [6 months]
Model application: 5 percent [6 months] Number of simulation staff Three
Network data sources The transportation network was available from a previous simulation model.
Network Size 500 links, 400 nodes, 40 freeway miles, 150 arterial miles, 300 signals, 150 stop/yield signs, 20,000 vehicle trips per day
Model features Buses, pedestrians/bicycles, trucks, pre-timed signals, actuated signals, dual-ring signals, signal progression, signal preemption, turn prohibitions, bus lanes, HOT lanes, HOV lanes, transit-only lanes, reversible lanes, traveler value of time, variable message signs, stop signs, work zones, incidents, lane closures, roundabouts, merging/weaving, time-of-day differences
Demand preparation Demand was available from a previous model.
Performance measures Network-wide: average speed, VMT, VHT, stopped/delay time
Link: (time-dependent) speeds, volumes, densities, travel times
Path: vehicle trajectories, travel times
Calibration results Very satisfactory: link capacities, link volumes (15 minute and hourly), turning movements, queue lengths, speed distribution, travel times, link delays, bike/pedestrian movements
Satisfactory: screenline volumes, transit ridership, truck/bus operations, travel distances
Model Maintenance Model is being maintained for future years by Utah DOT ($100K per year).
Major difficulties Obtaining HCM-consistent results
Ability to model a large urban area network
Accurately modeling traveler behavior
Achieving convergence or stabilization Conclusions Simulation results were very satisfactory.
Simulation results led to better capital investment decisions, changing or reversing an existing design/policy, and more effective or accurate answers.
Simulation results highlighted a particular inadequacy in similar studies conducted using traditional planning practices.
Study is cost-effective, but next study will be less expensive.
The Department will be inclined to do more studies of this nature in the future.
Best Practices in the Use of Micro Simulation Models
73
Arkansas State Highway and Transportation Department (AHTD) Hwy 112/Garland Ave
Reporting agency AHTD
Study scope Corridor
Simulation software Synchro (signal optimization), VISSIM (microscopic) Why was simulation used? Traditional modeling tools did not provide sufficient policy/analysis sensitivities.
Visualizing traffic was important for public outreach and decision-making.
Agency preference.
Study objective(s) Analyze several cross-section alternatives for a highway improvement/widening project in an urban area, including motor vehicle interactions with bus transit, bicyclists, and pedestrians
Time periods modeled AM peak period, PM peak period, weekday
Scenarios modeled Capacity improvements; transit, pedestrian, and bicycle improvements; operational improvements
Number scenarios Nine Simulation time Average: 1-2 hours
Planning horizons modeled Long-range forecast
Budget < $250K
Duration Ongoing since 09/2008
Budget breakdown by task and duration
Data preparation: 20 percent [2 weeks]
Model calibration and implementation: 30 percent [2 weeks]
Model application: 5 percent [6 months] Number of simulation staff Three
Network data sources Network was available from a previous simulation model.
Network Size Not available
Model features Buses, pedestrians/bicycles, trucks, pre-timed signals, actuated signals, signal progression, stop signs, time-of-day differences
Demand preparation Demand data was available from a previous simulation model.
Performance measures Network-wide: stopped/delay time
Path: vehicle trajectories, travel times
Other: cycle failures, queue lengths Calibration results Very satisfactory: link capacities, link volumes (15 minute and hourly),
screenline volumes, turning movements, queue lengths, bike/pedestrian movements, truck/bus operations
Model Maintenance The model is being maintained by AHDT.
Major difficulties Obtaining HCM-consistent results
Accurately modeling traveler behavior
Achieving reasonable calibration/validation
Conclusions Simulation results were very satisfactory and answered the intended questions for this study.
Simulation results led to better capital investment decisions, changing or reversing an existing design/policy, and more effective or accurate answers.
Visualization was very useful to decision-makers.
The simulation results highlight a particular inadequacy in similar studies conducted using traditional planning practices.
The Department will be inclined to do more studies of this nature in the future.
Best Practices in the Use of Micro Simulation Models
74
Kansas Department of Transportation (KDOT) Johnson County Gateway I-435/I-35/K-10
Reporting agency Kansas Department of Transportation (KDOT)
Study scope Corridor
Simulation software VISSIM (micro simulation)
Why was simulation used? Traditional modeling tools did not provide sufficient policy/analysis sensitivities.
Visualizing traffic operations was important for public outreach and decision-making.
Agency preference. Study objective(s) Analyze the present and future operations of a complex section of interchanges
where two interstates and another major highway that is highly congested come together, and create a new design that will function well into the future.
The study is intended to answer how to fix this bottleneck.
Time periods modeled AM peak period, PM peak period Scenarios modeled Capacity improvements
Number scenarios Six
Simulation time No data were available.
Planning horizons modeled Current year, long-range forecast
Budget > $1M
Duration Ongoing since 08/2008 Budget breakdown by task and duration
No data were available.
Number of simulation staff Three
Network data sources Transportation network was available from a previous simulation model. Network Size 1,093 links, 33 nodes, 13 freeway miles, 6 arterial miles, 19 signals, 2 stop/yield signs,
70,000 vehicle trips/day
Model features Trucks, pre-timed signals, stop signs, merging/weaving, time-of-day differences
Demand preparation demand data was available from a previous simulation model.
Performance measures Link: (time-dependent) speeds, volumes, densities, travel times Calibration results Satisfactory: link volumes, screenline volumes, travel times Model Maintenance This model is still in development so it will be maintained only if a use for it is needed
in the future. However, it will be saved and available if needed.
Major difficulties Ability to model a large urban area network
Accurately modeling traveler behavior
Achieving reasonable calibration/validation Conclusions Simulation results were satisfactory and answered the intended questions.
Simulation results led to better capital investment decisions and more effective or accurate answers.
Animation was very useful to decision-makers.
Simulation results did not highlight a particular inadequacy in similar studies conducted using traditional planning practices.
The next simulation-based study is expected to be as expensive.
The study was cost-effective.
The Department will be inclined to do more studies of this nature in the future.
Best Practices in the Use of Micro Simulation Models
75
San Diego Association of Governments (SANDAG) I-15 ICM Project
Reporting agency San Diego Association of Governments (SANDAG)
Study scope Corridor
Simulation software TransModeler (microscopic)
Why was simulation used? Agency preference
To link the travel demand model software with the traffic simulation software Study objective(s) Determine how to maximize person-throughput when dealing with recurrent and
non-recurrent congestion.
The goal is to use the software in a real-time traffic management center with variable message signs on the corridor warning drivers of an incident and providing route-path, temporal, and modal options for avoiding the congestion.
Time periods modeled AM peak period
Scenarios modeled Demand management, transit/pedestrian improvements, operational improvements
Number scenarios > Nine
Simulation time Average 1-2 hours
Planning horizons modeled Short-range forecast
Budget $500K-$1M Duration Ongoing
Budget breakdown by task and duration
Data preparation: 10 percent [2 months]
Model calibration and implementation: 60 percent [12 months]
Model application: 5 percent [1 month]
Number of simulation staff Seven
Network data sources TransCAD, Arc/INFO Network Size 936 links, 801 nodes, 52 freeway miles, 122 arterial miles, 181 signalized
intersections, 20 stop/yield signs, 355,458 vehicle trips
Model features Buses, signals, signal progression and preemption, turn prohibitions, HOT lanes, HOV lanes, reversible lanes, pre-trip information, on-board route information, variable message signs, stop signs, incidents, lane closures, merging/weaving
Demand preparation Estimated from traffic counts
Synthesized from planning-level trip tables
Performance measures Network-wide: average speed, stopped/delay time
Link: (time-dependent) speeds, volumes, travel times Calibration results Very satisfactory: queue lengths, link speeds
Satisfactory: link capacities, link volumes, travel times, link delays, turning movements
Model Maintenance Model is being maintained for future by SANDAG.
Intended to be used as an incident response for the Decision Support System. Major difficulties Accurately modeling traveler behavior
Achieving reasonable calibration/validation Conclusions Simulation results were satisfactory.
Simulation results led to better capital investment decisions, changing or reversing an existing design/policy, and more effective or accurate answers.
Animations of the problems are more affective at communicating the problems with the network than means such as maps, tables, and reports.
Data collection was inadequate, thus the next study is expected to be more expensive.
Study did not highlight an inadequacy in traditional planning practices.
Not cost-effective.
The agency will need to modify the procedures, budget, and schedule before the next study.
Best Practices in the Use of Micro Simulation Models
76
San Francisco County Transportation Authority (SFCTA) Doyle Drive Study
Reporting agency SFCTA
Study scope Subarea
Simulation software DynamEQ (simplified micro simulation-based DTA) Why was simulation used? Agency preference
Study objective(s) Identify route diversions and potential bottlenecks and queues as a result of the construction
Time periods modeled PM peak period
Scenarios modeled Capacity improvements Number scenarios Five
Simulation time Average: 5 hours
Planning horizons modeled Current year, short-range forecast
Budget < $250K
Duration Two months Budget breakdown by task and duration
Data preparation: 33 percent 3 weeks
Model calibration and implementation: 55 percent 4 weeks
Model application: 11 percent 1 week
Model visualization/results: 11 percent 1 week Number of simulation staff Three
Network data sources CUBE
Network Size 6,700 links, 2,900 nodes, 5 freeway miles, 40 arterial miles, 240 signalized intersections, 700 stop/yield signs, 160,000 vehicle trips/3 hours
Model features Buses, pre-timed signals, signal progression, turn prohibitions, stop signs, merging/weaving, time-of-day differences
Demand preparation Synthesized from planning-level trip tables
Estimated from traffic counts Performance measures Network-wide: average speed, VMT, VHT, stopped/delay time
Link: (time-dependent) volumes, travel times
Path: travel times
Other: cycle failures Calibration results Very satisfactory: path travel times, screen lines, link counts
Model Maintenance Model is being maintained by PB, SFCTA, and INRO
Major difficulties Very tight deadlines Conclusions Simulation results were very satisfactory and answered intended questions.
Simulation results led to more effective or accurate answers.
Visualization was useful to decision-makers.
Similar studies will be less expensive.
Study was very cost-effective. The super tight deadlines and strong involvement with software vendor helped.
The agency is inclined to do more studies of this nature in the future.