Citation
Calibration of a microscopic traffic simulation model for before-and-after traffic study

Material Information

Title:
Calibration of a microscopic traffic simulation model for before-and-after traffic study
Creator:
Bourdon, Brandon John
Publication Date:
Language:
English
Physical Description:
67 leaves : illustrations ; 28 cm

Thesis/Dissertation Information

Degree:
Master's ( Master of Science)
Degree Grantor:
University of Colorado Denver
Degree Divisions:
Department of Civil Engineering, CU Denver
Degree Disciplines:
Civil engineering

Subjects

Subjects / Keywords:
Traffic flow -- Computer simulation ( lcsh )
Traffic flow -- Computer simulation ( fast )
Genre:
bibliography ( marcgt )
theses ( marcgt )
non-fiction ( marcgt )

Notes

Bibliography:
Includes bibliographical references (leaves 66-67).
General Note:
Department of Civil Engineering
Statement of Responsibility:
by Brandon John Bourdon.

Record Information

Source Institution:
|University of Colorado Denver
Holding Location:
Auraria Library
Rights Management:
All applicable rights reserved by the source institution and holding location.
Resource Identifier:
48713451 ( OCLC )
ocm48713451
Classification:
LD1190.E53 2001m .B68 ( lcc )

Full Text
CALIBRATION OF A MICROSCOPIC TRAFFIC SIMULATION
MODEL FOR A BEFORE-AND-AFTER TRAFFIC STUDY
by
Brandon John Bourdon
B.S., South Dakota School of Mines and Technology, 1997
A thesis submitted to the
University of Colorado at Denver
In partial fulfillment
of the requirements for the degree of
Master of Science
Civil Engineering
2001


This thesis for the Master of Science
degree by
Brandon John Bourdon
has been approved
by
Bruce Janson
^Date
2L^ZlrV/


Bourdon, Brandon John (M.S., Civil Engineering)
Calibration of a Microscopic Traffic Simulation Model for a Before-and-After Traffic Study
Thesis directed by Assistant Professor Sarosh I. Khan
ABSTRACT
Traffic engineers use various microscopic traffic models to perform a variety of tasks. These
models are increasingly being used to analyze complex transportation networks effectively,
and to model a variety of geometric, traffic, and operational conditions. The recommended
methods to calibrate and validate a model vary considerably depending on the application of
the model. Many different methods of calibration and validation are discussed in the
literature. The purpose of this thesis is to examine the benefits of calibrating and validating a
microscopic model for a before-and-after traffic study. The main goal is to illustrate how the
level of detail included in network representation and calibration and validation affect the
outcome of an analysis. Several models representing existing conditions were developed at
various levels of detail. Each of these models were used for calibration, validation, and to
develop future scenarios. The delay and travel times were used to examine differences in
the models calibrated and validated at different levels. The results show that as the level of
service deteriorates, the level of calibration and validation plays a more significant role.
Although a more detailed level of representation results in higher modeling costs, it also
provides better validation results.
This abstract accurately represents the content of the candidates thesis. I recommend its
publication.
iii


CONTENTS
Figures............................................................................vii
Tables..............................................................................ix
Chapter
1. Introduction.....................................................................1
1.1 Objective.......................................................................1
2. Literature Review................................................................3
2.1 Calibration and Validation......................................................3
2.2 Rainer Wiedemanns Theory.......................................................6
2.3 VISSIM..........................................................................6
3. Study Site and Data Collection..................................................10
3.1 Study Site.....................................................................10
3.2 Data Collection................................................................13
3.2.1 Field Visits.................................................................14
3.2.2 GPS Survey Data..............................................................14
3.2.3 Traffic Data Collection......................................................17
3.2.3.1 Traffic Volumes.............................................................17
3.2.3.2 Video Data..................................................................17
3.2.3.3 Manual Data Collection..................................................... 17
3.2.4 VISSIM Output Data...........................................................20
4. Methodology.....................................................................21
4.1 Terminology....................................................................21
4.2 Model Development..............................................................23
iv


4.2.1 Model for Existing Conditions
24
4.2.1.1 Level 1....................................................................24
4.2.1.2 Level 2....................................................................24
4.2.1.3 Level 3....................................................................26
4.2.1.4 Level 4....................................................................26
4.2.1.5 Level 5....................................................................26
4.2.1.6 Level 6....................................................................26
4.2.2 Future Model Development......................................................27
4.3 Visual Basic Tools...............................................................30
4.3.1 VISSIM Batch Operations Module................................................30
4.3.2 Text to Excel File Format Converter...........................................30
4.3.3 Simulation File Creator.......................................................31
4.4 Statistical Analysis............................................................31
4.5 Preparation of Output...........................................................32
5. Results..........................................................................33
5.1 Models Representing Existing Conditions.........................................33
5.2 Examining the Five Additional Segments..........................................36
5.3 Future Scenarios................................................................41
5.3.1 Arterial Travel Times..........................................................41
5.3.2 Approach Delays...............................................................51
5.3.3 Cost Difference Analysis......................................................54
6. Conclusion and Recommendations...................................................57
Appendix
A.1 Program that Runs Batch VISSIM Simulations.......................................59
A.2 Program the Converts Text Output Files in the Excel Format......................60
v


A.3 Program that Creates Simulation Input Files...................................62
Glossary..........................................................................64
References........................................................................66
vi


FIGURES
Figure
Figure 3.1: Southwest Segment General Site Information.............................11
Figure 3.2: Northeast Segment General Site Information.............................12
Figure 3.3: Southwest Segment Geometric Conditions.................................15
Figure 3.4: Northeast Segment Geometric Conditions.................................16
Figure 3.5: Southwest Segment Balanced Peak Hour Turning Volumes...................18
Figure 3.6: Northeast Segment Balanced Peak Hour Turning Volumes...................19
Figure 4.1: Flow Chart of the Typical Calibration and Validation Process.............22
Figure 4.2: Levels of Analysis.......................................................25
Figure 4.3: Schematic of Existing Models Relation to the Future Models...............29
Figure 4.4: Access Form for Visual Basic Tools.......................................30
Figure 5.1: Travel Time Differences on Stout from Speer to 14th......................35
Figure 5.2: Travel Time Differences for Rights from Stout to California on 14th......35
Figure 5.3: Travel Time Differences on Stout from 14th to 15th.......................36
Figure 5.4: Travel Times on Stout from Entrance Point to Northbound Speer............37
Figure 5.5: Travel Times on Stout from Northbound Speer to the Exit Point............38
Figure 5.6: Travel Times on 14th Street..............................................38
Figure 5.7: Travel Times on 15th Street..............................................39
Figure 5.8: Travel Times on California Street to the Exit Point......................39
Figure 5.9: Future Scenario A Travel Times on 14th Street............................42
Figure 5.10: Future Scenario B Travel Times on 14th Street..........................42
Figure 5.11: Future Scenario C Travel Times on 14th Street..........................43
vii


Figure 5.12: Future Scenario A Travel Times on Northern Stout Segment................44
Figure 5.13: Future Scenario B Travel Times on Northern Stout Segment................44
Figure 5.14: Future Scenario C Travel Times on Northern Stout Segment................45
Figure 5.15: Future Scenario A Travel Times on Southern Stout Segment................46
Figure 5.16: Future Scenario B Travel Times on Southern Stout Segment................46
Figure 5.17: Future Scenario C Travel Times on Southern Stout Segment................47
Figure 5.18: Future Scenario A Travel Times on 15th Street...........................48
Figure 5.19: Future Scenario B Travel Times on 15th Street...........................48
Figure 5.20: Future Scenario C Travel Times on 15th Street...........................49
Figure 5.21: Future Scenario A Travel Times on California............................50
Figure 5.22: Future Scenario B Travel Times on California............................50
Figure 5.23: Future Scenario C Travel Times on California............................51


TABLES
Table
Table 2.1: Possible Calibration Parameters in VISSIM..................................8
Table 3.1: List of Data Collected....................................................14
Table 4.1: Signal 2000 Approach Delays and Level of Service..........................28
Table 5.1: Existing Approach Delays from VISSIM......................................40
Table 5.2: Future Scenario A Approach Delay..........................................52
Table 5.3: Future Scenario B Approach Delay..........................................52
Table 5.4: Future Scenario C Approach Delay..........................................53
Table 5.5: Cost Differences Based on Detail Level for North Segment of Stout.........55
Table 5.6: Cost Differences Based on Detail Level for 15th Street....................56
IX


1. Introduction
Traffic engineers use various microscopic traffic models to perform a variety of tasks. These
models are increasingly being used to analyze complex transportation networks effectively,
and to model a variety of geometric, traffic, and operational conditions. The various types of
quantitative outputs from simulation models can be used for several types of analysis. In
addition, its animation component can be used for effective presentation.
Traffic simulation can be used to complete a variety of tasks such as: 1). To investigate the
effects of a change in policy that would be either unsafe or too expensive to initiate. 2). To
collect extensive data for controlled experiments for research. 3). To perform before-and-
after traffic studies that allow traffic engineers to evaluate improvement alternatives.
1.1 Objective
The purpose of this thesis is to investigate the benefits of calibrating and validating a
microsimulation traffic model for a before-and-after study or alternatives analysis. The main
goal will be to illustrate how the level of detail included in network representation and
calibration and validation affect the outcome of an alternative analysis.
The study area selected for a case study for this research is located in and near the central
business district (CBD) of Denver, Colorado. This area includes heavily utilized bus and
light-rail transit (LRT) routes and several one-way arterials with closely spaced signals. It is
located between the Colorado Convention Center and the Currigan Hall (Denvers old
convention center).
Light-rail transit is becoming very popular in many large cities in the United States. Denver
currently has 14 miles of LRT, that primarily serves the southern portion of the metro area.
Most simulation models typically used in the United States model highway and transit
networks separately, thus making it difficult to model the interaction between the modes and
their response to comprehensive traffic control strategies. A few attempts have been made
to use a highway traffic simulation model to include LRT operations. However, these studies
report mixed results. The need to accurately model LRT is leading some to try different
traffic models. VISSIM is a German developed microscopic traffic simulation model that is
based on work completed at the University of Karlsruhe, in Germany, beginning in the early
1970s. The work of Rainer Wiedemann at the university forms the foundation of this traffic
model.
VISSIM is a German acronym for Traffic in Towns: SIMulation. VISSIM is a microscopic
model, which means that each of the vehicles is modeled discretely as a single entity. This
model is windows based and has a variety of features such as the ability to model LRT,
pedestrians, signal-priority, and perform dynamic traffic assignment. Finding previous
research that has been done with VISSIM has been difficult for several reasons. One of the
main reasons is that the software only began to be distributed commercially in the early
1990s. Its German development has meant that its use in the United States or in English
1


speaking countries, has been relatively limited. Another reason is the lack of research
conducted in the United States due to the considerable acceptance of CORSIM, a
microscopic traffic simulation model sponsored by the Federal Highway Administration.
Calibration and validation are two steps that are involved in building a traffic model. These
steps can prove to be very time consuming and lead to a significant portion of the modeling
costs. Often these topics are not discussed in the literature or are only discussed on a
limited basis. This has lead many in both the private and public sectors to have a different
impression regarding the methods that should be used to calibrate and validate a traffic
simulation model. The purpose of this thesis is to address the impact of various levels of
calibration and validation on a before-and-after traffic study. This will address some of the
ways a model can be calibrated and validated, as well as give some guidelines regarding the
importance of these modeling steps.
Most before-and-after traffic studies are completed using a similar process. A model
reflecting existing conditions is built, calibrated, and validated to replicate existing conditions.
Once this model is completed then that model is modified to reflect the various future
scenarios that are to be investigated. Model simulations are then completed and the
averaged output data of interest, from each future model scenario, are compared to help
determine which future scenario should be implemented.
The procedure followed in this thesis will be similar to the general procedure that is followed
to complete a before-and-after traffic study. Data collected within the study area is used for
input and to calibrate and validate the VISSIM traffic model to reflect existing conditions.
This step will differ from the typical study because various levels of detail will be used to
calibrate and validate the model representing existing conditions. This will result in six
calibrated and validated models representing existing conditions, rather than one model.
Using the several models, three models that reflect three different future scenarios will be
created. The three future scenarios will reflect different rates of traffic growth applied to the
traffic model. No other changes to the future models will be made. Model simulations will
then be completed and the results will be used to study the effects of the various degrees of
calibration and validation on the output of both the existing and various future model
scenarios.
2


2. Literature Review
There are several technical articles that address the areas on which this thesis focuses.
Articles that focus on calibration and validation, Rainer Wiedemanns theory, and VISSIM
were of particular interest for this research. These areas of previous research will be
addressed in three separate sections.
2.1 Calibration and Validation
The focus placed on the calibration and validation of a traffic simulation model varies
considerably throughout the literature. Many articles either avoid the subject all together or
only briefly mention the fact that the model used was calibrated and validated. The different
ways calibration and validation is addressed in the literature make it very difficult to
determine the best method to use when calibrating and validating a simulation model.
The Highway Capacity Manual provides the basis of many traffic analysis procedures used in
the United States. The 1997 Highway Capacity Manual [1] provides very little guidance in
the area of traffic modeling, except for the methods addressed in the manual itself. In the
2000 Highway Capacity Manual [2], Chapter 31 addresses traffic simulation and other
models. It gives a good overview of how an appropriate model should be selected, as well
as modeling terminology, and typical modeling steps. Similar information was addressed
previously in an article titled, Beyond the Capacity Manual [3], This article recommended
that a section covering traffic modeling topics be added to the 2000 Highway Capacity
Manual. A good deal of this article is included in Chapter 31 of the 2000 Highway Capacity
Manual. This article describes calibration and validation as an iterative process. It also
states five possible reasons a model may not be performing accurately, of which all but one
relate to significant problems with the selected model: The input data or calibration
parameters may be coded inaccurately or inadequately. Both of the previously mentioned
sources give an overview about how calibration and validation is performed.
Model calibration and validation is addressed in several articles where the comparison of
several traffic models is the focus. The traffic models INTEGRATION, CORSIM, and
WATSim were compared in two articles [4, 5]. The purposes were similar in both articles
with the primary difference being the size of the traffic network. In both articles the volumes
produced by the models at various locations were compared to field data to determine if the
model was calibrated. An iterative process of calibration and validation was used in both
studies. The parameters that were changed to achieve reasonable volume replication were
car-following parameters, lane-changing duration, lane-change activating distances, link
lengths, capacity of particular links, and the addition of a lane-striping file. Travel speeds,
travel time, density, and queue length were the measures of effectiveness (MOEs) that were
compared for each of the three models. Field speed data was also compared to model
output where field data was available. The output MOEs were based on a 15-minute
simulation period. The results show how the models behave with respect to one another.
One of the concluding points in [5] was the high number of parameters that need to be
calibrated in order to replicate actual traffic conditions. NETSIM and TransSim II were
3


compared using a network that had LRT [6], Each of the models was calibrated using one
set of data collected in the field. Calibration in this study involved adjusting model input and
defaults to match field observed values. An iterative process was used to adjust the free-
flow speed in NETSIM and the location of detectors in TransSim. The validation consisted of
statistical tests to determine if the models properly replicated field conditions. This
calibration data included field-observed means and distributions of queue discharge
headway, start-up lost time, free-flow speed adjustments, LRT acceleration and deceleration,
and the location of LRT detection. Then the results from both models were compared to a
set of field collected MOEs to validate the models. The validation MOEs collected in the field
were link travel times, network directional travel times, intersection mean stopped delay, and
mean percent stops. There was no clear definition of the time period the output MOEs were
based upon. The main purpose was to determine if either calibrated model performed
considerably better than the other when compared to field validation data.
Calibration and validation of established models, rather than models in the development
stage, are the focus of several articles in the literature. INTEGRATION was used to build a
large model of Salt Lake City [7], Due to the size and purpose of the model, traffic volumes
were used to calibrate the model. Validation and the parameters used to calibrate the model
were not discussed in this article. The calibration was performed by comparing hourly traffic
volume data output from the model to the field collected data. This data was plotted along a
line of unit correlation to visually inspect the quality of calibration. The benefits of calibrating
NETSIM at a macroscopic level were investigated by calibrating the model based on two
MOEs used in the two-fluid model for town traffic [8], These MOEs were measured in the
field and output by a modified version of NETSIM. The two-fluid model MOEs are very
macroscopic in nature and were based on the entire simulation which included 10 or 31 15-
minute periods depending on the model run. The uncalibrated model was run in NETSIM
with default parameters and the values of the two two-fluid model MOEs produced were
compared to the measured field values. Then several iterations using trial and error were
performed where various parameters were varied until a good match between the measured
and simulated values of the two-fluid MOEs were produced. Some of the changes made to
model parameters during the calibration process was the addition of special vehicle classes
to represent slower drivers, the addition of parking maneuvers, varied maximum speeds of
some driver classes, the addition of short-term events to represent bus stops, increased
headways for slower driver types, varied overall desired speed, and different driver types
during different time periods of the day. Along with showing the benefits of calibration using
the two-fluid model MOEs, the results for each 15-minute period were also shown using more
traditional MOEs (total delay, fuel consumption, and number of stops) for both the calibrated
and uncalibrated models. The results showed that calibration was worthwhile and that
significant differences existed between various output MOEs produced by the calibrated and
uncalibrated models. Calibration and validation of uncommon intersection designs was the
focus of an article authored by Boone and Hummer [9]. The three types of intersections that
were modeled were the roundabout, continuous green T-intersection, and the median U-turn
intersection Separate field data was collected for both calibration and validation. This data
was compared to the calibration parameters and model output to determine what calibration
steps were required, as well as which model or model configuration performed best.
Calibration was achieved by adjusting the model parameters to field measured values. The
validation involved comparing the model output MOEs to field measured values. The
calibration data used to complete this research was circulation speed in the roundabouts,
saturation flow rate data, and gap acceptance data. The validation data was traffic volumes,
travel times, stopped delay, time in queue, and the number of stops. The validation was
4


based on one-hour and half-hour time intervals depending on the site with no validation
being performed for the continuous green T-intersection.
The benefits of bus preemption were investigated in a study using NETSIM [10]. The first
task was to investigate the validity of using NETSIM to model a bus preemption situation.
NETSIM was found to be valid, so more detailed calibration and validation was performed.
This was done by adjusting various model parameters until a good match occurred between
field measured and simulated MOEs. The model parameters that were adjusted to achieve
calibration were not discussed in detail. The MOEs used for the validation were PM peak
hour values of average queue length, maximum queue length, and approach delay.
Simulation models are often used to produce data for other research. This is because field
data collection efforts can be very expensive and situations beyond the control of the data
collection team can taint the data (i.e. an incident). Sisiopiku [11] used NETSIM to produce
data so the relationship between arterial travel time and loop detector data could be
investigated. Field data was collected so calibration and validation could be completed. In
this article calibration is defined differently because it included both input data as well as
other model parameters. In depth calibration was not performed and most of the calibration
parameters were given assumed values due to the lack of field data. These calibration
parameters were free-flow speed, driver behavior, and traffic composition. The validation
output produced by the model was data for a 15-minute period that was used to create flow
versus detector occupancy, travel time versus flow, and travel time versus detector
occupancy graphs. Although the validation MOEs were used to compare the field measured
values to the model MOEs there was no iterative changes made to further calibrate and
validate the model. The simulated output was compared to field data in the previously
mentioned graphs and although differences in absolute values were apparent the general
trends experienced in the field were similar to those produced by NETSIM. This shows that
the differences between model output and field data are primarily due to calibration. This
previous point helps stress the importance of calibration if data from a traffic model is to be
used to develop empirical relationships.
Wong [12] used NETSIM to estimate the capacity and level of service in a similar fashion as
the Highway Capacity Manual procedures. The article describes the five types of parameters
that may be adjusted to calibrate the model: 1) Parameters which relate to vehicle and driver
characteristics. 2) Parameters that relate to the probabilities of drivers reactions. 3)
Parameters which relate to the weights to be applied to the mean values listed in the first
parameter type. 4) Parameters which relate to the generation of probabilities used in the
second parameter type. 5) Parameters which relate to the generation of driver types to be
used in parameter type three. The parameters that were adjusted to calibrate this model
were the mean queue discharge headway, probability of left-turn jumpers, delays due to
pedestrians, and the duration of strong pedestrian interaction. The calibration process was
defined as changes made to parameters that make the model better reflect local conditions.
The author suggests that the numerous parameters available for calibration make it
adaptable to a variety traffic conditions. Validation was not focused on in this article. This
article also gives the formula that was used in this thesis to determine the number of model
runs required to meet the desired level of confidence in the output. The formula given in this
article was taken from a document titled Statistical Guidelines for Simulation Experiments
[13]. This procedure will be discussed in detail in the methodology section of this thesis.
5


2.2 Rainer Wiedemanns Theory
The traffic modeling foundation of VISSIM has been developed primarily by Rainer
Wiedemann at the University of Karlsruhe, Germany. This work began in the early 1970s.
The traffic model is a psychophysical driver behavior model, in that at different vehicle
headways the following driver behavior varies. This model uses driver perception thresholds
to define the different car following behavior. The various car following situations are
uninfluenced driving, closing process, following process, and emergency situation. For the
most part, prior to Wiedemanns theory, car following equations only provided a deterministic
description of traffic flow [14].
Todosiev developed a model that used psychophysical spacing in the early 1960s. These
types of models have also been classified as multiregime car following models [15].
Wiedemann used some of the measurements made by Todosiev to calibrate his model when
it was initially developed [16]. The University of Karlsruhe has been involved in the
development of several traffic models over the years. An article by Leutzbach and
Wiedemann summarize some of the models that have been developed at the university [17].
Rainer Wiedemann authored an article that discussed the development of the ICARUS
model [18]. ICARUS is a microscopic model being developed to study the effects of RTI
measures on traffic flow. This article covers some of the theory that is incorporated into the
VISSIM model. PELOPS is another simulation program that has been developed. An article
that discusses PELOPS also addresses Wiedemanns theory in some detail [19].
2.3 VISSIM
The VISSIM traffic model is a German developed model, so many of the articles are in other
languages. Articles that cover research performed in the United States are relatively
uncommon because the model was first used in the United States in 1995 [20], So
compared to other simulation models such as CORSIM, developed in the United States,
there have only been a few studies that focus on VISSIM.
General information that explains the VISSIM model is available in a few documents.
Fellendorf [21] authored an article that covers the general modeling capabilities and
methodology that VISSIM employs. The VISSIM Version 3.5 Users Manual also extensively
covers the model [22],
VISSIM is a microscopic model which means each vehicle is modeled as a distinct entity.
VISSIM can be coded using both the windows based application or the input file directly. It
also has an advanced animated display to view the simulation visually. This animation has
the ability to be displayed in 3-D and also allows for different models of vehicles to be
displayed (i.e., corvette, pickup trucks, etc.).
VISSIM is modeled by coding links and connectors rather than links and nodes. Connectors
take the place of nodes in the VISSIM model. VISSIM can also have an image imported as a
background over which the network can be easily coded. In this model a figure was created
to scale in Microstation, converted to the bitmap file format, and imported into VISSIM. This
is very convenient because a lot of the data required by the model is already on the imported
figure.
6


One area where VISSIM is very advanced is in signal operations. VISSIM can model both
fixed time and actuated signals. A traffic signal head is placed for each lane in VISSIM
rather than on each approach. This allows for odd traffic signal phasing plans and provides
the modeler with more flexibility. There are two separate programs in VISSIM, the traffic
simulator and the signal state generator. The traffic simulator includes the car following and
lane changing logic, where the signal state generator determines the status of the traffic
signal displays. Each time step the signal state generator is called to see if a change in
signal indications is required. Vehicle Actuated Programming (VAP) is an add-on option that
allows for the coding of actuated traffic signals. Although it was not used to complete this
research, this module allows for complex traffic signalization schemes, such as signal priority
to be implemented.
The various available calibration parameters in VISSIM are of particular interest for this
thesis. Table 2.1 gives a list of some of these parameters and their location in the model.
These calibration parameters affect the model in a variety of ways. Some affect all inputted
traffic, others affect the model globally, and others effect traffic at a particular location. The
variety of parameters and their different effects on the model give the modeler many options
that can be used to calibrate the model. This must be kept in mind when looking at the
validation data. For example, if the model validation values are low at all locations a global
parameter may be the best way to calibrate the model.
There are several output files that can be created by VISSIM. Only the output files that are
toggled are created and most output files can be configured to report only certain data for
desired time intervals. All the files are output in a text semicolon delimited format which
saves a considerable amount of time during the data reduction and summarization process.
Since a separate text file is created for each type of output requested, the modeler avoids
having to sift through huge output files to find the desired output data.
VISSIM was the model used in two articles solely related to calibration and validation. Hoyer
and Fellendorf [23] looked at a process that uses image processing to collect field data. This
data is later used in an automated parameterization process that calibrates the model based
on validation data using a computerized process. This article defines calibration parameters
as those that can not be measured directly in the field, but rather are adjusted using
validation MOEs. Validation values are those MOEs that can be directly measured in the
field and do not have equivalent calibration parameters. This is a very basic iterative
procedure at this time, but may be a method that is used in the future to reduce the time
required to calibrate and validate traffic models. An example was included but the actual
parameters adjusted to achieve calibration during the automated process were not covered
in depth. Another article written by Fellendorf and Vortisch [24] validated VISSIM at two
different study sites. The purpose of the article is to show that VISSIM produces valid results
for two very different generically defined study sites rather than focusing on detailed
calibration and validation for a specific site. The article shows results of flow versus speed
diagrams and lane use diagrams for both sites. No time period was defined for the flow and
speed data but the lane use data was based on five-minute intervals. These sites were the
German Autobahn and a freeway in the United States. The results showed that VISSIM
produced MOEs that were very similar to the field data.
7


Calibration Parameter Location in VISSIM
Generate Exact Number Vehicles Vehicle Input
Lane Change Link Connection
Emergency Stop Link Connection
Desired Speed Distribution Desired Speed Decision
Minimum Gap Time Priority Rules
Minimum Headway Priority Rules
Maximum Speed Priority Rules
Configuration of Interrupting Section Priority Rules
Dwell Time Distribution Stop Sign/RTOR
Time Offset Transit Line
Desired Speed Distribution Transit Line
Slack Time Fraction Transit Line
Dwell Time Distribution Transit Line
Maximum Acceleration Vehicle Type
Desired Acceleration Vehicle Type
Maximum Deceleration Vehicle Type
Desired Deceleration Vehicle Type
Weight Distribution Vehicle Type
Power Distribution Vehicle Type
Desired Speed Distribution Traffic Composition
Time Steps Per Simulation Second Simulation Parameters
Average Standstill Distance Driving Behavior Parameters
Maximum Deceleration Driving Behavior Parameters
Additive Part of Desired Safety Driving Behavior Parameters
Multiple Part of Desired Safety Driving Behavior Parameters
Waiting Time Before Diffusion Driving Behavior Parameters
Minimum Headway (Front/Back) Driving Behavior Parameters
To Slower Lane if Collision Time Above Driving Behavior Parameters
Probability Factor Alpha Driving Behavior Parameters
Probability Factor Beta 1 Driving Behavior Parameters
Probability Factor Beta 2 Driving Behavior Parameters
Number of Observed Preceding Vehicles Driving Behavior Parameters
Table 2.1: Possible Calibration Parameters in VISSIM
8


Another area of research of considerable importance is the comparison of VISSIM to other
traffic simulation packages. One study compared VISSIM, CORSIM, and TRANSYT-7F [25].
The simulations were used to represent LRT for a large traffic network in Dallas, Texas. The
main purpose was to prove that VISSIM modeled automobile traffic well. If this proved to be
successful VISSIM would be used to complete a study that will investigate future LRT
extensions in the Dallas area. This article defined calibration and validation as an iterative
process. Some of the parameters used to calibrate the model were saturation flow rate,
desired speeds, yield priorities, turning decision points, lane usage, and vehicle lengths.
Travel times, queue delay, total delay, and train stops were used to compare model and field
results. The travel time and delay data was based on the peak hours and the train stops
were based on a two-hour peak period. The results showed that both VISSIM and CORSIM
produced similar results. Bloomberg and Dale have written two articles that compare
CORSIM and VISSIM [26; 27]. One article reported a study that was completed in Seattle,
Washington that involved using both models to help investigate possible future changes in
the roadway network near the Kingdome and Safeco Field. Calibration and validation was
only briefly discussed and extensive details were not given. The models were build using
both software packages and were completed by two separate teams. Output MOEs were
compared for the 2020 PM peak period with a baseball game scheduled. Route and system
travel time as well as throughput volumes from both models were used to compare results of
the various future scenarios. The study established that both models produced similar travel
time results and using two models helps increase the level of confidence in the results. The
other article did a more direct comparison of VISSIM and CORSIM. This article did not focus
on calibration and validation. Throughput, intersection level-of-service, and travel time data
were investigated using both models. The time interval of the output MOEs was not
discussed. This article points out that although there were some differences when the output
was compared, the majority of the results were similar in nature.
The other articles that look at VISSIM do not fit clearly into the previously discussed topics.
Anderson used VISSIM to produce data that could be used to help fit a queuing model
relationship [28], VISSIM was also used to investigate the effects of various types of network
signal control on various performance measures [29]. This study was completed in Dublin
and involved LRT. It looked at the effects of the various types of signal control on both
automobiles and LRT. VISSIM was used in another study to look at the effect of four future
alternatives involving both transit and roadway changes in Shoreline, Washington [30], Note
wrote an article that summarized the results of a study that focused on various guided
busway alternatives in Oregon [31]. This was a before-and-after study that looked at the
effect of adding a guided busway in the median of an existing arterial.
9


3. Study Site and Data Collection
3.1 Study Site
The study site is located in and near the CBD of Denver, Colorado. It runs from the
southwest to the northeast. Stout Street runs the length of the study area from the southwest
to the northeast. The length of Stout Street that is modeled in this study is approximately
2400 feet. Fourteenth and 15th Streets cross Stout Street on the northeast end of the study
area. The lengths of 14th Street and 15th Street modeled are about 1000 feet. These lengths
help put in perspective the size of the network that was modeled. The roadway network
within the study area is composed primarily of one-way arterials, which provide access to
several key trip generators located near or in the study area. The site includes several transit
modes including LRT and several heavily utilized bus routes. Figures 3.1 and 3.2 show the
roadway orientation, traffic-flow direction, adjacent land uses, and key access locations.
There are several types of adjacent land uses in or near the study area. A considerably
portion of the trip generators are parking lots or structures. The parking lots located toward
16th Street are used most often by commuters. The lots farther from 16th Street toward
Speer Boulevard are used by both commuters and those attending activities held at Currigan
Hall, the Colorado Convention Center, or the Denver Performing Arts Complex. The
buildings between 15th Street and 16th Street are businesses, lofts, and offices. The
buildings between 15th Street and 14th Street are generally occupied by small businesses.
The buildings to the northwest and southeast of Stout Street, between 14th Street and Speer
Boulevard, are Currigan Hall and the Colorado Convention Center, respectively. These are
both large conference centers. A mid-rise office building occupies the southeast corner of
the intersection at northbound Speer Boulevard and Stout Street. The only other notable
adjacent property is the Auraria Campus which includes the University of Colorado at
Denver, Metropolitan State College, and the Community College of Denver. The Auraria
campus is located on the southwest corner of the study site.
The north LRT line is a two-mile segment that runs from a residential area in north Denver to
downtown Denver. The south LRT line is a 12-mile segment that continues from downtown
to the south along Sante Fe Drive. This LRT line primarily transports commuters to and from
the CBD during the peak periods. Light-rail ridership is much higher in the peak direction of
travel. The LRT lines run the length of the study area from the Auraria station in the
southwest to 15th Street in the northeast.
Bus routes also run along most of the roads within the study area. The majority of these
stops serve only a couple of bus routes, but there are several heavily utilized stops spaced at
one-block intervals along 15th Street. The proximity of these bus stops to the CBD, the 16th
Street mall shuttle, and several LRT transit stops make this a heavily utilized area by
automobiles, transit, and pedestrians.
The study area can be divided into three distinct segments based on the roads and land uses
on the site. The southwest portion of the site is on the edge of the CBD and has few access
points. It is composed of major arterials designed to get high volumes of traffic to and from
10


Figure 3.1: Southwest Segment General Site Information


Figure 3.2: Northeast Segment General Site Information


the CBD. Stout Street is a one-way arterial that runs to the northeast the entire length of the
study area. Speer Boulevard runs along the southwest edge of the CBD. Speer Boulevard
includes two one-way arterials divided by Cherry Creek. It runs on an angled alignment from
the northwest to the southeast.
The central portion of the study area is a long segment of Stout Street that runs between
Speer Boulevard and 14th Street. Stout Street runs almost uninterrupted from Speer
Boulevard to 14th Street because there are very few access points and only one signalized
intersection along this segment at 13th Street. Stout Street is intersected by 13th Street near
the midpoint of the study area. Thirteenth Street is a one-way local street that provides
access from the parking facilities located under Currigan Hall and the Performing Arts
Complex.
The northeast end of the study area includes one block of streets located in the CBD of
Denver. This block includes Stout, California, 14th, and 15th Streets. California and Stout
Streets are both two-lane one-way arterials. Both 14th and 15th Streets are major arterials
that provide accesses to various parts of the CBD including Lower Downtown and Colfax
Avenue.
The study site includes several elements that are not often represented in a single traffic
model. This particular site is well suited to evaluate various components of the simulation
model due to its urban location as well as the several modes of transit employed within this
site.
3.2 Data Collection
Field data was collected for three separate purposes during the modeling process. The first
type of data is model input. This data is used to code the network and enter other required
inputs. The next type of data is used to calibrate the traffic model. Calibration data tends to
be more microscopic in nature. The final type of data is used to validate the model. This
data tends to be macroscopic in nature and is usually similar to the output produced by the
model.
There were several methods used to the collect the required data needed to complete this
study. Field visits are very important because general notes can be made that will later help
assess what types of data will need to be collected using other methods. GPS (geographic
positioning system) was used to collect static survey data. The information collected from
the field visits and the survey were used to design the traffic data collection effort. These
initial steps help assure that the proper traffic data required during the modeling process is
collected. The traffic data is included as a separate category because it involves data that is
very stochastic in nature. The other type of data used was output produced by VISSIM.
Table 3.1 lists some of data that was collected for this research. Data collected during the
field visits, the GPS survey, the traffic data collection effort, and from VISSIM output will be
discussed in separate sections.
13


DATA COLLECTED
Intersection Geometries
Bus Stop Information
LRT Station Information
Traffic Volume Data
Heavy Vehicle Percentages
Transit Bus Volumes
LRT Operation Data
Automobile and LRT Travel Time Data
Signal Timings and Offsets
Speed Limit Data
Access and Surrounding Land Use Data
Locations of Exclusive Pedestrian Phasing
Miscellaneous Field Notes
GPS Data
Construction Zone Layout
Table 3.1: List of Data Collected
3.2.1 Field Visits
Several field visits were completed in order to get a sense of particular features that required
special attention. General notes were taken regarding the AM and PM peak traffic
operations. These visits were used to determine the locations where traffic related data was
required. Field visits were also used to collect data that was missed during the GPS survey
or when other questions arose.
3.2.2 GPS Survey Data
One of the primary objectives of data collection was to collect intersection geometries and
other general site data that could be done using basic survey methods. A GPS unit was
used to collect the survey data. This data included items such as stop-bar locations, turn-
lane configurations, access locations, bus and LRT related data, speed limits, etc.
It was important to get the appropriate data into figure form so that future data collection
needs could be assessed. The GPS data was used as the base mapping in Microstation so
that figures of the study area could be created. Figures 3.3 and 3.4 show the pertinent
geometric and other site data for the study area. This Microstation figure was exported in
bitmap form and imported into VISSIM. The imported image in VISSIM provides the
background over which the network is coded. This feature is very helpful because most of
the geometric data does not have to be retrieved from other sources because a significant
quantity of the pertinent data can be included in the figure.
14


15
MATCHLINE


16


3.2.3 Traffic Data Collection
The data described so far involved static data related to the study area. The previous field
visits and GPS data helped determine what traffic related data had to be collected in the
field. The data collected during this phase of the data collection effort involved the field
measurement of several traffic MOEs that vary day to day. All the traffic data was collect on
August 30th, 2000 from 4:00 PM to 6:00 PM. The traffic data collected can be classified into
three categories: traffic volumes using counters, video data, and manually collected data
using preprinted forms. These three types of traffic data will be discussed in their own
respective sections.
3.2.3.1 Traffic Volumes
The traffic volume data were collected at all but one of the intersections within the study
area. A tube counter was also set to obtain traffic volumes near the LRT crossing at
Kalamath Street. The data was collected at five-minute intervals.
The raw traffic volume data was balanced to eliminate any problems with the counts and to
estimate the traffic volumes at accesses. This is a very important step when working with
traffic volume data because it can help determine if there are any problems with the data that
was collected. Figures 3.5 and 3.6 show the balanced peak hour traffic volumes.
3.2.3.2 Video Data
During the same time period that traffic volume data was collected, video data was being
collected at four locations. This would provide a recorded source of data that would be
available later so other types of data could be obtained. The primary purpose of the video
data was to determine vehicle travel times by tracing vehicles from one camera location to
the next camera location. The video data was also used to determine bus volumes and
heavy-vehicle percentages.
3.2.3.3 Manual Data Collection
The rest of the required data was collected manually during the same time period using
preprinted forms. Preprinted forms helped students collecting the data.
There were several types of data collected using these forms. The traffic signal timings and
offset information was collected using a stopwatch. All of the signals in this part of Denver
are pretimed so this made the data collection process easy. The other type of data collected
using the forms was related to the LRT operation. The times that the LRT entered and exited
the network were recorded. The LRT dwell times as well as the number of passengers
boarding and exiting at the LRT stations were also recorded.
17


18
MATCHLINE


19


3.2.4 VISSIM Output Data
The final type of data was output data from the microscopic model VISSIM. After each
model run a variety of output files may be created depending on the purpose of the model.
Several output files were used to determine the results of the study and during the calibration
and validation process. These output files are text files in a semicolon delimited format. This
allows them to easily be converted into Excel format for future analysis. There were four
types of output files used in this study: travel time, queue length, delay, and volume.
20


4. Methodology
There were several areas that needed to be addressed to develop the methodology for this
research. This section covers the following topics in individual subsections: discussion of
terms, development of a model to represent existing or current conditions and future
scenarios, additional visual basic tools to perform model runs, statistical analysis, and post-
processing of the model output.
4.1 Terminology
The definition of validate and calibrate are given in Websters Dictionary [32] as:
Calibrate -1. a. to set or check the graduation of (a quantitative measuring
instrument), b. to mark (a thermometer or other instrument) with indexes of degree or
quantity. 2. to determine the correct range for (a gun, mortar, etc.) by observing
where the fired projectile hits.
Validate to make valid; substantiate; confirm.
It is helpful to keep these definitions in mind as these terms and their relation to traffic
modeling is explained.
Calibration is the process of adjusting various modeling parameters, to replicate field
conditions. Calibration is usually completed using one or a combination of the following three
methods: 1). The calibration parameter values are measured at the study site or
measurements from a similar site are input into the model. 2) The calibration parameters are
adjusted using a trial and error process. 3). Additional detail or modeling features can be
added to the model if it appears that the addition of these features will provide more accurate
results. The parameters adjusted during the calibration process should only be changed to
reasonable values.
Validation is the process of comparing model output to field measured data. This step is
used to determine if the traffic model is performing well. If approach delays were collected in
the field the same approach delays would need to be output by the model. The values of the
field data and the model output would then be compared. If there are considerable
differences then the model must be calibrated.
Figure 4.1 is a flow chart that shows the calibration and validation process as well as an
initial process that may be used to check for coding errors. The first step is to ensure that a
relatively error free model has been built. Since microscopic models require a lot of data to
represent a network, this check for errors is very important. A macroscopic traffic MOE may
be used. The field and model estimated MOE may then be compared. Major differences
between the field and model MOE suggest errors in the input data coding. If an error is
found it should be corrected and the model should be run again. The field and model MOE
should then be compared again. This iterative process may be continued until reasonable
agreement between field and model MOE is achieved.
21


No
Figure 4.1: Flow Chart of the Typical Calibration and Validation Process
22


Separate data should be collected for both the calibration and validation steps. Calibration
parameters tend to be microscopic. A calibration parameter may be site specific. Typical
calibration parameters include start-up lost time, gap-acceptance factors, distributions
regarding driver type, and saturation flow rates. Validation data can be characterized as
macroscopic MOEs such as delay, queue length, and travel time.
Calibration and validation can be thought of as an iterative process. The validation
measured from the model and the field is compared after every change made during
calibration. This process of changing calibration parameters, running the model, and
comparing validation results may be continued until reasonable agreement between the
model and field data is observed. After one parameter is calibrated significant improvement
in performance may be experienced for one MOE but not for all.
The traffic animation module of a simulation model can be a very important tool both to
uncover coding errors and for calibration and validation. The animation should be viewed
initially to ensure that the conditions modeled seem reasonable. Viewing the animation is a
good way to see if the volume, signal timings and phases, and roadway geometry appear to
be coded correctly. During calibration and validation the animation may help the modeler
determine what parameters need to be calibrated to better replicate local conditions.
Calibration and validation are generally used for two different purposes when traffic modeling
is discussed. When a traffic model is first developed, generally limited calibration is
performed to set the range of appropriate default parameters to ensure realistic modeling of
driver behavior. Later calibration and validation is performed as the model is applied to
represent a specific network. This is performed to ensure site specific characteristics are
modeled. Since VISSIM is an established traffic simulation model the second type of
calibration and validation is the focus of this research.
The goal of calibration and validation depends on the application of the model. A large-scale
microscopic model to be used for planning purposes is usually run based on default
parameters to ensure field volumes are represented. If an arterial composed of three
intersections is being investigated to determine the effect of adding an additional signal the
model should be validated in more detail. In this case travel times, delay, or queue length
may be used as the validation MOEs. The effect of various levels of calibration and
validation on a before-and-after traffic study is the focus of this thesis.
4.2 Model Development
One common use of traffic simulation models is to perform before-and-after studies. In this
application a traffic model for existing conditions is developed, calibrated, and validated.
Then the existing calibrated model is adjusted to reflect several future scenarios, to
determine the best alternative. Often a future no-action model is built to reflect the existing
roadway network with only adjustments to the volumes to reflect growth. Planned
improvements such as signal timings and improvements, that are not part of the project
being modeled, are also coded. This model is usually the worst case scenario and is
compared to the various build alternatives. A build alternative will often involve
improvements such as an additional traffic lane, additional turn pockets, or more advanced
signalization phasing.
23


Since the purpose of this thesis is to evaluate the level of effort required to calibrate and
validate a model, a different approach is taken. The model representing existing conditions
is calibrated and validated to reflect six levels of detail. Figure 4.2 shows a flow chart that
helps explain the various models calibrated to different levels of detail. The calibration and
validation process undertaken and the future scenarios are presented in the next few
sections.
4.2.1 Model for Existing Conditions
4.2.1.1 Level 1
The most basic model representing existing conditions will be referred to as the Level 1
model. This model includes minimal model coding and calibration. This model may be used
to obtain macroscopic results used for a planning level analysis or may result from someone
who is inexperienced with microscopic modeling.
To build a microscopic simulation model several types of input data need to be coded. The
Level 1 model represents the roadway network geometry, volumes on all entering links on
the perimeter of the network, the turning percentages at all of the intersections, and the
signal timing data. The other coding is more specific in nature and controls items such as
transit routes, data collection types and locations, and movement priority at merge and
access points.
This model can best be described by what is not included. Fifteenth Street is a heavily
utilized transit route for bus service. If an area being modeled only has very limited bus
service often these routes are not coded. Even though Fifteenth Street has very heavy bus
traffic, the bus routes were not coded in the Level 1 model. Another part of the network that
was not coded was the source nodes at some access points on Stout Street between 14th
and 15th Streets. Often if the existing data shows only a small source or sink between two
traffic count locations these sources or sinks are not included. Finally, a construction zone
that closed two lanes on Fifteenth Street was not coded in this level.
4.2.1.2 Level 2
The next level of model detail focused on adding the source nodes, construction zone, and
bus routes not included in the Level 1 model. In addition the LRT offsets were modified.
When the LRT travel time data was collected in the field there was not any information as to
what part of the signal phasing the travel time data was referenced. Since the cycle length
was 60 seconds LRT start times had to be adjusted using a trial and error process by 1 to 59
seconds to determine if other start times better replicated field conditions. The northbound
LRT start times were adjusted first and the best results occurred when the start times were
reduced by 40 to 50 seconds. Then both the northbound and southbound LRT start times
were adjusted by minus 40, 45, and 50 seconds. The best results occurred when both the
northbound and southbound LRT start times were adjusted by minus 45 seconds.
24


Less Accurate
More Accurate
Figure 4.2: Levels of Analysis
25


4.2.1.3 Level 3
Speed distribution was the focus of the Level 3 model. Several changes were made that
affected the desired speeds of the modeled vehicles. The desired speed of all the vehicles in
the areas of the model where lower speed limits were posted were reduced by 5 mph. The
desired speed of the northbound LRT was increased by 5 mph for the higher speed limit
section. The location of the lower speed limit for the southbound LRT was moved several
hundred feet north of the actual field location.
4.2.1.4 Level 4
This level focused on better representing the flow profile. Usually peak hour volumes are
coded. In Level 4,15-minute volumes rather than hourly volumes were represented. This
involved changing the input volumes as well as the turning percentages. This level was
added so that the benefits of coding more detailed traffic volumes could be assessed.
The standard deviation of the vehicle input volumes based on 15-minute volumes for the
entire simulation period (including the initialization period) varied considerably depending on
the input location. The southbound Speer Boulevard standard deviation was highest with a
value of 303, while the standard deviation at 13th Street was lowest with a value of 22. The
other vehicle input locations experienced the following values: California Street 78,14th
Street 68, Stout Street 49, 15th Street 42, and northbound Speer Boulevard 133.
4.2.1.5 Level 5
The Level 5 model was calibrated further by adjusting the modeling parameters that affect
the saturation flow rate. Several model runs were made with different values of the additive
part of desired safety and the multiple part of desired safety, which are the modeling
parameters that affect the saturation flow rate in VISSIM. The values were adjusted so that a
lower saturation flow rate was modeled. This change resulted in about a 200 vehicle/hour
decrease in the saturation flow rate. Therefore the saturation flow rate was adjusted from a
default rate of 1850 vehicles/hour to 1640 vehicles/hour.
4.2.1.6 Level 6
A spillback condition was observed under the existing conditions at the periphery of the
network. Although the condition occurred outside the network, it affected the network MOEs.
The two ways to correct this problem are to expand the model or to try to represent the
condition within the model. Expanding the model was too time consuming and difficult due to
the lack of additional data. Therefore the problem that was occurring outside the study area
was represented in the model of existing conditions using other means.
To try to replicate this condition a dummy signal was coded on Stout Street toward 16th
Street. This dummy signal was set up to try to create a spillback condition toward 14th Street
so the travel times better replicated existing conditions. The signal timings were adjusted
several times until the best travel time similarities occurred.
26


4.2.2 Future Model Development
In order to investigate the effects of calibration and validation on before-and-after studies
three future scenarios were coded. The only changes from existing conditions involved
changes in the traffic input volumes and turning percentages. This was done so that the
effects on the results could be isolated to being caused solely by volume increases and
differences in the calibration and validation. The volumes were adjusted so that an
incremental degradation in approach level of service would be achieved.
In almost all modeling exercises a Highway Capacity Manual signalized analysis is
performed. This is done for an assessment of the quality of service, as well as to determine
the traffic signal timings for future scenarios. Signal timings are determined since simulation
models do not yet have the ability to optimize signal operations. Signal 2000 was used to
complete the Highway Capacity Manual analysis in this research.
Three future traffic models were built with increasing volume levels on most approaches.
The future scenario traffic volumes were determine using Signal 2000. Before the future
volumes were determined the existing condition Signal 2000 files had to be calibrated. The
first step required to determine the future volumes was to enter the existing conditions into
Signal 2000. Then the approach delays from the VISSIM model representing existing
conditions were compared to the approach delays output from Signal 2000. The existing
Signal 2000 approach delays were calibrated so that the delays produced by Signal 2000
matched the delays from VISSIM. The Signal 2000 files required calibration because Signal
2000 only looks at isolated intersections and the results from VISSIM reflect other factors on
a network and arterial basis.
After the existing Signal 2000 models were calibrated the future volumes for the three
scenarios had to be determined. These volumes were determined by first considering the
approaches that most restricted arterial traffic flow. The volumes at these approaches were
adjusted by trial and error in Signal 2000 so that the approach experienced significant
increases in delay for each future model scenario. The volumes determined at these
restricted traffic flow locations were then distributed to the other intersection approaches in
the network. Table 4.1 shows the existing and future Signal 2000 approach delays at several
of the modeled approaches. The three future scenarios are titled Future Model A, Future
Model B, and Future Model C. The future models each have higher traffic volumes than
existing conditions with Future Model A having the smallest volume increases, Future Model
C having the largest volume increases, and Future Model B having volume increases
between Future Model A and Future Model C.
The future models were built simply by taking the existing calibrated models and changing
the volume related inputs. There was no need for any further calibration or validation. Figure
4.3 is a flow chart that explains the relation of the models that reflect existing conditions to
the future models.
27


Approach Ex sting Future A Future B Future C
LOS Delay * LOS Delay * LOS Delay * LOS Delay *
Stout Approach to 14th B 12 D 42 E 65 F 82
Stout Approach to Southbound Speer C 21 C 27 C 34 D 46
14th Approach to Stout B 20 D 41 E 59 F 90
15th Approach to Stout D 38 E 64 E 75 F 96
14th Approach to California A 6 B 13 B 19 C 31
15th Approach to California C 23 C 26 C 28 C 34
Stout Approach to 15th C 24 D 40 E 68 F 94
California Approach to 14th D 39 E 56 E 78 F 115
California Approach to 15th C 27 C 33 D 40 E 61
Delay in Seconds/Vehicle
Table 4.1: Signal 2000 Approach Delays and Level of Service
28


Figure 4.3: Schematic of Existing Models Relation to the Future Models
29


4.3 Visual Basic Tools
Performing model simulations and working with output data can be a daunting task. There
were several small Microsoft Visual Basic programs written and incorporated into a Microsoft
Access form to help simplify this process. These programs used basic loop structures and
send key commands to complete the desired task. The send key command essentially
means the program types the desired keys in the proper order to complete the task.
Although this is not the most advanced programming method it worked relatively well. Figure
4.4 is a copy of the Access form in which these tools are incorporated. Each of the tools
available in the Access form will be described briefly in the following sections.
VISSIM REPLICATION TOOLS
RUN BATCH OPERATIONS =
i
tEkT'TO EkCELXOiyyERtER :
. i
REPLICATION FILE CREATOR'.
1 REPLICATIONS,
EXCEL
Figure 4.4: Access Form for Visual Basic Tools
4.3.1 VISSIM Batch Operations Module
The batch operations module prompts the user for the VISSIM input file locations, the
number of files that need to be run, and the estimated time each run will take. The program
then opens each input file in VISSIM and runs the simulation. After the VISSIM run is
completed the program waits until the estimated run time is reached and closes the file. This
process continues until the inputted number of model runs are completed.
4.3.2 Text to Excel File Format Converter
This module asks how many text files need to be converted from the text semicolon delimited
format to an Microsoft Excel worksheet. The location of the output text files is coded into the
programming text. This can be changed with relatively little effort. Each of the output text
files is opened in semicolon delimited format in Excel. The file is then saved in the Excel file
format. The location of the output files in Excel format is also already incorporated into the
program.
30


4.3.3 Simulation File Creator
This option is used to create multiple files with different random number seeds so that
multiple simulations can be performed. The program asks for the full file path for the current
input file. The user is also prompted to enter the number of files that need to be completed.
The initial file is opened in Microsoft WordPad and the random number seed is changed to a
different random number seed. The program has 50 possible random number seeds that
were determined using the RANDBETWEEN function in Excel. The file is than saved with a
different random number seed. This process continues, with a different random number
seed being input, until the desired number of input files have been created. The benefit of
this program is that it removes the opportunity for error. Another benefit is that if the same
number of simulation files are made for each model scenario the same random numbers will
be input into all the simulations for each model scenario.
4.4 Statistical Analysis
One area where statistical analysis is often required when performing a before-and-after
study is to determine the number of model runs that need to be completed in order to
produce output within a certain confidence level. The formula below was used by Wong [12]
and is from the Statistical Guidelines for Simulation Experiments, Volume 3 [13].
X=[(T1.a/2,n.1)(S)/E]2
(4.1)
where,
X = required number of runs
T = critical value for students t-distribution with (1-a)100% level of confidence and
(n-1) degrees of freedom
a = coefficient of confidence
n = number of samples from which S is computed
E = tolerable error
S = standard deviation
Equation 4.1 was used to determine how many simulations were required. Each model
scenario was initially run 20 times. This was done using the same set of random numbers
for each model scenario. After the first 20 runs were completed the output data was
averaged for each scenario and equation 4.1 was used to determine if more runs were
required. The results showed that more runs were necessary so each model scenario was
run an additional 30 times, which resulted in averaged data based on 50 runs.
Another step that must be completed in order to reduce the variance in the output is that
each model run should have an initialization period where the model is run before
summarizing any output MOEs. This period allows the vehicles to enter the network. A
general rule is that the initialization period should at least be as long as it would take for one
vehicle to traverse the length of the network. The initialization period used for all model runs
was 30 minutes.
31


4.5 Preparation of Output
After the model runs were completed and the text files had been converted into Excel format,
all of the related data had to be put into a format where it could be easily summarized. The
output that was summarized was travel time and delay. For each modeled scenario there
were 50 output files for each type of data. For each type of output data a new excel
workbook was created and each of the 50 output sheets was copied into the workbook. This
allowed for all of the output data for each type of output and each modeled scenario to be
used later for analysis. With all of the data in one location all of the pertinent data for the fifty
runs was averaged. The averaged data is what was required in order for the analyses to be
completed.
32


5. Results
The results of the analysis are covered in several sections. The first section presents the
results from the six models coded with different levels of detail and calibration, by comparing
the MOEs of the model representing existing conditions to the field measured MOEs for
selected arterial segments. The next section summarizes the MOEs for additional arterial
segments. The third section examines the effects of model detail on the future scenarios.
5.1 Models Representing Existing Conditions
There were three travel time segments where field data was collected for the automobile
traffic. These three segments are: Stout Street from northbound Speer Boulevard to 14th
Street, Stout Street from 14th Street to 15th Street, and 14th Street from Stout Street to
California Street. Figures 5.1, 5.2, and 5.3 show the percent difference for the three
segments between the simulated and the field travel times for each of the six models
representing existing conditions.
There are several points that can be drawn from Figure 5.1. This figure reports the
automobile travel times on Stout from northbound Speer Boulevard to 14th Street. It shows
that the existing Level 1 and existing Level 2 results were very similar. This is because the
addition of the construction zone, accesses, and bus routes did not affect this area of the
network. The most significant change occurred between Level 2 and 3 where there was a
change in the desired speeds. This is more significant on this segment due to the longer
length of the segment. The results from Levels 3 through 6 are very similar. The overall
trend shows that the Level 1 and Level 2 models underestimated travel time by 20 percent.
The Level 3 through 6 results are very similar but are about five percent higher than field
travel times.
The results presented in Figure 5.2 are very similar for each level of detail. This figure
reports the right-turning automobile travel times from northbound Stout Street to California
Street on 14" Street. As expected, most of the changes in the various levels do not have a
significant effect on this travel time segment because most of the significant changes occur
at other areas of the network.
Figure 5.3 summarizes the automobile travel times on Stout Street from 14th Street to 15th
Street. This figure shows minor differences between the levels. It is apparent that although
the dummy signal was added, the Level 6 model did not fully capture the actual field
conditions. Although the addition of the dummy signal did result in an improvement of about
five percent when compared to Level 1. Level 6 also performed more realistic toward the
end of the simulation. The beginning and ending times both produced very similar results to
the field data.
When comparing the differences between the field and model data, the percent difference
generally decreases as the model detail and calibration effort increases. The Stout Street
segment between northbound Speer Boulevard and 14th Street showed the most
improvement due to calibration and model detail. The segment travel times are about five
33


percent higher than field measured values for the entire hour for Levels 3-6. Levels 1 and 2
for the same segment resulted in a -20 percent difference from field conditions. The effect
of calibration and model detail on the right-turning vehicles from Stout Street to the
intersection of 14th and California Streets was not as significant. All of the model levels
performed within a five percent range of each other and the overall travel times were very
similar to the field measured values. The results for the Stout Street segment from 14th
Street to 15th Street show some improvement due to calibration and model detail. The
results for the entire hour show that Level 6 underestimates field travel times by about -20
percent where the Level 1 model underestimates field travel times by -25%.
Similar data is available to analyze the effects of model detail on the LRT travel times but
since the future models do not include any changes in the LRT volumes these results are not
included in this research.
34


Existing Conditions Automobile Traffic on Stout Street from
Northbound Speer Boulevard to 14th Street
Time (seconds)
- Level 6
- Level 5
Level 4
X- - Level 3
X- - Level 2
- Level 1
Figure 5.1: Travel Time Differences on Stout from Speer to 14th
Existing Conditions Automobile Traffic on 14th Street: Right-
Turns from Stout Street to California Street
c ~o
s
2
0>
CD

o
c

T3
C
re
o

re
E
v>
LU

XJ
o
S
Time (seconds)
- Level 6
m- - Level 5
Level 4
X- - Level 3
X- - Level 2
-Level 1
Figure 5.2: Travel Time Differences for Rights from Stout to California on 14
th
35


Existing Conditions Automobile Traffic on Stout
Street from 14th Street to 15th Street
c o
0
3 il
0 o
DO c re
0 o c XJ 0 4-*
0 re
k. 0 E
it
Q 4-1
c 0
0 o
O L. O
0 2
Q.
- Level 6
m- - Level 5
Level 4
- Level 3
*- - Level 2
-Level 1
Time (seconds)
Figure 5.3: Travel Time Differences on Stout from 14th to 15th
5.2 Examining the Five Additional Segments
In order to investigate the impact of model calibration on other parts of the network, several
additional arterial travel time segments were examined. The additional segments include:
Stout Street from the entrance of the network to northbound Speer Boulevard, Stout Street
from northbound Speer Boulevard to the exit of the network, 14th Street, 15th Street, and
California Street to the exit point of the network on 14th Street. Figures 5.4 through 5.8 show
the travel times for each of the six model levels representing existing conditions.
For all five segments, Levels 1 and 2 produced very similar results at all locations except 15th
Street. These results are reasonable because the major changes from Level 1 to Level 2
was the addition of the bus routes and construction zone on 15th Street. The changes made
from Level 1 to Level 2 resulted in roughly a 50% increase in the travel times on 15th Street.
For all of the segments except 15th Street the difference between Levels 2 and 3 are also
significant. The desired speed adjustments increased the travel times at most locations by
about 10%. This increase was about 20% for the travel time segment from Speer Boulevard
to the exit point of the network. The increase is higher on this link due to the longer length of
the travel time segment. The other travel time segments all have very similar lengths. The
reason the desired speed had little effect on 15th Street is due to the congestion experienced
on that segment.
There was very little change from Level 3 to Level 4 due to the addition of the 15-minute
traffic volumes. This modification generally resulted in a change of a several seconds in the
resulting travel times. This is the result of the standard deviations for the input volumes not
being significant enough to affect the travel times.
36


The decrease in the saturation flow rate in Level 5 also resulted in minor changes in the
travel times. The only location where significant changes occurred was on 15tH Street. This
resulted in a 5-10% increase in the travel times depending on the time period. The increase
in travel times is again due to congestion that is already occurring on 15th Street.
The addition of the dummy signal in Level 6 only significantly affected the Stout Street
approach to 15th Street. This caused the slope of the graph to change slightly and resulted in
a 5% increase in the travel times at the 3600 and 4500 time periods.
Some of the approach delays are summarized in Table 5.1. The approach delays do not
change much from Level 1 to Level 6 except at the Stout Street approach to 14,fi Street and
the 15th Street approach to Stout Street. The Stout Street approach to 14th Street
experienced an 80% increase in delay from Level 1 to Level 6. This increase in delay is
primarily due to the adjustments made to the desired speed parameters. The other location
where significant changes occurred was the 15th Street approach to Stout Street. A 130%
increase in travel times were experienced on this approach. This increase in delay is
primarily due to the addition of the bus routes and the construction zone on 15th Street.
Therefore, for these approaches the delay or the LOS is worse for higher level models.
There was one overall general trend that was experienced for all the models that represented
existing traffic conditions with few exceptions. As level of detail represented by the models
increased the travel times experienced increased as well. The degree to which model detail
effected the results was dependent on the differences in the model levels being compared,
as well as the network location on which the output data was based.
Existing Conditions Arterial Travel Time: Stout Street
from Entrance Point to Northbound Speer Boulevard
Level 6
Level 5
Level 4
* Level 3
Level 2
-Level 1
Time (seconds)
Figure 5.4: Travel Times on Stout from Entrance Point to Northbound Speer
37


Existing Conditions Arterial Travel Time: Stout Street
from Northbound Speer Boulevard to Exit Point
Level 6
* Le\^l 5
Level 4
* Level 3
Level 2
Le\l 1
Figure 5.5: Travel Times on Stout from Northbound Speer to the Exit Point
Existing Conditions Arterial Travel Time: 14th Street
- Level 6
*- - Level 5
Level 4
- Level 3
#- - Level 2
-Level 1
Time (seconds)
Figure 5.6: Travel Times on 14th Street
38


Travel Time (seconds) Travel Time (seconds)
Existing Conditions Arterial Travel Time: 15th Street
2700 3600 4500 5400
- Level 6
- Level 5
Level 4
X- - Level 3
- Level 2
- -Level 1
Time (seconds)
Figure 5.7: Travel Times on 15th Street
Existing Conditions Arterial Travel Time:
California Street to Exit Point
- - Level 6
- Level 5
Level 4
x- - Level 3
*- - Level 2
- -Level 1
2700 3600 4500 5400
Time (seconds)
Figure 5.8: Travel Times on California Street to the Exit Point
39


Approach Level 1 Level 2 Level 3 Level 4 Level 5 Level 6
LOS DELAY LOS DELAY LOS DELAY LOS DELAY LOS DELAY LOS DELAY
Stout Approach to 14th Stout Approach to Southbound Speer 14th Approach to Stout 15th Approach to Stout 14th Approach to California 15th Approach to California Stout Approach to 15th California Approach to 14th California Approach to 15th B 11 B 12 B 19 B 18 B 20 B 20
C 21 C 21 C 21 C 21 C 21 C 21
B 19 B 19 B 19 B 19 B 20 B 20
B 20 D 38 D 39 D 39 D 47 D 46
A 6 A 6 A 6 A 6 A 6 A 6
C 22 C 23 C 23 C 23 C 24 C 24
C 22 C 24 C 22 C 22 C 23 C 23
D 36 D 36 D 39 D 39 D 39 D 40
C 26 C 27 C 26 C 27 C 31 C 30
Table 5.1: Existing Approach Delays from VISSIM
40


5.3 Future Scenarios
The impact of the various levels of model detail on the future scenarios is examined in this
section. The effect of the various levels of model detail on arterial travel time and delays is
discussed in two separate sections. The final section compares the differences in commuter
costs due to different arterial travel times produced by the various levels.
5.3.1 Arterial Travel Times
The travel time for the same arterial locations discussed in the previous section is again
summarized for each of the future scenarios modeled. Each of these locations and the
results for the future models is discussed in the next several paragraphs.
The travel time results on 14th Street for all of the future scenarios are shown for each of the
six levels in Figures 5.9 through 5.11. For each of the three future scenarios the general
shape of the travel time graphs for each level were very similar. The results from Level 1 and
Level 2 are almost identical for all three future models. This is because the addition of the
access points, bus routes, and constructions zone did not occur on 14th Street. Levels 3 and
4 also produced similar results while Level 4 produced slightly longer travel times. Both
Levels 3 and 4 produced higher travel times than Levels 1 and 2. The higher travel times in
Level 3 are due to lower desired speeds and Level 4 was due to the use of 15-minute traffic
volumes. The Level 5 and 6 models also produced nearly identical results. This is because
the addition of the dummy traffic signal in the Level 6 model did not affect this segment.
Both Level 5 and 6 results were higher than all the previous levels. The differences between
the various levels of detail become more apparent as the level of congestion increases.
41


Future A Arterial Travel Time: 14th Street
(A
T3
C
o
o
o
42-
0)
E
70 60
Level 6
f 1"" 5 x |
50 ! 1 m Level 5
40 Level 4

30 x Level 3
20 - - Le\el 2
10 - -- - - - Lerel 1
0 i
2700
3600 4500
Time (seconds)
5400
Figure 5.9: Future Scenario A Travel Times on 14th Street
Future B Arterial Travel Time: 14th Street
- - Level 6
m- - Level 5
Level 4
x- - Level 3
- Level 2
- Level 1
l____________________________________________ _______________________
Figure 5.10: Future Scenario B Travel Times on 14th Street
42


Future C Arterial Travel Time: 14th Street
Level 6
Level 5
Level 4
Level 3
* Level 2
Level 1
Time (seconds)
Figure 5.11: Future Scenario C Travel Times on 14th Street
The arterial travel times on Stout Street from northbound Speer to the exit point in the
network are shown in Figures 5.12 through 5.14. Level 1 resulted in the lowest travel times
for all of the future scenarios. Level 2 shows higher travel times than Level 1 and the
differences between Level 1 and Level 2 were considerably greater for Scenario C than
Scenario A. This change is primarily due to the addition of the access points on Stout Street
between 14th and 15th Streets. The Level 3 and Level 4 models produced similar results
and generally produced slightly higher travel times than Level 2. The addition of the 15-
minute volumes in Level 4 did change the shape of the travel time line slightly from the Level
3 results in Figures 5.12, 5.13, and 5.14 when compared to the Level 3 results in the same
figures. The changes in the saturation flow rate did increase the travel times considerably for
Level 5, but the shape of the travel time results were similar to the Level 4 results. The
addition of the dummy signal had significant effects on the northern Stout Street segment.
This addition increased the travel times experienced in the Level 5 model runs by about
100% for Scenario A (Figure 5.12) and about 50% for Scenarios B (Figure 5.13) and C
(Figure 5.14) when compared to Level 5. The reason for the dramatic increase in travel
times was that the approach where the volumes were adjusted to develop the degradation in
the approach level of service was the Stout Street approach to 15th Street. The dummy
signal was actually the signal that caused the greatest restriction to traffic flow on Stout
Street.
43


Figure 5.12: Future Scenario A Travel Times on Northern Stout Segment
Future B Arterial Travel Time: Stout Street from Northbound
Speer Boulevard to Exit Point
- - Level 6
- Level 5
Level 4
K- - Level 3
Hr- - Level 2
- Level 1
2700 3600 4500 5400
Time (seconds)
Figure 5.13: Future Scenario B Travel Times on Northern Stout Segment
44


Figure 5.14: Future Scenario C Travel Times on Northern Stout Segment
Figures 5.15 through 5.17 show the future travel time results for Stout Street from the
entrance point in the network to northbound Speer Boulevard. Levels 1 through 5 produced
similar results for both future Scenarios A and B. Future Scenario C produced travel time
results that showed more significant differences between Levels 1 through 4. The changes
made between Levels 1 and 2 did produce an increase in travel time primarily toward the end
of the modeled time period for Scenario C. Levels 3 and 4 produced similar results to Level
2 with only minor differences in the travel times for Scenario C. Level 5 caused an increase
in the travel times for all future scenarios. The most significant increase occurred for future
Scenario C. This is because Scenario C experienced the greatest amount of congestion,
and the saturation flow rate has the most significant effect on traffic beginning to flow from a
stopped in-queue position. The Level 6 results show a large increase in the travel times due
to the addition of the dummy signal. This indicates that the effect of the dummy signal is
extending beyond the travel time segment from northbound Speer on Stout Street to the exit
point of the network. The reason for this large queuing effect was because the Stout Street
approach to 15th Street was the approach where the volumes were adjusted to develop the
degradation in traffic operations, but the dummy signal was actually the signal that caused
the greatest restriction to traffic flow on Stout Street. This caused traffic to queue and
spillback beyond the intersection of Stout Street and 14th Street on the Stout Street
approach.
45


Future A Arterial Travel Time: Stout Street from Entrance
Point to Northbound Speer Boulevard
...Level 6
Level 5
Level 4
Level 3
* Level 2
Level 1
Time (seconds)
Figure 5.15: Future Scenario A Travel Times on Southern Stout Segment
Future B Arterial Travel Time: Stout Street from Entrance
Point to Northbound Speer Boulevard
Level 6
-m Level 5
Level 4
Level 3
Level 2
Level 1
Time (seconds)
Figure 5.16: Future Scenario B Travel Times on Southern Stout Segment
46


Future C Arterial Travel Time: Stout Street from Entrance
Point to Northbound Speer Boulevard
- Level 6
- Level 5
Level 4
X- - Level 3
Level 2
- - Level 1
Time (seconds)
Figure 5.17: Future Scenario C Travel Times on Southern Stout Segment
The travel time results for 15th Street are shown in Figures 5.18 through 5.20. The largest
changes in travel times due to model detail occurred between Levels 1 and 2. This is
expected because 15th Street is where the bus routes and the construction zone were
added. The Level 3 results show a similar shape as Level 2 with the only difference being a
slight increase in travel times for the Level 3 results. This reflects the lowering of the desired
speed parameter. The results for Level 3 only differ slightly from the Level 2 results due to
the congestion and relatively short length of the link. Level 4 resulted in lower travel times
than Level 3 toward the beginning of the simulation period. This is because the volume at
the end of the initialization period was considerably lower than the rest of the 15-minute time
periods. Levels 5 and 6 produced the highest travel times and were virtually the same for all
future scenarios.
47


Future A Arterial Travel Time: 15th Street
Level 6
Level 5
Level 4
Level 3
Level 2
Level 1
Time (seconds)
Figure 5.18: Future Scenario A Travel Times on 15th Street
Future B Arterial Travel Time: 15th Street
Level 6
Level 5
Level 4
* Level 3
Level 2
Level 1
Time (seconds)
Figure 5.19: Future Scenario B Travel Times on 15th Street
48


Future C Arterial Travel Time: 15th Street
Level 6
Level 5
Level 4
* Level 3
* Level 2
Level 1
Time (seconds)
Figure 5.20: Future Scenario C Travel Times on 15th Street
The travel times for the future scenarios on California Street are presented in Figures 5.21
through 5.23. The Level 1 model by far produced the lowest travel times for all three future
scenarios. Level 2 caused a significant increase in the travel times on this arterial segment.
This is because the right-turning vehicles from California to 15th Street experienced
considerable congestion due to the addition of the bus routes and construction zone on 15th
Street. Level 3 only caused a slight increase in the travel times. Level 4 produced the
second lowest travel times out of all six levels with the biggest difference occurring at the
beginning of the simulation period. This is because the 15-minute volumes again were
considerably lower during the initialization period. Level 6 produced nearly identical results
as the Level 5 model. This is because the addition of the dummy signal had no measurable
effect on California Street.
49


Future A Arterial Travel Time: California
Street to Exit Point
Level 6
Level 5
Level 4
Level 3
Level 2
Level 1
Time (seconds)
Figure 5.21: Future Scenario A Travel Times on California
Future B Arterial Travel Time: California
Street to Exit Point
* Level 6
Level 5
Level 4
Level 3
* Level 2
Level 1
Time (seconds)
Figure 5.22: Future Scenario B Travel Times on California
50


Future C Arterial Travel Time: California
Street to Exit Point
- Level 6
- Level 5
Level 4
X - Level 3
- Level 2
- - Level 1
Time (seconds)
Figure 5.23: Future Scenario C Travel Times on California
5.3.2 Approach Delays
The approach delay data is shown in Tables 5.2 through 5.4 for each of the future scenarios.
The delay output shows similar results to the travel time data. These tables make it easy to
see which of the levels have the greatest effect on the approach delays. These tables also
show the corresponding level of service. It is apparent that the Signal 2000 results in Table
4.1, even when calibrated, did not always produce future approach delays with similar values
to the VISSIM results. This gives some insight as to why simulation models are used to
study areas that involve more that one or two intersections, especially when congestion is
present. The other reason these figures are included is to allow one to compare how the
intersection approaches modeled for each of the future scenarios are operating according to
the Highway Capacity Manual procedures.
The future approach delays show the significance of the additional detail added to the model.
This is most significant when comparing the existing approach delays, shown in Table 5.1,
with the future approach delays. There were only two notable changes in the existing
approach delays when comparing these delays for each of the levels of detail. The two
specific changes were discussed in Section 5.2 The future delay data shows significant
changes in delay and level of service in some cases. This shows that as growth rates used
for future analysis become larger the effects of model detail become more pronounced.
51


Scenario A
Level 1 Level 2 Level 3 Level 4 Level 5 Level 6
Approach LOS Delay LOS Delay LOS Delay LOS Delay LOS Delay LOS Delay
Stout Approach to 14th B 18 C 21 C 33 C 34 D 46 F 203
Stout Approach to Southbound Speer C 23 C 23 C 23 C 23 C 24 C 34
14th Approach to Stout C 22 C 22 C 22 C 22 C 24 C 24
15th Approach to Stout C 23 F 93 F 98 F 88 F 100 F 101
14th Approach to California A 6 A 6 A 7 A 8 A 8 A 8
15th Approach to California C 24 D 43 D 46 D 39 E 57 E 58
Stout Approach to 15th C 29 D 44 D 42 D 40 D 48 F 111
California Approach to 14th D 37 D 35 D 36 D 37 D 35 D 36
California Approach to 15th C 28 F 110 F 119 E 79 F 114 F 116
Table 5.2: Future Scenario A Approach Delay
Scenario B
Level 1 Level 2 Level 3 Level 4 Level 5 Level 6
Approach LOS Delay LOS Delay LOS Delay LOS Delay LOS Delay LOS Delay
Stout Approach to 14th C 30 E 60 F 90 E 79 F 162 F 329
Stout Approach to Southbound Speer C 24 C 24 C 24 C 26 C 35 F 97
14th Approach to Stout C 23 C 23 C 23 C 23 C 26 C 27
15th Approach to Stout C 23 F 102 F 102 F 96 F 107 F 107
14th Approach to California A 6 A 6 A 8 A 8 A 9 A 8
15th Approach to California C 26 E 71 F 80 E 58 F 97 F 87
Stout Approach to 15th D 40 E 65 E 74 E 60 E 68 F 118
California Approach to 14th D 39 C 34 C 35 D 36 C 35 C 35
California Approach to 15th C 28 F 130 F 133 F 103 F 131 F 136
Table 5.3: Future Scenario B Approach Delay
52


Scenario C
Level 1 Level 2 Level 3 Level 4 Level 5 Level 6
Approach LOS Delay LOS Delay LOS Delay LOS Delay LOS Delay LOS Delay
Stout Approach to 14th F 84 F 184 F 193 F 175 F 233 F 359
Stout Approach to Southbound Speer B 31 D 41 D 53 D 41 E 78 F 134
14th Approach to Stout C 24 C 24 C 25 C 25 C 30 C 31
15th Approach to Stout C 25 F 105 F 105 F 104 F 108 F 110
14th Approach to California A 6 A 6 A 9 A 9 B 10 A 10
15th Approach to California C 29 F 139 F 142 F 123 F 141 F 142
Stout Approach to 15th D 51 E 76 E 77 E 72 E 73 F 118
California Approach to 14th D 42 C 33 C 34 C 35 C 33 C 34
California Approach to 15th C 34 F 137 F 137 F 126 F 147 F 144
Table 5.4: Future Scenario C Approach Delay
53


5.3.3 Cost Difference Analysis
This section examines the effects of the various levels of detail on the future scenarios based
on the difference in travel costs for each scenario examined. Travel cost was estimated for
20 workday years for a 2-hour peak period assuming a $10 hourly average value of time.
The two arterial segments that experience the most significant effects on travel cost due to
the different detail levels are included in Tables 5.5 and 5.6. Table 5.5 show the cost data
for the north segment of Stout Street between northbound Speer Boulevard and the exit point
of the network. Table 5.6 shows the cost data for 15th Street. The tables are set up to show
the difference in the travel costs estimated based on the various models. The first table in
each figure presents the cost difference for the model representing existing conditions. The
second, third, and fourth tables present the travel cost differences for Scenarios A, B, and C
respectively. The rows and columns represent the comparison levels. An example using
Table 5.5 for the Stout Street travel times helps explain the use of the tables. Referring to
the second table in Table 5.5, if for future Scenario A, a Level 1 model is used instead of a
Level 6 model for a study, then the travel cost would be underestimated by $4,773,553.
The costs help show the significance of model detail and its effect on the results. If the
simulation model is going to be used to select one out of several future scenarios, in a
before-and-after traffic study, without using the results to justify the selection in terms of
some type of cost analysis than more model detail may not be beneficial. On the other hand
if the results are going to be used to justify the cost of the improvement in the form of a
benefit/cost analysis than more extensive calibration may be required.
54


Existing Level 1 Level 2 Level 3 Level 4 Level 5 Level 6
Level 1 Level 2 Level 3 Level 4 Level 5 Level 6 33334 191513 180481 228294 256933 158179 147147 194960 223599 -11031 36781 65420 47812 76452 28640
Scenario A Level 1 Level 2 Level 3 Level 4 Level 5 Level 6
Level 1 Level 2 Level 3 Level 4 Level 5 Level 6 320068 614460 596465 950762 4773553 294393 276397 630694 4453485 -17995 336302 4159093 354297 4177088 3822791
Scenario B Level 1 Level 2 Level 3 Level 4 Level 5 Level 6
Level 1 Level 2 Level 3 Level 4 Level 5 Level 6 1050229 1926156 1471974 2863574 6047143 875927 421745 1813345 4996914 -454182 937418 4120987 1391600 4575169 3183569
Scenario C Level 1 Level 2 Level 3 Level 4 Level 5 Level 6
Level 1 Level 2 Level 3 Level 4 Level 5 Level 6 2288190 2477558 2116916 2771067 5144649 189368 -171273 482878 2856459 -360642 293509 2667091 654151 3027732 2373581
All Values in Dollars
Table 5.5: Cost Differences Based on Detail Level for North Segment of Stout
55


Existing Level 1 Level 2 Level 3 Level 4 Level 5 Level 6
Level 1 Lewi 2 Level 3 Level 4 Level 5 Level 6 451759 524973 547099 673678 662611 73214.2 95340.7 221919 210852 22126.5 148705 137638 126578 115512 -11067
Scenario A Level 1 Level 2 Level 3 Level 4 Level 5 Level 6
Level 1 Level 2 Level 3 Level 4 Level 5 Level 6 2066597 2373653 1912667 2506425 2527306 307056 -153930 439828 460709 -460986 132772 153653 593759 614639 20881
Scenario B Level 1 Level 2 Level 3 Level 4 Level 5 Level 6
Level 1 Level 2 Level 3 Level 4 Level 5 Level 6 2923511 3173435 2482103 3463529 3260894 249924 -441408 540018 337383 -691332 290094 87459 981426 778791 -202635
Scenario C Level 1 Level 2 Level 3 Level 4 Level 5 Level 6
Level 1 Level 2 Level 3 Level 4 Level 5 Level 6 4194723 4339554 3836984 4212373 4221951 144831 -357739 17650 27228 -502570 -127182 -117604 375389 384967 9578
All Values are n Dollars
Table 5.6: Cost Differences Based on Detail Level for 15th Street
56


6. Conclusion and Recommendations
The calibration and validation of a microsimulation model is a very important part of the
design of any simulation study. The application of the model guides the design of the
calibration and validation process. If the model is to be used to determine the best future
alternative from a list of several or for sketch planning purposes than less time will be
required on calibration and validation. If the results from an alternative analysis or before-
and-after study are to be used for detailed design and to justify the cost of possible
improvements, in terms of a cost benefit ratio, then more detailed calibration and validation
will be required. The size of the model is also an important factor that determines the type
of calibration and validation that should be required. If the network size is relatively small it is
much easier to perform calibration and validation on a more detailed level.
The literature review shows that calibration and validation is used differently. Validation is
more clearly defined and it is the process of comparing model MOEs to field data to
determine if the model is properly capturing field conditions. The time periods on which the
validation data is based can also be a way to more accurately calibrate and validate a model.
Calibration and validation based on hourly values will be much easier to achieve than using
data based on 5-minute intervals. Calibration is the process of changing model parameters.
Calibration can be achieved in a variety of ways: 1) The parameters can be adjusted to
match field measured values. 2) Additional detail can be added to the model to better
capture traffic conditions. 3) Parameters can be adjusted using an iterative process after
checking the validation data. These methods can be used separately or in combination to
achieve the level calibration and validation required for a particular study.
This study was designed so that the effects of calibration and validation on future and
existing results could be easily assessed. This is not the case when performing a simulation
study in most cases. By only changing volumes in the future scenarios and the level of detail
and calibration it was easy to establish a cause and effect relationship between calibration
and validation and the various future scenarios.
This study looked at a particular study site and although not all of the results found in this
study will apply for every microsimulation study some general points can be established.
As the LOS deteriorates the level of calibration plays a more significant role.
Although more detail will result in higher modeling costs it will also achieve more
accurate representing validation results.
As the variability in flow rate increases it pays to use flow rate data for shorter
intervals, e.g. 15-minute volumes.
Travel costs may be underestimated for congested segments if a model is not well
calibrated.
57


There are several recommendations that can be made after performing this study.
Establish a good data collection plan well before collecting the data. This plan
should include the data required for model input as well as calibration and validation.
Consider the purpose of the model when determining the level of calibration
and validation that should be performed. A long-range planning model usually
requires much less calibration and validation, when compared to a detailed model
that will be used in part to determine future roadway configurations.
Perform a Highway Capacity Analysis (using the Highway Capacity Manual methods,)
as early in the process as possible to help evaluate existing and estimated future
operating conditions. This can be used to determine which areas in the network may
require special attention during the modeling process.
58


Appendix
A.1 Program that Runs Batch VISSIM Simulations
Written by Brandon Bourdon
Private Sub CommandO_Click()
Dim strt As Single
Dim ennd As Single
Dim differe As Single
numfiles = InputBoxfHow many files do you need to run?")
filepath = InputBoxfWhat is the file path?")
tminput = InputBoxfWhat is the anticipated execution time (in seconds)?")
For counter = 1 To numfiles
return val = Shell("C:\program files\PTV_Vision\VISSIM350\Exe\VISSIM.EXE", 1)
SendKeys ("%{f}o~")
SendKeys (filepath)
SendKeys ("~+{tab}{home}")
For num = 1 To counter
If num > 1 Then SendKeys ("{down}")
Next num
SendKeys ("~~")
SendKeys ("%{s}c~"), wait:=True
differe = 0
strt = Timer
Do Until differe > tminput
endd = Timer
differe = Format(endd strt, "fixed")
DoEvents
Loop
AppActivate returnval
SendKeys ("~")
SendKeys ("%(f)e~")
Next counter
59


A.2 Program the Converts Text Output Files in the Excel Format
Written by Brandon Bourdon
Private Sub Command7_Click()
Dim strt As Single
Dim ennd As Single
Dim differe As Single
numfiles = lnputBox("How many files do you need to run?")
For counter = 1 To numfiles
returnval = Shell("C:\program files\microsoft office\office\excel.EXE", 3)
SendKeys ("%{f}o~"), wait:=True
SendKeys ("c:\fut_runs")
SendKeys ("~"), wait:=True
SendKeys ("{tab}"), wait:=True
SendKeys ("{downXup}"), wait:=True
SendKeys ("{enter}"), wait:=True
SendKeys ("+{tab}"), wait:=True
SendKeys ("+{tab}"), wait:=True
For value = 1 To counter
If value > 1 Then SendKeys ("{down}"), wait:=True
Next value
SendKeys ("{enter}"), wait:=True
SendKeys ("%{d}"), wait:=True
SendKeys ("{enter}"), wait:=True
SendKeys ("%{t}"), wait:=True
SendKeys ("%{m}"), wait:=True
SendKeys ("%{f}"), wait:=True
SendKeys ("%{f}a"), wait:=True
SendKeys ("c:\fut_runs\output"), wait:=True
SendKeys ("{enter}"), wait:=True
SendKeys ("{end}"), wait:=True
SendKeys ("{backspace}"), wait:=True
SendKeys ("{left 4}"), wait:=True
SendKeys ("{delete}"), wait:=True
SendKeys ("{home}"), wait:=True
SendKeys ("{DEL}"), wait:=True
SendKeys ("{tab}"), wait:=True
SendKeys ("{up 4}"), wait:=True
SendKeys ("{enter}"), wait:=True
SendKeys ("%(s)"), wait:=True
SendKeys ("%{f}x"), wait:=True
differe = 0
strt = Timer
Do Until differe > 5
endd = Timer
60


differe = Format(endd strt, "fixed")
DoEvents
Loop
Next counter
End Sub
61


A.3 Program that Creates Simulation Input Files
Written by Brandon Bourdon
Private Sub Command8_Click()
numreps = lnputBox("How many replications do you need to create?")
FullPath = InputBoxfWhat is the file path?")
For counter = 1 To numreps
If counter = 1 Then var1 = 172083412
If counter = 2 Then var1 = 179334828
If counter = 3 Then var1 = 518633824
If counter = 4 Then var1 = 1561635041
If counter = 5 Then var1 = 2096381075
If counter = 6 Then var1 = 358376244
If counter = 7 Then var1 = 388277468
If counter = 8 Then var1 = 1410923883
If counter = 9 Then var1 = 1785031343
If counter = 10 Then var1 = 1036570441
If counter =11 Then var1 = 117695685
If counter =12 Then var1 = 835596523
If counter =13 Then var1 = 166020805
If counter =14 Then var1 = 776419229
If counter = 15 Then var1 = 1035179194
If counter = 16 Then var1 = 1488738201
If counter =17 Then var1 = 1388468204
If counter =18 Then var1 = 321741285
If counter =19 Then var1 = 885299154
If counter = 20 Then var1 = 1201055105
If counter = 21 Then var1 = 311494867
If counter = 22 Then var1 = 445306508
If counter = 23 Then var1 = 1611129779
If counter = 24 Then var1 = 637315205
If counter = 25 Then var1 = 884957404
If counter = 26 Then var1 = 1687907839
If counter = 27 Then var1 = 802580189
If counter = 28 Then var1 = 227930732
If counter = 29 Then var1 = 971035797
If counter = 30 Then var1 = 837052141
If counter = 31 Then var1 = 1574944242
If counter = 32 Then var1 = 1791234878
If counter = 33 Then var1 = 1823546729
If counter = 34 Then var1 = 1286259945
If counter = 35 Then var1 = 1697169481
If counter = 36 Then var1 = 1554250057
If counter = 37 Then var1 = 1946027738
If counter = 38 Then var1 = 1413400924
If counter = 39 Then var1 = 338647843
62


If counter = 40 Then var1 = 1545259932
If counter = 41 Then var1 = 212423515
If counter = 42 Then var1 = 1167303599
If counter = 43 Then var1 = 1821204561
If counter = 44 Then var1 = 1909442072
If counter = 45 Then var1 = 785451448
If counter = 46 Then var1 = 1681477554
If counter = 47 Then var1 = 2084262263
If counter = 48 Then var1 = 1615407249
If counter = 49 Then var1 = 1838450930
If counter = 50 Then var1 = 720307881
returnval = Shell(C:\program files\accessories\wordpad.exe", 1)
SendKeys ("%{f}o"), wait:=True
SendKeys (FullPath), wait:=True
SendKeys ("~"), wait:=True
SendKeys ("{down 6}"), wait:=True
SendKeys ("{right 17}"), wait:=True
SendKeys ("{insert}"), wait:=True
SendKeys (var1), wait:=True
SendKeys ("%{f}a"), wait:=True
SendKeys ("{home}"), wait:=True
SendKeys ("c:\fut_runs\output\")
SendKeys ("{end}"), wait:=True
SendKeys ("{left 4}"), wait:=True
SendKeys (counter), wait:=True
SendKeys ("~"), wait:=True
SendKeys ("%{f}x"), wait:=True
Next counter
End Sub
63


Glossary
The following terms have been taken from the Highway Capacity Manual [2].
Access Point An intersection, driveway, or opening on the side of a roadway.
Approach A set of lanes at an intersection that accommodates all left-turn, through, and
right-turn movements from a given direction.
Arterial A signalized street that primarily serves through-traffic and that secondarily
provides access to abutting properties, with signal spacings of 2.0 miles or less.
Calibration The process of comparing model parameters with real-world data to ensure
that the model realistically represents the traffic environment. The objective is to
minimize the discrepancy between model results and measurements or
observations.
Cycle A complete sequence of signal indications.
Default value A representative value that may be appropriate in the absence of local data.
Dwell time The time a transit unit (vehicle or train) spends at a station or a stop measured
from stopping to starting.
Geometric condition The spatial characteristics of a facility, including approach grade, the
number and width of lanes, lane use, and parking lanes.
Level of service A qualitative measure describing operational conditions within a traffic
stream, based on service measures such as speed and travel time, freedom to
maneuver, traffic interruptions, comfort, and convenience.
Light rail transit (LRT) A metropolitan electric railway system operating single cars or
short trains along exclusive rights-of-way at ground level, on aerial structures, in
subways, or occasionally in streets; an LRT can also board and discharge
passengers at track or car floor level.
Headway The time between two successive vehicles as they pass a point on the roadway,
measured from the same common feature of both vehicles.
Measure of effectiveness (MOE) A quantitative parameter indicating the performance of a
transportation facility.
Microscopic model A mathematical model that captures the movement of individual
vehicles.
64


Offset The time between the start of individual green times on a specified time datum in a
system of signalized intersections.
Performance measure A quantitative or qualitative characteristic describing the quality of
service provided by a transportation facility or service.
Pretimed control A signal control in which the cycle length, phase plan, and phase times
are preset to repeat continuously.
Queue A line of vehicles waiting to be served by the system in which the flow rate from the
front of the queue determines the average speed within the queue. Slowly moving
vehicles joining the rear of the queue are usually considered part of the queue.
Saturation flow rate The equivalent hourly rate at which previously queued vehicles can
traverse an intersection approach under prevailing conditions, assuming the green
signal is available at all times and no lost times are experienced.
Simulation model A computer program that uses mathematical models to conduct
experiments with traffic events on a transportation facility or system over extended
periods of time.
Start-up lost time The additional time consumed by the first few vehicles in a queue at a
signalized intersection above and beyond the saturation headway, because of the
need to react to the initiation of the green phase and to accelerate.
Transit stop An area where passengers await, board, alight, and transfer between transit
units.
Travel time The average time spent by vehicles traversing a highway segment including
control delay.
Validation Determining whether the selected model is appropriate for the given conditions
and for the given task; it compares model prediction with measurements or
observations.
Volume The number of vehicles passing a point on a lane, roadway, or other traffic-way
during some time period.
65


References
1. Special Report 209: Highway Capacity Manual, 3rd ed. (1997 update). TRB, National
Research Council, Washington, D.C., 1998.
2. Highway Capacity Manual, (2000 update). TRB, National Research Council, Washington,
D.C., 2000.
3. Eiefteriadou, L., J. D. Leonard II, G. List, H. Lieu, M., Thomas, R. Giguere, G. Johnson,
and R. Brewish. Beyond the Highway Capacity Manual: Framework for Selecting
Simulation Models in Traffic Operational Analyses. In Transportation Research Record
1678, TRB, National Research Council, Washington, D.C., 1999, pp. 96-106.
4. Prevedouros, P. D., and Y. Wang. Simulation of Large Freeway and Arterial Network
with CORSIM, INTEGRATION, and WATSim. In Transportation Research Record 1678,
TRB, National Research Council, Washington, D.C., 1999, pp. 197-207.
5. Wang, Y., and P. D. Prevedouros. Comparison of INTEGF1ATION, TSIS/CORSIM, and
WATSim in Replicating Volumes and Speeds on Three Small Networks. In
Transportation Research Record 1644, TRB, National Research Council, Washington,
D.C., 1998, pp. 80-92.
6. Venglar, S. P., D. B. Fambro, and T. Bauer. Validation of Simulation Software for
Modeling Light Rail Transit. In Transportation Research Record 1494, TRB, National
Research Council, Washington, D.C., 1995, pp. 161-166.
7. Rakha, H., M. Van Aerde, L. Bloomberg, and X. Huang. Construction and Calibration of
a Large-Scale Microsimulation Model of the Salt Lake Area. In Transportation Research
Record 1644, TRB, National Research Council, Washington, D.C., 1998, pp. 93-102.
8. Denney Jr., R. W., J. C. Williams, S. Bhat, and S. A. Ardekani. Calibrating NETSIM for
a CBD Using the Two Fluid Model. Proc., Advanced Traffic Management Conference,
St. Petersburg, Florida, 1993, pp. 169-184.
9. Boone, J. L., and J. E. Hummer. Calibrating and Validating Traffic Simulation Models for
Unconventional Arterial Intersection Designs. In Transportation Research Record 1500,
TRB, National Research Council, Washington, D.C., 1995, pp. 184-192.
10. Khasnabis, S., R. R. Karnati, and R. K. Rudraraju. NETSIM-Based Approach to
Evaluation of Bus Preemption Strategies. In Transportation Research Record 1554,
TRB, National Research Council, Washington, D.C., 1996, pp. 80-89.
11. Sisiopiku, V. P., N. M. Rouphail, and A. Santiago. Analysis of Correlation Between
Arterial Travel Time and Detector Data from Simulation and Field Studies. In
Transportation Research Record 1457, TRB, National Research Council, Washington,
D.C., 1994, pp. 166-173.
12. Wong, S. Y. Capacity and Level of Service by Simulation: A Case Study of TRAF-
NETSIM. Proc., International Symposium on Highway Capacity, Karlsruhe, Germany,
1991, pp. 467-483.
13. Torres, J. F., A. Halati, A. I. Gafarian, and S. Guevrekian. Statistical Guidelines for
Simulation Experiments, Vol. 3. JFT Associates. FHWA, U.S. Department of
Transportation, 1983, pp. 12.
14. Leutzbach, W. Some Remarks on the History of the Science of Traffic Flow. Workshop
on Traffic and Granular Flow, Germany, 1995, pp. 3-10.
66


15. Zhang, Y, L. E. Owen, and J. E. Clark. Multiregime Approach for Microscopic Traffic
Simulation. In Transportation Research Record 1644, TRB, National Research Council,
Washington, D.C., 1998, pp. 103-115.
16. Reiter, U. Empirical Studies as Basis for Traffic Flow Models. Proc., Second
International Symposium on Highway Capacity, Sydney, Australia, 1994, pp. 493-502.
17. Leutzbach, W., and R. Wiedemann. Development and Applications of Traffic Simulation
Models at the Karlsruhe Institut fur Verkehrwesen. In Traffic Engineering and Control,
Vol. 27, No. 5, 1986, pp. 270-278.
18. Wiedemann, R. Modeling of RTI-Elements on Multi-Lane Roads. Proc., Drive
Conference, Brussels, Belgium, 1991, pp. 1007-1019.
19. Ludmann, J., D. Neunzig, and M. Weilkes. Traffic Simulation with Consideration of
Driver Models, Theory and Examples. In Vehicle System Dynamics: International
Journal of Vehicle Mechanics and Mobility, Vol. 27, No. 5-6, 1997, pp. 491-516.
20. Bauer, T. Lessons Learned from VISSIM Applications in the United States. PTV Vision
User Group Meeting 2000, Karlsruhe, Germany, 2000.
21. Fellendorf, M. VISSIM for Traffic Signal Optimization. In Traffic Technology
International 96, Surrey, United Kingdom, 1996, pp. 190-192.
22. PTV Planung Transport Verkehr AG. VISSIM 3.50 Users Manual, Karlsruhe, Germany,
2000.
23. Hoyer R., and M. Fellendorf. Parametrization of Microscopic Traffic Flow Models
Through Image Processing. Proc., IFAC Symposium, Chania, Greece, 1997, pp. 889-
894.
24. Fellendorf, M. and P. Vortisch. Validation of the Microscopic Traffic Flow Model VISSIM
in Different Real-World Situations, http://www.itc-world.com. Accessed May 21, 2001.
25. Moen, B, J. Fitts, D. Carter, and Y. Ouyang. A Comparison of the VISSIM Model to
Other Widely Used Traffic Simulation and Analysis Programs, http://www.itc-world.com.
Accessed May 21, 2001.
26. Bloomberg, L, and J. Dale. A Comparison of the VISSIM and CORSIM Traffic
Simulation Models on a Congested Network. Prepared, 7&h Annual Meeting of the
Transportation Research Board, Washington, D.C., 2000.
27. Bloomberg, L. and J. Dale. A Comparison of the VISSIM and CORSIM Traffic
Simulation Models. Prepared, Institute of Transportation Engineers Annual Meeting,
Nashville, Tennessee, 2000.
28. Anderson J. Travel Time Prediction in Urban Road Networks. Proc., of the IFAC
Symposium, Chania, Greece, 1997, pp. 1109-1114.
29. Fellendorf. M, C. Mac Aongusa, and M. Pierre. LRT Priority within the SCATS
Environment in DublinA Traffic Flow Simulation Study. Proc., Third International
Conference on Urban Transport and the Environment for the 21st Century, Italy, 1997,
pp. 53-62.
30. Dale, J., T. Bauer, T. Bevan, K. McKinley, and T. Slind. Evaluating Arterial Street
Transit Preferential Treatments. Prepared, 7(f Annual Meeting of the Institute of
Transportation Engineers, Nashville, Tennessee, 2000.
31. Note, I. Integration of a Center-Running Guided Busway into an Arterial Street.
http://www.itc-world.com. Accessed May 21, 2001.
32. Random House Webster's College Dictionary. Random House, New York, 1992.
67