See discussions, stats, and author profiles for this publication at: https://www.researchgate.

net/publication/328942453

Calibration Methodology of Microsimulation Model for Unsignalized

Intersection Under Heterogeneous Traffic Conditions

Conference Paper · December 2017

CITATION READS

1 860

4 authors:

Madhumita Paul Vaibhav Charan

Indian Institute of Technology Kharagpur Indian Institute of Technology Roorkee
23 PUBLICATIONS   88 CITATIONS    1 PUBLICATION   1 CITATION   

SEE PROFILE SEE PROFILE

Vishal Soni Indrajit Ghosh

Indian Institute of Technology Roorkee Indian Institute of Technology Roorkee
1 PUBLICATION   1 CITATION    81 PUBLICATIONS   610 CITATIONS   

SEE PROFILE SEE PROFILE

Some of the authors of this publication are also working on these related projects:

Microsimulation modelling View project

Effect of Countdown Display on Traffic Flow Characteristics and Driver Behaviour at Signalised Intersections View project

All content following this page was uploaded by Madhumita Paul on 14 November 2018.

The user has requested enhancement of the downloaded file.

 Calibration Methodology of Microsimulation Model for Unsignalized
Intersection Under Heterogeneous Traffic Conditions

Paul M., M.Tech,1 Charan V.,2 Soni V. ,2 and

Ghosh I., Ph.D, A.M.ASCE3

1
Ph.D. Candidate, Department of Civil Engineering, Indian Institute of Technology (IIT) Roorkee,
Uttarakhand, 247667, e-mail: [email protected]
2
B.Tech Student, Department of Civil Engineering, Indian Institute of Technology (IIT) Roorkee,
Uttarakhand, 247667, e-mail: [email protected]
2
B.Tech Student, Department of Civil Engineering, Indian Institute of Technology (IIT) Roorkee,
Uttarakhand, 247667, e-mail: [email protected]
3
Assistant Professor, Department of Civil Engineering, Indian Institute of Technology (IIT)
Roorkee, Uttarakhand, 247667, Ph: +91-1332-285533; e-mail: [email protected]

ABSTRACT
Indian unsignalized intersections operate quite differently in comparison to their western
counterparts. To address traffic heterogeneity, lane indiscipline, non-priority rules, lack of
movement control and non-uniform road geometry that exist on Indian roads, microsimulation is
the best-suited technique. While a few attempts have been made on the calibration of signalized
intersection under heterogeneous traffic, unsignalized one is still an unexplored area. Hence, the
present study proposes a calibration methodology of PTV VISSIM simulation model for an
unsignalized intersection under heterogeneous traffic condition. The 85th percentile accepted gap
time is utilized as a performance measure in calibration. Calibration parameters are identified using
Morris sensitivity analysis, and their optimum values are obtained using Genetic Algorithm (GA).
Simulation results show that VISSIM provides reasonable accepted gap time for the unsignalized
intersection. Overall, it is concluded that, using the proposed methodology, the developed network
performs well and replicates field situation efficiently.

Keywords: unsignalized intersection, heterogeneous traffic, gap acceptance, PTV VISSIM,

calibration

INTRODUCTION
In the present scenario of ever-growing traffic and increase in the use of road transportation, there
arises the need of a tool to analyze any complex road network and traffic flow. The problem
becomes more critical in developing countries like India where heterogeneous traffic condition
prevails. Heterogeneous traffic is characterized by the “mixture of vehicles having diverse static
(length, width, etc.) and dynamic (acceleration/deceleration, speed, etc.) properties. These vehicles
include motorized as well as non-motorized types, and their composition is highly transient” (Minh
and Sano, 2003). The diversity in such type of traffic makes them challenging to model. Among
all road networks, modelling of unsignalized intersections is more difficult, as traffic signals are
not present and drivers of self-organizing traffic flow need to choose an appropriate time for
carrying out various maneuvers. Interestingly, in India, unsignalized intersections operate quite
differently in comparison with their western counterparts. Be it an all-way or two-way stop-
controlled one, traffic signs associated with the priority rules such as the stop and yield signs are

1
not respected by the drivers due to the non-existence of enforcement. Priorities are basically
established by the drivers’ perspective of situation. Traffic flow theories seek to describe this
complex traffic behaviour mathematically, but still, it is difficult to analyze by using conventional
evaluation techniques. Thus, modelling of such traffic, especially at unsignalized intersection
locations, is a challenging task as it involves the simultaneous study of a vast number of
microscopic traffic parameters affecting driver behaviour.
Traffic microsimulation technique has been used as the most effective method for analysis
of complex traffic behaviour. It is a computerized analytical tool that performs highly detailed
analysis of activities through a series of sophisticated algorithms incorporated into a model and
capture the real world traffic interaction. For the purpose of modelling, there are different
microsimulation software readily available such as AIMSUN, DRACULA, SUMO, PARAMICS,
PTV VISSIM, etc. Among such microsimulation software, PTV VISSIM becomes the more
popular and reliable simulation package utilized by researchers in traffic engineering field. One of
the key advantages of VISSIM is that it has a high level of flexibility to represent any motorized
and non-motorized vehicle class, interactions between individual vehicles and complex driving
behaviour phenomena. Additionally, it permits users to model in-detail configurations of road
geometry as well as drivers’ behavioural characteristics. The road networks in VISSIM can be
created either as lane oriented or space oriented, i.e., vehicles can travel anywhere on the road
without any lane restriction. Due to all of these facilities mentioned above, VISSIM is gaining
huge popularity to model heterogeneous traffic besides homogeneous one
In many developing countries like India, a diverse mix of vehicles is present with various
geometrical issues. Therefore, the model parameters in VISSIM are first needed to be calibrated
in order to replicate real life traffic behaviour in the simulation. Calibration states the variability
in driver behaviour, vehicular characteristics as well as, to some extent, the effect of road
geometry. In most of the western countries, the calibration of different simulation models is
primarily done for homogeneous traffic conditions, where good lane discipline and road geometry
exist (Dowling et al., 2004; Park and Qi, 2005). Earlier, Hossain (2001), Mathew and
Radhakrishnan (2010), Manjunatha et al. (2013), Siddharth and Ramadurai (2013) attempted the
calibration of VISSIM for heterogeneous traffic condition. These studies provide a broad idea
about model calibration and model parameter optimization under heterogeneous condition but all
of them focused primarily on signalized intersections where traffic movements are less
complicated in comparison to Indian unsignalized intersections. In a few previous studies,
calibration is done at unsignalized intersections under the homogeneous condition where priority
rules are obeyed by drivers (Caliendo and Guida, 2012; Liu et al., 2012). However, calibration of
unsignalized intersection under heterogeneous traffic conditions is still an unexplored area.
Till date, several methodologies had been adopted by researchers to calibrate VISSIM
simulation model (Fellendorf and Vortisch, 2001; Gomes et al., 2004; Park and Qi, 2005; Mathew
and Radhakrishnan, 2010; Manjunatha et al., 2013; Siddharth and Ramadurai, 2013), etc. The
calibration of traffic microscopic simulation models is such a process, from where the optimal
parameters are identified and matched with the field data so that model can accurately represent
field measured or observed traffic condition. Hence, a vital step in calibration is optimization of
model parameters. Different algorithms had been applied in past studies to find the optimum values
of the model parameters. Dabiri and Abbas (2016) optimized signal timing in VISSIM using
Particle Swarm Optimization (PSO). Qin et al. (2016) calibrated a signalized intersection network
in VISSIM using Orthogonal GA (OGA), and they compared this method with the conventional
GA. In a significant amount of past studies carried out in developed countries, GA had been

2
utilized for automatic calibration and to determine model parameters’ optimum values (Kim et al.,
2005; Park and Qi, 2005; Ge and Menendez, 2014), etc. Similarly, in India, under heterogeneous
traffic conditions, researchers selected GA as an optimization tool in order to calibrate the VISSIM
model (Mathew and Radhakrishnan, 2010; Manjunatha et al., 2013; Siddharth and Ramadurai,
2013). In the process of calibration, optimization task includes certain microscopic traffic
parameters as performance measures. By comparing and minimization differences of these
measures obtained from simulation models with field one, the accuracy of calibration is defined.
Delay and traffic flow are the performance measures most commonly used in the calibration of
VISSIM model for signalized intersection under heterogeneous traffic condition.
Hence, the present study aims to demonstrate the 'state of the art' calibration and validation
techniques of VISSIM model for unsignalized intersection under heterogeneous traffic condition
where lack of lane discipline, the absence of priority rules, poor traffic control, etc. results in
unsystematic traffic flow. For this purpose, calibration is done considering all vehicle class present
in the road network under heterogeneous traffic conditions and by considering a microscopic
parameter, accepted gap time, as a performance measure. The vital steps of the proposed
methodology include automated sensitivity analysis, calibration of VISSIM model parameters and
validation of the calibrated model.
METHODOLOGY
In a developing country like India, calibration of a microsimulation model is a tricky task,
especially under heterogeneous traffic. Any error in the process can result into the unsatisfactory
performance of the model and hence model will be unable to generate desired field condition.
Additionally, in case of Indian unsignalized intersection, the absence of priority rules, lack of
movement control, drivers’ decision according to own prospection, etc. prevail which altogether
present a challenging task to calibrate such road network. To address these above-mentioned
characteristics of Indian unsignalized intersections, an efficient calibration of VISSIM simulation
model is carried out using microscopic parameter, gap time accepted by the least prioritized minor
road vehicles.
In this context, a good representation of field traffic, as well as road geometry, are done in
the simulation model to reflect the heterogeneity in vehicle mix, as well as unusual road geometry
such as non-uniform road width. After preparing the basic geometry of road network and setting
the composition of different vehicle class, selection of parameters that can affect the performance
of network are carried out using sensitivity analysis. Identified sensitive parameters are then
incorporated into the simulation model. Consequently, using an optimization technique, GA,
calibrated parameters along with their optimum values are determined for the selected unsignalized
intersection under heterogeneous traffic condition. Currently, use of GA is increasing for model
optimization purpose by many researchers in the field of traffic engineering. GA uses the process
of natural selection that states the survival of the fittest, in order to generate high-quality
optimization solution. As per HCM (2010), several traffic parameters such as control delay, gap
acceptance, and capacity are acknowledged as the standard performance measures for the
unsignalized intersection. Only one study is found, where modelling of U-turn movements with
non-traversable median cross section was done using VISSIM for unsignalized intersections in
discipline traffic environment (Liu et al., 2012). However, in the present study, the proposed
calibration method is applied at an Indian unsignalized intersection, where no traffic movement is
prioritized by a stop or yield signs. For calibration purpose, the 85th percentile accepted gap time
is calculated. This particular value is considered as accepted gap time and utilized in calibration
process as the performance measure. Later, the model is validated using another set of field data
3
taken from the same intersection for different time period. If the network can replicate the field
situation within acceptable error range, the calibration process is considered to be completed.
Otherwise, the process is to be repeated until the satisfactory level of performance is reached.
Usually, an error within 5% is considered acceptable in network calibration.
The major steps followed in the proposed methodology shown in Figure 1. It is to be noted
that VISSIM has two separate car-following models namely Wiedemann 74 model (W-74) and
Wiedemann 99 model (W-99). As per VISSIM user manual (PTV VISSIM 7.0), the W-74 model
is mainly suitable for urban traffic. The selected location is a semi urban unsignalized intersection
located at the intercity highway. Hence, in the present study, the W-74 model is used to calibrate
an unsignalized intersection under heterogeneous traffic condition.

Figure 1. Proposed methodology of calibration

MEASUREMENT OF ACCEPTED GAPS
In order to calibrate the drivers’ behaviour in the microscopic simulation model, accepted gap time
is selected as an effective parameter. According to HCM (2010), gap time is one of the
performance measures at the unsignalized intersection which indicates the time in secs, for the

4
front bumper of the second vehicle of two successive vehicles to reach the starting point of the
front first bumper of the first. However, in the present study, the conventional calculation process
of the accepted gap is modified based on the characteristics of unsignalized intersection prevailing
in developing countries like India. The reason is that in India a diverse mix of motorized and non-
motorized vehicle classes with a wide range of static and dynamic characteristics are observed.
Additionally, due to the absence of lane discipline, vehicles are found to travel side by side of one
another on the minor road. When high traffic volume exists on the major road, many minor roads
vehicles those are having the least priority in any unsignalized intersection lose their patience for
long waiting time. As a result, they accept smaller gaps by forcing the major road vehicles to slow
down in order to cross the intersection. There are several occasions when even a single gap in the
major road is accepted by more than one vehicle moving parallel to each other. After crossing the
major road traffic, these vehicles merge into the far lane following a single line, one after another.
At the same time, it is observed in the field that minor road vehicles try to cross multiple streams
at a time. Thus, the gap acceptance behaviour is very complex at Indian unsignalized intersections.
Hence, a modified procedure is adopted to measure the gap accepted by a minor road vehicle while
crossing a stream (or multiple streams) or merging with a stream. Initially, they look for a gap in
the near side lanes and only after moving further into the conflict area, look for a gap in the far
side major lanes. Hence, the accepted gap in such situations is measured from the exit of major
road vehicle at the near side to the arrival of the vehicle on the far side of the major road after the
subject vehicle (minor road vehicle) has completed the required maneuver. This is explained in
Figure 2.

Figure 2. Measurement of accepted gap

Using this procedure, gap value is calculated for each minor road vehicle. Later, the 85th percentile
gap value is identified, considering all the accepted gaps, and utilized as a performance measure
for model calibration of the study site.
SITE SELECTION AND DATA COLLECTION
For the study purpose, a 3-legged unsignalized intersection located in sector 19, Dwarka, National
Capital Region (NCR) is selected. The intersection is provided with channelized left turning lanes
(for left-hand driving condition). The major road of the selected site has three lanes on one side of
the median. The site has no influence of roadside parking, bus stops or any other side frictions and
very limited pedestrian and cyclist movements.

5
 From the field, road geometry as well as traffic data are collected to develop, calibrate and
validate VISSIM simulation model. Geometric details such as numbers of lanes, lane width and a
certain length of lanes are collected and measured to develop the road network in VISSIM.
Videography technique is used for the data collection process. A high-definition video camera is
set up at an elevated point to obtain the clear view of the study location. Being located on an
intercity highway, no distinct peak or off-peak hours are observed for this site. Therefore, data are
collected for a particular daytime (from 10 A.M. to 2 P.M.) on weekdays under good visibility
conditions. Out of four hours data, the first two-hour data is used for calibration while the next
two-hour data is used in validation. Recorded video is played on a large television screen at a frame
rate of 25 frames per second. Necessary data such as hourly traffic volume, vehicle composition,
speed, acceleration-deceleration characteristics of each vehicle, and accepted gap values for the
least prioritized minor road vehicles are extracted. Gaps values are calculated as the time difference
between the exit of a vehicle at the near stream of major road and the arrival of next far stream
major road vehicle at the reference line placed in the middle of the conflict area. This reference
line is simulated exactly with a data collection point in microsimulation models to get accepted
gap time accurately from the simulation. At the study site, six vehicle classes are observed such as
car, big car, motorized 2-wheeler (2w), 3-wheeler (3w), Light Commercial Vehicles (LCV), and
Heavy Vehicles. LCV includes pick-up trucks whereas HV represents bus and truck. The details
of the selected intersection along with characteristics of vehicle classes are provided in Table 1.
Table 1. Details of selected intersection and vehicular characteristics
Vehicle HV
Site Characteristics 2w Car Big Car 3w LCV
Class Bus Truck
Airport-
NHAI-
Proportion
Site AIFF 12 72 11 1 2 0.70 1.3
(%)
Intersect
ion
Type of length, width 1.8, 4.4, 4.89, 3.2, 5, 10.1, 7.5,
3-legged
Intersection (m) 0.5 1.5 2.14 1.4 2.05 2.43 2.35
3-lane
Vehicular Characteristics

1-way Desired
Traffic
on the speed 45 50 55 20 25 35 35
movement
side of (km/hr)
median
Maximum
Approach
9 Acceleration 2.6 3.1 3.4 1.5 1.3 1.6 1.1
Width (m)
(m/s2)
No of lanes Desired
on the side 3 Acceleration 2.3 2.7 3 0.8 0.7 1.1 0.7
of median (m/s2)
Average
Maximum
Traffic
3736 Deceleration -2.1 -2.8 -2.5 -1.2 -1.1 -1.6 -1.4
Volume
(m/s2)
(veh/hr)
85th
Desired
percentile
5.42 Acceleration -1.8 -1.9 -1.8 -0.6 -0.6 -0.7 -0.7
accepted
(m/s2)
gap (sec)

6
Table 1 shows the details of selected intersection and traffic characteristics including its width,
traffic volume, vehicle dimensions (Chandra and Parida, 2004; PTV VISSIM 7.0), desired speeds,
and acceleration/deceleration, etc.
NETWORK PREPARATION IN VISSIM
A simulation model of a three-legged unsignalized intersection is developed in VISSIM. The
primary step of model development is to define the road geometry by tracing a google image of
the selected intersection. The dimensions of different approaches are set accordingly in the
network. The simulated approaches are prepared with links, and all the links are joined together
using connectors (curved and straight lines). In developing countries like India, the major
geometrical issues are non-standard lane width and a non-equal fraction of total approach width.
As a result, it is important to address the geometry as per field in the simulated network to simulate
the selected unsignalized intersection. For both major and minor roads, left turning lanes are
provided separately as per the field geometry. Once the geometry of the road network is defined,
traffic flow parameters for each approach have been assigned. Routing decisions for turning traffic
(left, straight and right) are marked with appropriate proportions. Traffic volume and vehicles
composition are set in accordance with the recorded data from the site. Vehicle classes such as
motorized 2w, car, big car, bus, truck etc. are available in VISSIM to represent the motorized
vehicle categories. On the other hand, some 3-D models of the vehicle classes, which run only in
heterogeneous traffic such as 3-wheeler and LCV such as pickup truck (e.g., TATA 407) are not
readily available in VISSIM. Hence, these are modeled by considering the car as a base model and
changing its static and dynamic characteristics. Moreover, the dimensions of these two new vehicle
types are also integrated into the model. Field measured speed and acceleration/deceleration of
each vehicle class have been provided in the network along with their desired and maximum
values. Conflict areas are then identified and priority rules are decided in those areas. There is a
possibility of three types of conflicts such as the interaction of through vehicles travelling on the
major road with right turning vehicles from the minor road as well as from major road, right turning
vehicles from the major road with right turners from the minor road. In the first two cases, through
moving traffic on the major road have priority whereas in the third case right turning vehicles from
the major road have priority over minor road vehicles.
After setting up geometric design, vehicles composition, and priority rules, next step is to
establish the driving behaviour parameters. VISSIM has inbuilt characteristics for a homogeneous
traffic flow. Hence, in order to set it for heterogeneous traffic that prevails in India, several changes
need to be done. Firstly, vehicles are allowed to place anywhere in the lane and overtake along the
right or left of the slower vehicle. This is done by setting W-74 driving behaviour parameter
‘desired position at free flow’ to any among other options and for overtaking on the same lane, it
is allowed for both (left and right). The reduced speed areas are also introduced near an intersection
to exhibit field traffic scenario. All the above-mentioned steps ensure the operation of VISSIM
according to heterogeneous traffic condition. At last, data collection points are set in VISSIM at
the exact location where field observations are made and from the simulated model, necessary data
are extracted accordingly. A snapshot of a simulated network of study site is shown in Figure 3.

7
 Figure 3. A snapshot of simulated network
SELECTION OF PARAMETERS FOR CALIBRATION
Before utilizing VISSIM simulation model for estimation of the 85th percentile accepted gap value
of minor road vehicles, the model needs to be calibrated and validated based on field data to ensure
that it provides reasonable accepted gap time for the unsignalized intersection. One of the criteria
in simulation model calibration is that the calibrated parameter must be within a reasonable range.
For that purpose, initially, several behavioural parameters belonging to the following, lane change,
lateral and signal control models are considered to perceive their influence on gap value. Later, for
each selected parameter their respective ranges are defined to carry out the sensitivity analysis.
The last step of model calibration is to adjust the selected parameter within the selected range until
a reasonable correspondence between field data and the output of the simulation model is achieved.
However, to ascertain the need for calibration, the model is calibrated first with its default setting
and the accepted gap is determined from VISSIM result (.mer file). There is a significant error
(21.20%) observed between accepted gap from the field and simulated one that obtained using
default settings of model parameters Hence, further calibration is done by changing its default
settings in order to minimize the error between field-measured 85th percentile gap value and
simulated one. In order to determine which model parameters of VISSIM have a significant
influence on gap value, a total of twelve model parameters are selected for sensitivity analysis.
These are the three main parameters of the W-74 car following model such as the average standstill
distance (ax_average), the additive and multiplicative factors of safety distance (bx_add and
bx_mult) which are essential to calibrate to reflect the driving behaviour. Additional parameters
such as look ahead distance (minimum) and look back distance (minimum) are selected. Lane
change model parameters such as maximum deceleration (own and trailing vehicle), waiting time
for diffusion, min. headway (front/rear) and safety distance reduction factor are selected. Among
lateral movement parameters, lateral distance at 0 kmph (standing) and 50 kmph (driving) are
considered from the following model.

IDENTIFICATION OF SENSITIVE PARAMETERS

Sensitivity analysis is carried out to identify the sensitive parameters for network w.r.t. a given
gap time. There are several sensitivity analysis techniques, e.g., Morris method, ANOVA
(Analysis of Variance) and FAST (Fourier Amplitude Sensitivity Testing). In the present study,

8
Modified Morris elementary method (EE method) has been conducted to identify the sensitive
parameters taking gap acceptance as a measure of sensitivity. This method is proposed by
Campolongo et al. (2007) and it is a fast and efficient method as compared to more demanding
ANOVA and FAST techniques. For each of the selected parameters, their respective domains are
defined, and some of are selected from the previous studies as shown in Table 2. EE method is
based on ‘one at a time’ approach in terms of their domain, where the value of only one parameter
is increased or decreased, and other values are kept constant. To automate the sensitivity analysis,
the required script is written in Python 3.6 with the help of its popular sensitivity library (SALib).
The EE method uses the numerical approximation of the partial derivative of the model
function w.r.t each input variablexi, . The model function for a model ‘y’ with input parameters x1,
x2, x3…xn , is expressed as y = f(x1 , x2 , … , xn ). Then by moving each input parameter xi, OAT by
a grid interval of δ the output of model function becomes y = f(x1 , x2 , … , xi−1 , xi + δ. ê,i … , xn ).
Finally, the EE of ith parameter based on partial derivative formulae is defined as Ei (𝐱).and
expressed as Equation 1.

EEi (𝐱) = [f(x1 , x2 , … , xi−1 , xi + δ ∙ ê,i … , xn ) - f(x1 , x2 , … , xn )]/ δ

Where, x = input parameters, (𝑥1 , 𝑥2 , … , 𝑥𝑛 ) and 𝑒̂𝑖 is the unit vector in direction of the ith
parameter, defined as 𝑒̂𝑖 = (0,0,…0,1,0..0). The set of EE values that are generated from a given
set of variables’ domains create several trajectories. Consequently, a total of ‘r’ optimized
trajectories are selected from these random trajectories. According to Campolongo et al. (29), for
an n-dimensional input vector and ‘r’ optimized trajectories, an 𝑟 ∙ (𝑛 + 1) × 𝑛 matrix is
obtained. This is further analysed into the morris:analyze function to obtain values such as µ
(arithmetic mean), σ (standard deviation), and µ* (absolute mean) corresponding to each input
parameter. Parameters resulting in a high value of µ* and σ are affirmed as the higher sensitive
ones to model output which is accepted gap time in the present study. Hence, for the developed
model the parameters which are found sensitive to gap time are shown in Table 2 with a right
check mark ().

TABLE 2 Parameters list and results of sensitivity analysis

Parameters'
Sensitive
Model Parameters Default value selected
Parameters
domain
W74ax 2 0 to 4.00 
Following

W74bxAdd 2 0 to 4.00 
W74bxMult 3 0 to 5.00 ×
LookAheadDistMin 0m 0 to 30 
LookBackDistMin 0m 0 to 30 
MaxDecelOwn -4.00 m/s2 -1 to -5 
Lane Change

MaxDecelTrail -3.00 m/s2 -1 to - 4 

DiffusTm 60 25 to 65 ×
MinHdwy 0.5 0.1 to 1 
SafDistFactLnChg 0.6 0.1 to 1 ×
Lat
era

LatDistStandDef 1 0.1 to 0.7 ×

l

9
 LatDistStand(50) 1 0.1 to 0.7 ×

Table 2 shows that among twelve selected parameters, total seven model parameters are sensitive
to accepted gap.
MODEL CALIBRATION USING GA AND VALIDATION
After sensitivity analysis, the calibration process is done where sensitive parameters of VISSIM,
those affect the behaviour of the network, are adjusted till the model replicates field conditions.
Calibration is an optimization problem in which an iterative procedure is performed to determine
the calibrated parameters and their optimum values. The process of parameter calibration can be
tedious and time-consuming. More importantly, it is difficult to guarantee that the optimum
solution can be achieved by manually adjusting the parameters. For this purpose, GA is used,
which is a very popular optimization technique, and it has the ability to speed up the parameter
calibration procedure to enhance the chance of achieving the optimum solution. For the present
study, it is run in MATLAB by integrating VISSIM with the help of COM interface for automating
the calibration process. GA is a heuristic optimization technique that emulates the mechanics of
evolution and genetics to make a set of data (population) fitted over some generations by
systematically removing unfit members and fusing the fitted ones to obtain better offspring (new
values). Therefore, a fitness function value is defined as the percentage error between the 85th
percentile accepted gap value computed from the simulation result and a fixed target accepted gap
value that obtained from the field. A low error value indicates a higher fitness. To get the best
outcome, multiple VISSIM runs are conducted and GA minimizes the error value until it observed
to be low. Subsequently, it provides the corresponding input model parameters along with their
optimum values. The calibrated parameters and their optimum values that obtained after absolute
calibration are given in Table 3. In optimization process, GA includes the population size of 20
with a total generation of 50 and stall generation is taken as 25. After calibration, the 85th percentile
accepted gap value is obtained and found to be very close to the field obtained value with a
percentage error of 1.38 %, as shown in Table 3.
The calibrated model is validated with a different data set other than the one used for
calibration. This external validation is done with the next two-hour data (12 noon – 2 p.m.) as the
first two hours data (10 a.m -12 noon) is used for calibration. The absolute error between the 85th
percentile accepted gap time obtained from the calibrated model and field one is computed. The
validation result obtained and the error is found to be very low (3.10%). Comparison of gap time
found from validation, calibration along with the field ones are also shown in Table 3.
Table 3. Results of calibration and validation
Calibrated Parameters Optimum Values
W74ax 2.242
W74bxAdd 1.223
LookAheadDistMin 26.54
LookBackDistMin 22.15
MaxDecelOwn -3.341
MaxDecelTrail -3.025
MinHdwy 0.535
Comparison of the 85th percentile accepted gap values
Field Accepted Gap Simulated Gap Validated Gap
Time Time Time
10
 85th percentile gap
5.42 5.49 5.58
(sec)
percentage error (%) - 1.38 3.10

CONCLUSION
The present study proposes a methodology to calibrate VISSIM model for unsignalized
intersection under heterogeneous traffic condition. The behavioral characteristics of unsignalized
intersection in the developing countries like India are quite different from its western counterpart.
Modelling of such road network is still an unexplored area. For this purpose, a 3-legged
unsignalized intersection located in the National Capital Region (NCR) is selected. A microscopic
parameter, gap acceptance, is utilized as a measure of sensitivity and to calibrate the model of
selected road network. This is the first attempt made to investigate the performance of unsignalized
intersection in simulation model using gap time. Before carrying out the calibration, primarily, the
simulated intersection network is prepared by considering all the characteristics of heterogeneous
traffic conditions along with standard and non-standard vehicle classes (3w and LCV), all unique
features of each vehicle class and addressing other atypical issues, i.e., non-uniform road width,
and non-lane based traffic movements, etc. Conflict areas and priority rules are also provided in
the network.
However, the default parameter values that are incorporated in VISSIM or any other
simulation software need modification in accordance to the road network and traffic conditions
that prevail at that section to simulate field situation adequately. Hence, the vital steps of the
proposed methodology comprised of automated sensitivity analysis, calibration of VISSIM model
parameters using the 85th percentile accepted gap time of the minor road vehicles and validation
of the calibrated model. For sensitivity analysis, twelve parameters including three main
parameters W-74 car following model are selected, and among them, in total, seven are found to
be sensitive to 85th percentile gap value. Consequently, calibration is carried out using GA to find
out the calibrated parameters and their optimum values are obtained by minimizing the error
between the simulated and field gap time. The calibrated model is validated using gap data for
other time duration which is not used in calibration and error is observed to be less than 5 percent.
As no distinct peak and off hours exists at that location, the outcome implies that the proposed
model is applicable for whole day traffic of that particular site as well as for the locations having
similar traffic characteristics and road geometry. The present study also provides a comprehensive
idea about model calibration for unsignalized intersection under heterogeneous traffic condition.
Additionally, the findings of the present study can be used as a guiding tool for microscopic traffic
simulation practitioners in calibrating VISSIM models for operational analysis at any unsignalized
intersection under heterogeneous traffic conditions. This study can be further extended by
considering more number of intersections, individual accepted gap time as per movement priority
for a better understanding of the effect of various VISSIM parameters. The effects of individual
vehicle class on model parameters can also be identified by conducting sensitivity analysis and
calibration w.r.t each vehicle class.
REFERENCES
Caliendo, C., and Guida, M. (2012). “Microsimulation Approach for Predicting Crashes at
Unsignalized Intersections Using Traffic Conflicts.” Journal of Transportation
Engineering, 138(12), 1453–1467.
Campolongo, F., Cariboni, J. and Saltelli, A. (2007) “An effective screening design for sensitivity

11
 analysis of large models.” Environmental modelling & software, 22(1), 1509-1518.
Chandra, S., and Parida M. (2004) “Analysis of urban road traffic through simulation.” Indian
Highways, 87-102.
Dabiri, S. and Abbas M. (2016). “Arterial traffic signal optimization using Particle Swarm
Optimization in an integrated VISSIM-MATLAB simulation environment.” In 19th
International Conference on Intelligent Transportation Systems (ITSC), IEEE, 766-
771.
Dowling, R., Skabardonis, A., Halkias, J., McHale, G., and Zammit, G. (2004). “Guidelines for
calibration of microsimulation models: framework and applications.” Transportation
Research Record: Journal of the Transportation Research Board, 1876, 1-9.
Fellendorf, M., and Vortisch, P. (2001). “Validation of the microscopic traffic flow model VISSIM
in different real-word situation.” In 80th Annual Meeting of the Transportation Research
Board, Washington, DC.
Ge, Q., and Menendez, M. (2014). “An efficient sensitivity analysis approach for computationally
expensive microscopic traffic simulation models.” International Journal of
Transportation, 2(2), 49-64.
Gomes, G., May, A., and Horowitz, R. (2004). “Congested freeway microsimulation model using
VISSIM.” Transportation Research Record: Journal of the Transportation Research
Board, 1876, 71-81.
HCM (2010), “Highway Capacity Manual”, Fifth Edition, Transportation Research Record:
Transportation Research Board of the National Academies, Washington, D.C.
Hossain, M. (2001). “Estimation of saturation flow at signalised intersections of developing cities:
a micro-simulation modelling approach.” Transportation Research Part A: Policy and
Practice, 35(2), 123- 141.
Kim, S.J., Kim, W. and Rilett, L. (2005). “Calibration of microsimulation models using
nonparametric statistical techniques.” Transportation Research Record: Journal of the
Transportation Research Board, 1935, 111-119.
Liu, P., Qu, X., Yu, H., Wang, W., and Cao, B. (2012). “Development of a VISSIM simulation
model for U-turns at unsignalized intersections.” Journal of Transportation
Engineering, 138(11), 1333-1339.
Manjunatha, P., Vortisch, P. and Mathew, T.V. (2013). “Methodology for the Calibration of
VISSIM in Mixed Traffic.” In 92nd Annual Meeting of the Transportation Research
Board, Washington, DC.
Mathew, T. V., and Radhakrishnan, P. (2010). “Calibration of microsimulation models for
nonlane-based heterogeneous traffic at signalized intersections.” Journal of Urban
Planning and Development, 136(1), 59-66.
Park, B., and Qi., H. (2005). “Development and evaluation of a procedure for the calibration of
simulation models.” Transportation Research Record: Journal of the Transportation
Research Board, No. 1934, 208–217.
PTV. (2006). VISSIM Version 7.0. manual, Planung Transport Verkehr AG, Karlsruhe, Germany.
Qin, Y., Dong, H., Zhang, Q., and Yang, Y. (2016). “Parameter Calibration Method of
Microscopic Traffic Flow Simulation Models based on Orthogonal Genetic Algorithm.”
In DMS, 55-60.
Siddharth, S. M. P., and Ramadurai, G. (2013). “Calibration of VISSIM for Indian heterogeneous
traffic conditions.” Procedia-Social and Behavioral Sciences, 104, 380-389.

12

View publication stats