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

net/publication/281290567

Methodology for the Calibration of VISSIM in Mixed Traffic

Conference Paper · January 2013

CITATIONS READS
70 2,755

3 authors, including:

Peter Vortisch
Karlsruhe Institute of Technology
169 PUBLICATIONS   1,774 CITATIONS   

SEE PROFILE

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

Virtual Reality and Pedestrian Simulation View project

Understanding the conditions of acceptability of walk-sharing View project

All content following this page was uploaded by Peter Vortisch on 28 August 2015.

The user has requested enhancement of the downloaded file.

 Methodology for the Calibration of VISSIM in Mixed Traffic

Pruthvi Manjunatha
IIT Bombay
Mumbai – 400 076, India
email: pruthvim.pro@gmail.com

Peter Vortisch
Karlsruhe Institute of Technology
Karlsruhe-76131, Germany
Email: peter.vortisch@kit.edu

Tom V Mathew (Corresponding Author)

IIT Bombay
Mumbai – 400 076, India
email: tvm@civil.iitb.ac.in

4293 Words + (4 Figures + 3 Tables)*250 = 6043 Words

1stAugust, 2012

TRB 2013 Annual Meeting Paper revised from original submittal.

 Manjunatha, Mathew, Vortisch 1

1 ABSTRACT
2 Mixed traffic, characterized by diverse vehicles, changing composition, lack of lane discipline, etc. is best modeled
3 by micro simulation. However, the majority of the leading micro simulation packages and their calibration
4 methodologies have been developed considering less complex homogeneous traffic. Hence, a methodology for
5 calibrating a micro simulation model for mixed traffic is proposed. Driver behavior in mixed traffic is observed and
6 adjustments were made to represent in the simulation. Calibration parameters were identified using multi parameter
7 sensitivity analysis, and the optimum values for these parameters were obtained by minimizing the error between the
8 simulated and field delay using a genetic algorithm. Multiple criteria were included in the optimization formulation
9 by constraint insertion. The proposed methodology is illustrated using VISSIM, a widely used micro simulation
10 software. Signalized intersections with different traffic characteristics from Mumbai are taken as case study.
11
12 INTRODUCTION
13 Traffic micro simulation models are quickly becoming the standard for the evaluation and development of road
14 traffic management and control systems worldwide. But skeptics often view simulation modeling as an inexact
15 science at best and an unreliable “black-box” technology at worst (1). This skepticism usually results from
16 unrealistic expectations of the capabilities of simulation models and use of poorly calibrated models. Due to the
17 complexities of mixed traffic, not only are there difficulties in collecting and analyzing real world data but also there
18 is a lack of readily available procedures to calibrate simulation models. It is a challenge to understand driver
19 behavior in heterogeneous traffic and calibrate the micro simulatorsaccordingly. The need for a reliable, calibrated
20 model to asses various traffic operations and control in the heterogeneous traffic was the driving force behind this
21 study.
22
23 The heterogeneous traffic is characterized by a mix of vehicles having diverse static (length, width, etc.) and
24 dynamic (acceleration/deceleration, speed, etc.) properties. These vehicles include nonconventional motorized as
25 well asnonmotorized vehicles, and their composition is highly transient. Another distinguishing aspect of such
26 traffic is the absence of lane marking and lane discipline resulting in complex movement of vehicles especially at
27 intersections. Often, the lane widths are not uniform. Characteristics and modeling issues of such traffic are well
28 documented in the literature (2).
29
30 Most early literature of calibration of micro simulators were primarily confined to homogeneous traffic conditions
31 where good lane discipline is observed. Earlier, researchers informally calibrated simulation models and often used
32 default parameters resulting in large errors (3). Realizing its importance, Hellinga (1), Cohen (4), Dowling et al., (5),
33 Zhang et al., (6) and others suggested general methodologies and techniques for calibration. Along with a set of
34 guidelines Milam et al.,(7) implemented the calibration methodology in CORSIM. Toledo et al., (8) used O-D flow
35 data to calibrate MITSIMLab.
36
37 Among the studies on calibration of VISSIM in particular, Fellendorfand Vortisch (9) provided a discussion of the
38 car following and driver behavior logic that is incorporated in the VISSIM. The paper included a detailed analysis
39 of the Wiedemann driver behavior model implemented in the VISSIM. Park and Schneeberger (3) used Latin
40 Hypercube sampling along with a linear regression model to generate scenarios and solve to match the travel times
41 in field and simulation.
42
43 Parameter optimization is one of the important calibration techniques, and various algorithms have been applied for
44 solving this problem by obtaining the optimal values for the parameter sets used in calibration. Genetic algorithms
45 (GA) are widely used for solving a wide variety of optimization problems including those related to the calibration
46 of simulation models. GA was explored to determine a suitable combination of parameter values for PARAMICS by
47 Lee et al. (10). Kim and Rilett (11) illustrated a GA-based approach to traffic simulation model calibration using ITS
48 data. Two microscopic simulation models, CORSIM and TRANSIMS, were calibrated for two freeway segments in
49 Houston, Texas. Menenni at. al. (12) use a genetic algorithm to calibrate VISSIM based on speed-flow-data. Park
50 and Qi (13) proposed a general methodology to calibrate micro simulation models for isolated intersections using
51 VISSIM as well as CORSIM as an example. In this study, Latin Hyper Cube method was used to generate scenarios
52 and the solution parameter set was obtained by use of GA. The same methodology was tested on a large scale
53 network of co-ordinated signals by Park et al., (14). Mathew andRadhakrishnan (15), made a studyon calibration of
54 VISSIM with respect to heterogeneous traffic. A sensitivity analysis was conducted to select parameters and their
55 ranges. An optimization formulation was introduced to find a solution parameter set so as to minimize the
56 intersection delay. Both Wiedemann 74 and 99 models were calibrated at three intersections.

TRB 2013 Annual Meeting Paper revised from original submittal.

 Manjunatha, Mathew, Vortisch 2

57 Lownes and Machemehl (16) considered link capacity as measure of effectiveness and conducted experiments with
58 six different combinations of car following parameters of the Wiedemann-99 model in VISSIM. Two way Analysis
59 of Variance (ANOVA) was conducted to find the individual parameters producing significant effect on the value of
60 the capacity as well as the interaction among the parameters themselves.Most of the earlier calibration studies use a
61 single measure of performance for the sake of simplicity. Recently, multi criteria approaches were adopted by
62 Duong et al., (17) in VISSIM and Park and Kwak (18) in TRANSIMS respectively.
63
64 Traffic representation is considered important in literature and a visual check is generally suggested after the
65 calibration process. The present study involves the identification and observation of unique features of mixed traffic
66 and techniques to incorporate them in VISSIM. In principle, the work done in this study can be transferred toother
67 micro simulators, as long as they support a continuous lateral movement of vehicles and support the calibration
68 methods by the appropriate user interface, measures of performance output and the ability to interface and interact
69 with an external application.
70
71 METHODOLOGY
72 Calibration of a micro simulation model for mixed traffic requires special procedures to address the unique
73 characteristics of such traffic.Accordingly a methodology is proposed which includes representation of vehicles,
74 geometry and traffic, followed by identification of calibration parameters by multi parameter sensitivity analysis,
75 setting their ranges heuristically and determining the parameter values by an optimization model. The traffic
76 representation addresses several distinct features of mixed traffic. The proposed methodology is presented in the
77 form a flow chart in Figure 1.First, to ascertain the need for calibration, the model is to be simulated with the default
78 setting (pre calibration) and the delay values are to be obtained. These values are compared with field values; if the
79 error is insignificant (which is generally unlikely), then the model with default settings can be adopted without any
80 further calibration. If the error is significant, the calibration steps are to be followed.
81
82 Vehicle Representation
83 Simulation models typically come with a set of standard types of vehicles such as car, bus, truck, and motorcycle.
84 However, in the case of mixed traffic, several non standard vehicle types such as motorized and non motorized three
85 wheelers exist and these can significantly affect the simulation results. Therefore, the first step in the simulation is to
86 accurately define the static and dynamic characteristics of every vehicle type in terms of length, width, acceleration
87 and deceleration, and speed ranges. For a more accurate modeling, one may consider parameters such as axle
88 configuration, turning radius, etc., for each vehicle type.
89
90 Geometric Representation
91 The next step is to represent the intersection accurately. The foremost aspect is to accurately represent the geometry
92 defined by the number of approaches, width of each approach, and turning space, the space occupied by each turning
93 movement in the intersection. The second aspect is the representation of the signal control system. This is
94 comparatively easier since most of the signal systems are similar to, or a simplified form of, the signal systems for
95 homogeneous traffic. The representation includes the cycle time, green time, and red time for each movement group,
96 amber time, and phase sequence.
97
98 Traffic representation
99 This phase involves identifying the local characteristics of the traffic and fine tuning the elements of networks so
100 that the traffic in the simulation behaves similar to the one in the reality. One can observe different additional
101 movements in terms of lane changes, smaller vehicles seeping through etc. The available parameters in the
102 simulation model may not be sufficient to replicate certain special movements by the vehicles in mixed traffic, but
103 depending on the flexibilityof network modeling, one can try to bring the behavior in the simulation as close as
104 possible to reality.
105
106 Selection of parameters and their range
107 Once the vehicle types and intersections are modeled, the next step is the identification of the calibration parameters.
108 A multi-parameter sensitivity analysis is conducted with link capacity being the measure of sensitivity. Pairs of car-
109 following parameters are considered and a two-way ANOVA is done to check for the significant parameters as well
110 as significant interaction within the parameters.

TRB 2013 Annual Meeting Paper revised from original submittal.

 Manjunatha, Mathew, Vortisch 3

111
112 FIGURE 1: Proposed methodology
113 Each parameter needs a lower and upper bound for the value it can takeso that the optimization model needs to
114 search in lesser spacewhich makes the procedure computationally efficient. Such arange is necessary to ensure
115 realistic performance of the simulation model. A heuristic method is proposed to determine the range ofparameters
116 for the subsequent use by optimization algorithm. If a parameter has a positive effect on delay, then thatvalue is
117 increased until the delay reaches aprespecified upperbound; this value is set as the upper limit for the parameter. In
118 asimilar manner, the value of this parameter is reduced in steps,until the delay reaches the lower limit. On the other
119 hand, if theparameter has negative influence on the delay, then an oppositeprocedure is adopted to get the lower and
120 upper limits. This iterative process is continued, and a range is fixed. While doing this, care should be taken to
121 seethat the parameter limits do not result in unrealistic behavior (vehicle jumping red, for instance, vehicles stopping
122 perpendicular to each other during red time, unrealistic long queues). During this process, if the searchis able to
123 yield delay values within the precision set, then calibration of the parameters can be considered completed.
124
125 Measures of Effectiveness and Parametric Optimization
126 The measures of effectiveness can be any number of, and any combination of, travel times, flows, capacities, delay
127 values, and queue lengths. Measures of effectiveness should be selected such that effects on individual parts of the
128 network, such as intersections or ramps, and effects on the entire network are captured. Multiple measures of
129 effectiveness can be adopted by weighting each of them based on certain criteria or making one of them as the
130 objective for optimization and others as constraints (17).The HCM also suggests suitable measures according to the
131 type of facility. For signalized intersection Control Delay is the suggested measure of effectiveness by the HCM.
132
133 In this study, the estimation of the parameter values is formulated as an optimization problem which minimizes the
134 cumulative absolute error between the observed and simulated delays for all the intersections. The link capacities are
135 fixed within a range and they are the measure introduced by constraint insertion technique.
136 The formulation is given in the form of a mathematical program:
137 Minimize abs (1)
138 Subjected to for all j (2)
139 lc< 10 (3)
140 Where, =average delay obtained from the simulation model forintersectioni;

TRB 2013 Annual Meeting Paper revised from original submittal.

 Manjunatha, Mathew, Vortisch 4

141 =observed field delay for intersectioni;

142 and are lower and upper ranges of parameter j; and
143 lc= Percentage error in the estimation of link capacity of the approach under consideration
144 n=number of intersections under consideration.
145
146 According to the terminology of Dowling et al. (5), this formulation can be called a global calibration. Local
147 calibration is a special case of the above formulation when n=1. The different algorithms to search for the parameter
148 set satisfying the conditions in the optimization formulation are iterative averaging, Genetic Algorithm (GA),
149 Simultaneous Perturbation Stochastic Approximation (SPSA) etc. GA is proposed as the solver for the above
150 formulations.
151
152 Calibration
153 First the GA code produces candidate solution sets randomly (initial chromosomes), then these values are fed into
154 simulator via an interaction interface and the feedback is received in the form of Measure of Effectiveness (MOE, in
155 this case is delay, See Figure 2). Based on the feedback, the value of the fitness function is calculated. The new set
156 of chromosomes will produced will depend on the fitness function. After iteration, the value of the fitness function
157 improves till the convergence. More literature on this can be found at Deb (19).
158
159 Validation
160 A new data set corresponding to different traffic and, preferably,geometric conditions should be used for validating
161 the simulationmodel. The absolute error inthe values of the measures of effectiveness is estimated. The model can be
162 used confidently if this error is within certain limits. Field measured and simulated delays are compared and
163 generally a variation of simulation results within 15% of the field measured delay is considered acceptable.
164
165 IMPLEMENTATION
166 The methodology is evaluated by implementation in the traffic simulator VISSIM. The traffic flow model in
167 VISSIM is a discrete, stochastic, time step based microscopic model, with driver-vehicle-units as single entities. The
168 model contains a psycho-physical car following model for longitudinal vehicle movement and a rule-based
169 algorithm for lateral movement. The model is based on the continued work of Wiedemann (20). The basic idea of
170 the Wiedemann model is the assumption that a driver can be in one of four driving modes: Free Driving,
171 Approaching, Following, and Braking. VISSIM implements two variants of this model, the so called Wiedemann-74
172 and Wiedemann-99 models. The models differ in the way they define the thresholds or action-points where the
173 driver changes his driving behavior and in the amount of stochasticity in the driver model. Owing to the advantages
174 and flexibilities Wiedemann-99 model offers, it is used in the present study. Also, due to varying characteristics of
175 different vehicle types in mixed traffic, the values of the car following parameter set are defined separately for each
176 individual vehicle type.
177
178 Apart from car following parameters, the behavior of traffic in simulation of mixed traffic depends mainly on two
179 other factors: Lateral distancesand speed-acceleration profiles of the vehicle types (21). Lateral safety distances are
180 assumed to be linearly dependent on speed and are defined in VISSIM by setting the values for speeds of 0 km/h and
181 50km/h. The lateral distances can be specified separately for any pair of vehicle types. A separate speed-acceleration
182 curve is defined for each vehicle type. Field measurements with utmost care and accuracy are necessary to define
183 these parameters in VISSIM, as a little variation in these can result in a varied output in simulation (Figure 3a and
184 3b).
185

TRB 2013 Annual Meeting Paper revised from original submittal.

 Manjunatha, Mathew, Vortisch 5

186
187
188 FIGURE 2a,2b: Variation of simulation output with change in lateral distances
189
190 Figure 2a shows the variation of simulation output (in terms of delay) with change in lateral distances for two
191 different sets of maneuverability. Maneuverability is indirectly achieved through three VISSIM parameters: the
192 collision time gained by lateral decision, the minimum speed to make that decision and the time between successive
193 direction changes. Figure 2b shows the variation of simulation output (in terms of delay) with change in lateral
194 distances for two different sets of driver aggressiveness. Driver aggressiveness is changed by altering speed-
195 acceleration profiles of the vehicles.In the next section, some of the significant features of the mixed traffic are
196 studied along with the attempts to simulate the same in VISSIM.
197
198 Traffic Representation
199 The unique features of mixed traffic are identified and observed.Some of them are difference in lane behavior,
200 maneuverability of smaller vehicles and stop line violation, ineffective free left turn etc. Techniques to incorporate
201 the mentioned features in micro simulation are also discussed. Although VISSIM was developed for western traffic,
202 the flexibilities provided by the simulator in defining the network elements can be taken advantage of, to simulate
203 the peculiarities of the driver behavior in mixed traffic condition.
204
205 Lane behavior
206 When a vehicle joins the queue, the main stimulus is the front gap irrespective of the lane in which it is available. As
207 a result, it has been observed that straight ahead vehicles occupy any position across the link based on the available
208 space.
209 Unlike lane based system, where lane changing is a continuous process i.e. when a driver needs to change lane in
210 order to overtake a slow leader, he takes lane changing decision only if it is possible to perform complete maneuver
211 in one attempt. On the contrary, in mixed traffic and non-lane based operation as there is no restriction on
212 positioning a vehicle at any place across the link width and therefore the driver who would like to change lane does
213 not bother about completing the maneuver in one attempt. From the field observation it is seen that a driver changes
214 lane gradually according to the available opportunity. Though this behavior is not possible to replicate completely in
215 the simulation, the vehicles can be made to look for newer options quickly using “Collision time gain”, “minimum
216 longitudinal speed” for lateral movement and “time between direction changes” parameters. Though the right side
217 must be used for overtaking according to the traffic rules, vehicles approaching a queue use both left and right sides
218 depending on the available front gaps. VISSIM has an “overtake on both sides” option to simulate the kind of
219 behavior.
220
221 Maneuverability of smaller vehicles and Stop line violation
222 Two-wheelers use inter-vehicular space to come in front of the queue. The “seepage” of two-wheelers can be
223 simulated in VISSIM by giving appropriate lateral distances to be maintained at rest and 50 km/h respectively and
224 setting the parameters collision time gain, minimum longitudinal speed for lateral movement and time between
225 direction changes. The parameter values are changed based on vehicle types. Also, the vehicles stop 2 to 4 meters
226 ahead of the stop line during the red signal. This gives the motivation for two-wheelers to seep through inter-
227 vehicular gaps. This behavior has a significant effect on how the different vehicle types position themselves in the
228 queue. This can be simulated by having two stop lines in VISSIM, the first one only for buses and cars and the
229 second for all vehicle types.
230

TRB 2013 Annual Meeting Paper revised from original submittal.

 Manjunatha, Mathew, Vortisch 6

231 Ineffective free left turn

232 When a “free left turn” is allowed at an intersection the vehicles turning left proceed along their route accordingly.
233 But several seconds into the red phase, through and right turning vehicles exhibit a “greedy” behavior by occupying
234 the lanes reserved for left turning traffic. Several seconds into the red, this behavior makes the concept of “free left
235 turn” ineffective. (The bus in the extreme rightinFigure 3 is a part of through traffic and it has made the free left
236 ineffective).Though there are no specific parameters available to simulate such behavior, using the flexibilities in
237 VISSIM in defining the network elements through links, connectors and stop lines this can be achieved.
238

239
240 FIGURE 3: Ineffective Free Left in Simulation and in Reality
241 Resolution requirement
242 Resolution is the simulation time step; to adequately model the complex movements in mixed traffic, the
243 calculations have to be at smaller time steps. VISSIM offers simulation at different time steps according to the need
244 from 1 time step per second to 10 time steps per second.A resolution of 10 time steps per second is adopted in the
245 case of mixed traffic, which requires more computational effort.
246
247 Data requirements
248 The data required for the proposed methodology are intersection geometry, signal timing and phasing, vehicle types,
249 traffic inflow, proportion of turning traffic, traffic composition, and delay at intersections and link capacity of
250 approach under study. Field delay is measured using the technique recommended by the highway capacity manual
251 (22). In this technique, the vehicles in queue are counted from the start of red until the last vehicle in the queue
252 clears the stop line. The counting is done for intervals of 20 s. The cumulative vehicle-in-queue count is obtained
253 and averaged over the total number of vehicles passing through the intersection during the observation period to get
254 the delay. The link capacity of approach under study is obtained by taking the vehicle count at saturated/over
255 saturated cycles and converting it into hourly flow. While the simulation capacity is estimated using the VISSIM
256 traffic counting tool available in the simulation immediately downstream of the stop line. This definition of capacity
257 is supported by the guidelines adopted by FHWA (5).
258
259 Multi Parameter Sensitivity Analysis
260 Once the network is created and model inputs are provided from the data, the model is run with default parameter
261 set. The default set usually fails to provide results to provide results closer to the reality. Hence the next steps to
262 choose parameters and their ranges for calibration have to be undertaken. Multi parameter Sensitivity Analysis is
263 done in this step. Apart from the individual effects of the driver behavior parameters on the capacity of the link in
264 simulation, their combined effect is also sometimes significant (16). Parameters with Significant effect in current
265 study were found to be CC0, CC1, CC2, CC7 and CC8. There was no significant interaction among each of them at
266 an alpha level of 0.05. After observing the effect of each parameter on the capacity of the link, the selection of
267 parameter and its range is done. A total of 13 variables are used in algorithm.
268
269
270 Tuning of Parameters
271 The interaction module feeds the GA populations into VISSIM simulation via COM interface and sends the received
272 delay output to GA as feedback. The GA parameters like cross over and mutation probabilities are chosen
273 heuristically. A general roulette wheel technique is used in selection of parameters. Currently, VISSIM COM
274 interface does not have access to all network elements, which can limit the range of parameters used in the
275 calibration.

TRB 2013 Annual Meeting Paper revised from original submittal.

 Manjunatha, Mathew, Vortisch 7

276
277 CASE STUDY
278 The methodology proposed in the previous chapter is implemented for two signalized intersections in Mumbai. The
279 aim of this case study is to evaluate the effectiveness of the calibration methodology. The intersection details are
280 collected along with the video; data is extracted from the video later.
281
282 Vehicle Representation
283 VISSIM needs inputs for vehicle characteristics in terms of geometry as well as desired speed, and accelerations
284 (23). The video is seen in frames and while queue discharges or builds up, vehicle trajectories are extracted and
285 speed, acceleration and lateral distances are then calculated. The vehicle characteristics given as an input to VISSIM
286 is listed in Table 1.
287
288 TABLE 1: Vehicle Characteristics and Intersection Details
289 Min. Lateral Distance Traffic
Desired
290 Vehicle Length Breadth (m) Composition(%)
Speed
291 Type (m) (m)
0 km/h 50 km/h (kmph) Powai Vashi
292
293 Two
1.8 0.6 0.3 0.7 35 30 22
294 Wheeler
295 Rickshaw/
2.6 1.4 0.4 0.8 30 11 38
3W
296
297 Car 4.0 1.6 0.5 0.9 40 43 32
298 LCV 5.0 1.9 0.5 0.9 30 6 3
299
Bus/HCV 10.3 2.5 0.6 1.0 35 10 5
300
301 Volume(vph): 16002150
302
303 Analysis of Results
304 The resultant calibrated data set and the convergence of Genetic Algorithmis shown in Table 2 and Figure 4
305 respectively.As the first step of evaluation, the performance of a default and calibrated parameter set needs to be
306 compared. As stated in the earlier, this calibration procedure is necessary when a simulation model with a default
307 parameter set does not reproduce the field condition. For the same intersection in Vashi, the data was collected on a
308 different day and the data was used for validating the calibrated model. The traffic composition and input volume
309 were changed in the simulation input, rest remaining the same. The new data was taken on during less saturated flow
310 and hence the delay time is lesser. The result from calibrated model was found close to the one in the field. (Table 3)
311
312

313
314 FIGURE 4: Convergence of Genetic Algorithm
315
316
317
318

TRB 2013 Annual Meeting Paper revised from original submittal.

 Manjunatha, Mathew, Vortisch 8

319
320 TABLE 2: Calibrated Parameter Set
Parameter Default 2W 3W Car/LCV Bus/HCV
CCO (Standstill distance) 1.50 0.47 1.00 1.20 2.00
CC1 (Headway Time) 0.90 0.43 1.2 0.81 0.94
CC2 (Following Variation) 4.00 4.42 4.42 6.84 6.84
CC3 (Threshold for entering following)* -8.00 -8.00 -8.00 -8.00 -8.00
CC4 (Negative following threshold)* 0.35 0.35 0.35 0.35 0.35
CC5 (Positive following threshold)* -0.35 -0.35 -0.35 -0.35 -0.35
CC6 (Speed dependency of Oscillation)* -11.44 -11.44 -11.44 -11.44 -11.44
CC7(Oscillation acceleration) 0.25 0.20 0.20 0.77 0.77
CC8 (Standstill acceleration) 3.50 2.30 2.30 2.30 2.30
CC9 (Acceleration at 80km/h)* 1.50 1.50 1.50 1.50 1.50
321 *Insignificant parameters left at Default values
322
323 TABLE 3: Results of the Calibration
Observed Absolute
Intersection Type Simulated Delay(s)
Delay(s) Error(s)
Default 130 79
Powai
Globally Calibrated 50 51 1
Default 160 59
Calibration
Globally Data Set 103 101 2
Calibrated Validation
Vashi Data Set 46 51 5
Calibration
Locally Data Set 102 101 1
Calibrated Validation
47 51 4
Data Set
324
325 Global and Local Calibration
326 When the formulation in (1) to (3) is solved for only one network, i.e. when n=1, then it is called local calibration.
327 The modules were adjusted for a single intersection and the Genetic Algorithm code is run again to get a new
328 solution considering only theVashi network. The comparison between globally and locally calibrated models in
329 terms of delay can be seen in Table 4. Though there is a slight improvement in the estimated delay in the locally
330 calibrated model, it is not significant. The delay values of both globally and locally calibrated models and the
331 parameter sets obtained as solutions are similar. This might be due to similarities in the driver behavior at both the
332 intersections.
333
334 CONCLUSIONS
335 The study involved the identification and observation of unique features of mixed traffic and techniques to
336 incorporate them in micro simulation.The multi parameter sensitivity analysis was found to be an effective way of
337 finding the significant parameters and the interactions between them. The methodology was implemented in
338 VISSIM with Genetic Algorithm for solving the optimization formulation and was shown to produce reliable results.
339 The flexibility VISSIM offers in defining the network elements allows the users to model some of the complex
340 maneuvers commonly adopted by the drivers in mixed traffic. The similarities in the results obtained by local and
341 global calibration suggest the similarity in driver behavior at both the intersections and a significant improvement
342 was seen in delay timesfrom that of the default parameter set. Both lateral and longitudinal clear spaces were found
343 to be lower than the default values, reflecting the compactness of the mixed traffic in the intersections under

TRB 2013 Annual Meeting Paper revised from original submittal.

 Manjunatha, Mathew, Vortisch 9

344 study.Beyond the flexibilities that present micro simulators offer, their ability to simulate the mixed traffic is still
345 limited, additional parameters are required to accurately model some of the behavior in mixed traffic.
346
347 The calibration can be further carried out for a corridor and network level; the methodology and measures of
348 effectiveness can be suitably modified.Implementation of the observations of this study can be made in different
349 micro simulators based on their traffic model, user interface, measures of performance output and the ability to
350 interface and interact with an external application. A model representing mixed traffic accurately can be built on the
351 lines of the observations made in the present study and can be either incorporated in an existing simulation package,
352 for example as an external driver behavior model in VISSIM or in a custom-made simulation environment.
353
354 Acknowledgements
355 The authorswould like to thank Lukas Kautzsch, PTVand the German Academic Exchange Programme (DAAD) for
356 supporting the project.
357
358 REFERENCES
359 1.Hellinga, B. R. Requirements for the Calibration of Traffic SimulationModels. Canadian Society of Civil
360 Engineering, University of Waterloo, 1998.
361 2.Arasan, V.T. and Koshy, R.B. Methodology for Modelling Highly Heterogeneous Traffic Flow. Journal of
362 Transportation Engineering, ASCE, vol. 131, No. 7, 2005, pp 544-551.
363 3.Park, B. and Schneeberger, J. D. Microscopic Simulation Model Calibration and Validation Case Study of
364 VISSIM Simulation Model for a Coordinated Actuated Signal System. In Transportation Research Record: Journal
365 of theTransportation Research Board, No. 1856, Transportation Research Board of the NationalAcademies,
366 Washington, D.C., 2003, pp. 185–192.
367 4. Cohen, S. L. An Approach to Calibration and Validation of Traffic Simulation Models. In 83rd Annual Meeting
368 Preprint. CD-ROM, Transportation Research Board of the National Academies, Washington, DC,2004
369 5. Dowling, R., Skabardonis, A., and Alexiadis, V. Calibration ofmicrosimulation models.In Transportation
370 Research Record: Journal of theTransportation Research Board, No. 1876, Transportation Research Board of the
371 NationalAcademies, Washington, D.C., 2004, pp. 1–9.
372 6. Zhang, Y. and L. E. Owen. Systematic Validation of a MicroscopicTraffic Simulation Program. In 83rd Annual
373 Meeting Preprint CD-ROM, Transportation Research Board of the National Academies, Washington, D.C., 2004.
374 7. Milam, R. T., and F. Choa. Recommended Guidelines for the Calibrationand Validation of Traffic Simulation
375 Models. Fehr & Peers AssociatesInc., Roseville, California, 2000.
376 8. Toledo T., Koutsopoulos H.N., Davol A., Ben-Akiva M., Burghout W., Andreasson I., Johansson T. and Lundin
377 C. Calibration and validation of microscopic traffic simulation tools: Stockholm case study.In Transportation
378 Research Record: Journal of theTransportation Research Board, No. 1831, Transportation Research Board of the
379 NationalAcademies, Washington, D.C., 2003, pp. 65–75.
380 9. Fellendorf, M. and Vortisch P. Validation of the Microscopic Traffic Flow model VISSIM in different Real-
381 World Situations. In 80th Annual Meeting Preprint CD-ROM, Transportation Research Board of the National
382 Academies, Washington, D.C., 2001.
383 10. Lee, D. H., Xu, Y., and Chandrasekar, P. Parameter Calibration for PARAMICS Using Genetic Algorithm.
384 In 80th Annual Meeting Preprint CD-ROM, Transportation Research Board of the National Academies,
385 Washington, D.C., 2001.
386 11. Kim, K-O, and Rilett, L.R. A Genetic Algorithm Based Approach to Traffic Micro-simulationCalibration Using
387 ITS Data. In 83rd Annual Meeting Preprint CD-ROM, Transportation Research Board of the National Academies,
388 Washington, D.C., 2004.
389 12.Menneni, S., Sun, C., Vortisch, P. Microsimulation Calibration Using Speed Flow Relationships. In 87th Annual
390 Meeting Preprint CD-ROM, Transportation Research Board of the National Academies, Washington, D.C., 2008.
391 13. Park, B. and Qi, H. Development and Evaluation of a Procedure for the Calibration of Simulation Models. In
392 Transportation Research Record: Journal of theTransportation Research Board, No. 1934, Transportation Research
393 Board of the NationalAcademies, Washington, D.C., 2005, pp. 208–217.
394 14. Park, B., Jongsun, W., and Ilsoo, Y. Application of Microscopic Simulation Model Calibration and Validation
395 Procedure: A Case Study of Coordinated Actuated signal System.In Transportation Research Record: Journal of
396 theTransportation Research Board, No. 1978, Transportation Research Board of the NationalAcademies,
397 Washington, D.C., 2006, pp. 113–122.

TRB 2013 Annual Meeting Paper revised from original submittal.

 Manjunatha, Mathew, Vortisch 10

398 15. Mathew, T.V. and Radhakrishnan, P. Calibration of Microsimulation Models for Nonlane-Based Heterogeneous
399 Traffic at Signalized Intersections.Journal of Urban Planning and Development, ASCE, Vol. 136, No.1, 2010, pp.
400 59-66.
401 16. Lownes, N. E., & Machemehl, R. B. VISSIM: A Multi-Paramater Sensitivity Analysis. Proceedings of the
402 38th Winter Simulation Conference, 2006, pp. 1406-1413.
403 17. Duong,D., Saccomanno,F., and Hellinga,B. Calibration of traffic model for simulating safety performanceIn
404 89th Annual Meeting Preprint CD-ROM, Transportation Research Board of the National Academies, Washington,
405 D.C., 2010.
406 18. Park, B and Kwak, J. Calibration and validation of TRANSIMS microsimulator for an urban arterial
407 network.KSCE Journal of Civil Engineering, Vol 15, No 6, 2011, pp 1091-1100
408 19. Deb, K. Optimization for engineering design: Algorithms and examples. Prentice-Hall, New Delhi, 1995.
409 20. Wiedemann, R. Modeling of RTI-Elements on Multi-lane Roads”. In Advanced Telematics in Road Transport,
410 edited by the Commission of the European Community, DG XIII, Brussels, 1991.
411 21. Jie, L., Zuylen, H. J., Shoufeng L. Calibration of a Traffic Simulation for a Chinese city, Proceedings of
412 2ndInternational Conference on Models and Technologies for Intelligent Transportation Systems, Leuven, Belgium,
413 2011.
414 22. TRB. The highway capacity manual, In Transportation Research Record. 209, Transportation Research Board,
415 Washington, D.C., 2000.
416 23. PTV.VISSIM 5.40 User Manual. Karlsruhe, Germany, 2011.

TRB 2013 Annual Meeting Paper revised from original submittal.

View publication stats