Simulation Modelling Practice and Theory 37 (2013) 18–29

Contents lists available at SciVerse ScienceDirect

Simulation Modelling Practice and Theory

journal homepage: www.elsevier.com/locate/simpat

Comparative study on simulation performances of CORSIM

and VISSIM for urban street network
Daniel(Jian) Sun a,⇑, Lihui Zhang b, Fangxi Chen c
a
State Key Laboratory of Ocean Engineering, School of Naval Architecture Ocean and Civil Engineering, Shanghai Jiao Tong University, No. 800 Dongchuan
Road, Shanghai 200240, China
b
School of Transportation and Logistics, Dalian University of Technology, No. 2 Linggong Road, Ganjinzi District, Dalian 116024, China
c
Education Development and Community Office, Shanghai Jiao Tong University, No. 800 Dongchuan Road, Min-Hang District, Shanghai 200240, China

a r t i c l e i n f o a b s t r a c t

Article history: With the progress of simulation technologies, many transportation simulation packages
Received 2 January 2013 were developed. However, little information is available to the users in applying these
Received in revised form 16 May 2013 models to the most appropriate situations, or even seldom with the simulation accuracy
Accepted 17 May 2013
of the individual model. This study conducts a comparative analysis of two popular simu-
Available online 15 June 2013
lation models (VISSIM and CORSIM), based on their simulation performances on an urban
transportation network. Road network and field traffic data from North Bund, Hongkou
Keywords:
District, Shanghai, China were used as the simulation background and input. Sensitivity
Micro-simulation model
Urban transportation network
analysis was carried out to compare the performance of both models based on four key
Comparative study indices, namely software usability, average control delay, average queuing length, and
Signal intersections cross-sectional traffic volume. Advantages of each simulator were identified based on com-
Sensitivity analysis parison analyses of simulations with different levels of congestion and intersection geospa-
tial scales. The main performance difference was found lying in the default parameter
configuration within the models, including driver behavior settings, traffic environment
settings, and vehicle types, etc. Consequently, it was recommended that analysts should
choose their appropriate tools based on intersection type and level of saturation within
the simulation case.
Ó 2013 Elsevier B.V. All rights reserved.

1. Introduction

As an important branch and application of traffic engineering studies, traffic simulation has become one major tool in
testing the operational performance of transportation facilities and evaluating planning strategies [1,2]. Micro-simulation
packages model individual driver for their behavior and decision making procedure, so as to approximate road traffic con-
ditions. Consequently, they are more applicable on small urban area or special transportation facility studies [3,4]. The tech-
nology has wide applications to provide new solutions to urban transportation problems, because of its extraordinary
advantages in cost effective, safety, reproducibility, and easiness of use. In general, microscopic simulators can be divided
into commercial products and the ones for laboratory research [5]. Of those for commercial purposes or similar, CORSIM,
VISSIM, AIMSUN, and PARAMICS are the most important ones [2,3,5,6]. Particularly, the first two are more widely used in
China, comparing to the latter ones. However, with the increasing diversity of the micro-simulation packages and their per-
formance discrepancy on driver/vehicle behavior modeling, traffic analysts have rather difficulties in choosing an appropri-
ate simulation platform. This study conducts a technical comparative analysis of CORSIM and VISSIM. The comparison is

⇑ Corresponding author. Tel.: +86 13918769316; fax: +86 21 34206674.

E-mail address: jiansun@ufl.edu (Daniel(Jian) Sun).

1569-190X/$ - see front matter Ó 2013 Elsevier B.V. All rights reserved.
http://dx.doi.org/10.1016/j.simpat.2013.05.007
 Daniel(Jian) Sun et al. / Simulation Modelling Practice and Theory 37 (2013) 18–29 19

mainly focused on simulation performance, in terms of four selected key measuring indices (i.e. software usability, average
control delay, average queuing length, and cross-sectional traffic volume), thus to provide guidelines in selecting micro-sim-
ulation models for urban arterials traffic operational analyses.
The remainder of the paper is structured as follows: Section 2 summarizes literatures related to the overall performance
and vehicular behavior within the two micro-simulation models (CORSIM and VISSIM). Next, an experimental case study
was carefully designed in Section 3, and various simulations were conducted with the field data collected from North Bund,
Hongkou District, Shanghai, China. Section 4 presents the comparative analyses and findings based on the results from multi-
ple runs of simulation in CORSIM and VISSIM, in terms of four selected measures of effectiveness (MOEs). Finally, conclusions
and recommendations for future work are provided in Section 5.

2. Literature review on traffic simulation models

As two widely used micro-simulation packages, CORSIM and VISSIM are perhaps more similar than they are different.
Bloomberg and Dale [7] compared the two models on specific measures, such as throughput and intersection level of service
(LOS), and found the predictions from the both models were similar, while both are different from the HCM predictions.
Especially, for congested intersections with complex geometrics, the micro-simulation models are more appropriate than
the HCM methodology. However, only general qualitative comparison was documented in the study, with no detailed quan-
titative impacts provided.
The main differences found between CORSIM and VISSIM are in vehicular and driver behavior, primarily in the car-fol-
lowing and gap acceptance logic [7,8]. As it is widely studied, vehicular movements in the simulation models were decided
by car following model in longitudinal and lane-changing model transversely [9–11]. Car-following models deal with the
time and space relationships of two consecutive vehicles in the same lane, and control the motion (e.g. time headway) of
the lag car [12–14]. In CORSIM, each simulated driver type has an ideal safety distance, and the distances between vehicles
are updated for every simulation time step. In general, aggressive drivers would accept a much smaller gap than their defen-
sive counterparts. On the contrary, VISSIM implemented Widemann psycho-physical car following model (1974 and 2000).
Whenever the lag driver notices the gap is smaller than the psycho (safety) distance, he/she would brake and decelerate. The
speed of the lag vehicle reduces until it becomes lower than that of the lead vehicle. However, due to uncertainty involved in
deciding the exact speed of the lead vehicle, the procedure would last until the gap between the two vehicles attains another
psycho (safety) distance, and then the lag driver starts to accelerate again [15,16].
Lane changing affects the distribution of vehicles across lanes [17–19]. Compared to car-following models, in which the
behavior of the lead vehicle is relatively unaffected by the lag one, the lane changing process depends on many scenarios and
parameters (such as merge, weaving), and hence it is more complex [7]. In CORSIM, the maneuvers are divided into manda-
tory lane changing, discretionary lane changing, and random lane changing [20]. The maneuver-related parameters include
maximum deceleration rate, average lane changing duration, minimal acceptable gap, etc. On the contrary, VISSIM divided
lane changes into necessary lane changing (to reach the next connecting road of a scheduled route) and free lane changing
(because of more room/higher speed). The behavior parameters include the maximum acceptable deceleration for the sub-
ject vehicle and the lag vehicle on the target lane, desired safety distance of the lag vehicle on the target lane (depending on
the speeds of both lag vehicle and subject vehicle).
Other recent comparative studies on microscopic traffic simulation models were mainly focused on the calibration and val-
idation efforts [21,22]. For example, Abbas et al. [21] compared car-following models, including Gazis-Herman-Rothery (GHR),
Gipps, Wiedemann, Intelligent Driver and velocity difference, by using the calibration effort to individual drivers with the nat-
uralistic data, and found Wiedemann model and the velocity difference model showed consistency across a variety of drivers.
Park and Schneeberger [23], Park et al. [24] proposed an experimental design and genetic algorithm (GA) based parameter
calibration procedure for simulation model calibration and validation. By using a coordinated actuated signalized network,
the performance was evaluated by comparing the distribution of simulation outputs with an average value of field data. The
travel time was used for calibration, while the maximum queue length measure was used for validation. The results indi-
cated that the default parameters of both VISSIM and CORSIM simulation models were not able to replicate field travel times.
Only by one (for CORSIM) or two (for VISSIM) trials of calibration, the simulation outputs of the calibrated parameters were
able to replicate field measured validation condition.
Yang et al. [22] calibrated parameters for four car-following models: Newell’s model, Pipes’ model, optimal velocity mod-
el, and the linear GM model with vehicle trajectories provided from NGSIM. However, the emphasis is to demonstrate that
the proposed calibration methods yield consistent results across different car-following models, without touching the appli-
cation of these models in transportation operation and management.
In summary, with respect to microscopic simulation models on urban streets, the effectiveness of operation performance
has not been studied in much detail. Existing research focuses more on calibration effort of theoretical driver behavior mod-
els, without guidelines for real world practical users. This paper focuses more on statistically rigorous comparisons of the
two models, applied to the same network with field data to test roadways and intersections of different functional classifi-
cations. Furthermore, a complete sensitivity analysis of performance measures (e.g., control delay, average queue length,
etc.) was carried out based on varying demand volume levels, thus to provide recommendations in assisting analysts to
choose their appropriate simulation tools.
20 Daniel(Jian) Sun et al. / Simulation Modelling Practice and Theory 37 (2013) 18–29

3. Simulation experimental design and implementation

The simulation background of this study was selected as the North Bund Area, Hongkou District, Shanghai, China. As its
unique history and one of the key areas around Huangpu River, the area is known as the unique shipping service clustering
area, with typical urban transportation characteristics. The daily traffic operation and management are especially complicate
because of the location of seaport for international trade.
The study area is bounded by Haining Road and Zhoujiazui Road from north, East Daming Road from south, Dalian
Road from east, and Wusong Road from west, with the entire area as an irregular rectangle. A sketch of the site is shown
in Fig. 1. Field surveys were conducted to collect simulation input data on April 14, 2011, and May 10, 2011, respec-
tively. Providing the complexity of data requirement, the first survey (on April 14) focused on static transportation infor-
mation, including channelization of the intersections, road segment, bus station and schedules, and the average travel
time between selected OD for calibration purpose. The second survey was mainly focused on real time traffic and signal
information, such as signal timings, intersection throughput, etc. Additionally, at intersection level, the entry link based
delays, maximum and average queue length, and cross-sectional volume for entry links were also collected from the two
studying subjects – Xinjian Road & Tangshan Road Intersection (X & T Intersection), and Wusong Road & Haining Road
Intersection (W & H Intersection).

3.1. Design of simulation experiment

The network was coded both in CORSIM and VISSIM, according to the layout shown in Fig. 2. Simulation runs were per-
formed with the same field survey data, such as road network, signal timings, and demand volumes. By input field survey
data, such as the road network, signal timings, demand volumes, simulations were run with the same input values. The se-
lected key measures of performance were produced and compared, along with their field collected counterparts, so that the
capabilities of two simulators under various situations can be attained.
During the experimental design, in order to reflect the performance of both simulators in a more detailed way, we in-
tended to use four different levels of input volumes. That is, based on the field collected traffic volume from the 13 exits
and entries, sensitivity analyses were conducted for four levels of input volumes, denoted as low volume, off peak volume,
peak volume, and over-saturated volume. Then, different output for the key simulation indices were generated and analyzed.
During the implementation, the field collected off-peak and peak traffic were used, while traffic volumes for the other two
levels were manipulated based on the survey data volumes, by subtracting 300, adding 300 (unit: pcuvph), respectively, as
follows:

 Situation 1: field traffic off-peak volume  300, with a minimum value of zero;
 Situation 2: field traffic off-peak volume;
 Situation 3: field traffic peak volume;
 Situation 4: field traffic peak volume + 300.

Fig. 1. Layout of the study area network in North Bund, Hongkou District, Shanghai, China.
 Daniel(Jian) Sun et al. / Simulation Modelling Practice and Theory 37 (2013) 18–29 21

Situation 3 Situation 4
60 80
70
50
Average Delagy (s)

Average Delagy (s)

60
40
50
VISSIM
30 40 VISSIM
Field Data
30 CORSIM
20 CORSIM
20
10 10
0 0
East West North South East West North South

Situation 1 Situation 2
20 40
18 35
Average Delagy (s)

Average Delagy (s)

16
30
14
12 25
VISSIM
10 VISSIM 20
Field Data
8 CORSIM 15 CORSIM
6
10
4
2 5
0 0
East West North South East West North South
(a)
Situation 1 Situation 2
60 90
80
50
Average Delagy (s)

Average Delagy (s)

70
40 60
50 VISSIM
30 VISSIM
40 Field Data
CORSIM
20 30 CORSIM
20
10
10
0 0
East West North South East West North South

Situation 3 Situation 4
120 250

100
200
Average Delagy (s)

Average Delagy (s)

80
150
VISSIM
60 VISSIM
Field Data
100 CORSIM
40 CORSIM

20 50

0 0
East West North South East West North South
(b)
Fig. 2. Control delays in (a) X & T Intersection, and (b) W & H Intersection.

By randomly setting different simulation seeds, multiple runs of simulation were conducted. The minimal number runs of
simulation was determined as eight, by setting level of confidence as 95%, and confidence interval as 3 times of the sampling
standard deviation. Consequently, ten runs of simulation were carried out for each situation, with the simulation period as
3600 s, and interval as 200 s. For each situation, the average values for the selected key indices were obtained for further
comparative studies.
22 Daniel(Jian) Sun et al. / Simulation Modelling Practice and Theory 37 (2013) 18–29

3.2. Selection of key simulation indices

Considering the objective of the study, along with the availability of various MOEs from each simulator, four indices, for
the comparative analysis, were chosen as software usability plus other three intersection level measures, namely control de-
lay, average queue length, and cross-sectional volume.

3.2.1. Software usability

For the two simulation application software, the content of usability includes network edibility, model calibration effort
and simulation output measurements. Both for beginners and professional simulation engineers, this index has great signif-
icance in saving the practice time acquired for proficiency, improving efficiency, and expediting commercial promotion
progress.

3.2.2. Average control delay

Control delay is an important MOE for efficiency and level of service at urban intersections, which not only reflects the
effectiveness of signal control and traffic designs, but also reports travelers’ perceived impedance and quality of service, en-
ergy consumption and environment protection. Consequently, accurate replicate of control delay is an important task of ur-
ban intersection simulation.

3.2.3. Average queuing length

Queuing at intersections means the decrease of average travel speed, lower transportation network efficiency and level of
service, which indirectly infers the reduction of vehicle travel efficiency and even the loss of economics. Studies on queuing
characteristics, mainly analyzing the queuing vehicle number or length and delay time, can assist to design queuing facility,
evaluate intersection level of service and signal timings, etc. The output of simulation on queuing length includes average
queuing length, maximal queuing length, and queuing vehicle stop times. This study selected the average queuing length
as the key index, as it’s more meaningful comparing to others.

3.2.4. Cross-sectional traffic volume

Volume is one of the fundamental parameters in describing the characteristics of the traffic. Long time continuous obser-
vations on cross-sectional volume can help to understand traffic distribution on time, spatial variation, thus to provide nec-
essary information for transportation planning, and traffic management and control. In addition, volume is a direct index to
evaluate the production of transportation facilities, and the observed volume is the empirical result of demand, capacity and
traffic flow.

3.3. Network simulation and calibration

By completing road network construction, route choice strategies selection, signal phasing and timing configuration, and
bus station and route setting, the background network was built within the two software, respectively. The network was
firstly simulated using peak hour traffic (Situation 3) for model parameter calibration. The purpose is to tune up each effec-
tive parameter based on the field traffic situations, so that the simulation model can replicate the real life, thus to minimize
the difference between simulation output and field measured values [23,24].
The main calibration variables in VISSIM include emergency stop distance, lane change distance, minimal space headway,
average stop distance, and waiting lane changing duration, etc. The minimal space headway was selected as the calibration
parameter in this study, which represents the minimal distance to the lead vehicle in target lane during a lane-changing
maneuver. The default value in VISSIM is 0.5 m, while the field observation found a much larger value. The main model cal-
ibration variables in CORSIM include lane change time, acceptable gap, acceleration/deceleration, vehicle normal decelera-
tion, minimal time headway, free flow speed etc. The minimal time headway was selected for this study, as the default value
of minimal time headway is 1 s. in CORSIM, while the field observation found a much larger one. Calibration scheme for the
selected parameters in both simulators are listed as follows:

 VISSIM Scenario 1: minimal space headway = 0.5 m;

 VISSIM Scenario 2: minimal space headway = 1 m;
 VISSIM Scenario 3: minimal space headway = 2 m;
 CORSIM Scenario 1: minimal time headway = 1 s;
 CORSIM Scenario 2: minimal time headway = 1.5 s;
 CORSIM Scenario 3: minimal time headway = 2 s.

During the calibration, this study used the route average travel time as the control parameter. The field value was col-
lected from the preselected route on May 10, 2011, peak hours (6 pm), starting from Dalian Road & Zhoujiazui Road, to Xinj-
ian Road, East Daming Road, ending at East Daming Road & Wusong Road. The route (about 3.2 km) was intentionally
selected to include major arterials, such as Zhoujiazui Road, Xinjian Road, and East Daming Road etc., to pass typical inter-
sections (total number = 4). Nine runs were driven, with the travel time collected as 100 0100 , 90 5000 , 90 2400 , 90 5600 , 90 1500 , 100 0400 ,
 Daniel(Jian) Sun et al. / Simulation Modelling Practice and Theory 37 (2013) 18–29 23

100 2600 , 90 4500 and 90 0400 , respectively and then the average travel time was calculated as 90 4500 (585 s). The number of runs (9)
2 2
was validated to satisfy the minimal requirement N P 1:96e2 S (N is the minimal num of runs, S2 is the variance calculated from
9 runs, e is the error as the 5% of the average travel time, and the confidence level is set as 95%, corresponding to the critical
value of 1.96).
The initial network was simulated within each software package by using the average travel time as the calibration index.
The objective is to adjust the default value of the selected calibration parameter(s), namely the minimal space headway in

Table 1
Simulation Travel Time (T.T.) from the two simulators.

Simulator Index Scenario 1 Scenario 2 Scenario 3

CORSIM Avg. Route T.T. (sec/veh) 477.48 569.76 665.368
Relative Error (%) 18.38 2.61 13.74
VISSIM Avg. Route T.T. (sec/veh) 499.4 579.9 686.8
Relative Error (%) 14.63 0.87 17.4

Table 2
Comparison of simulated control delays between CORSIM and VISSIM.

Location Mean C. D.(s) St. Dev.(s) F-test T-test Is C. D. Statistically

different?
VISSIM CORSIM VISSIM CORSIM Pooled
X & T Intersection
Situation 1
East 8.7 8.5 0.942 0.771 0.860 1.493 0.520 No
West 10.7 9.2 1.203 0.522 0.927 5.311 3.618 Yes
North 9.6 10.8 0.963 0.837 0.902 1.322 2.974 Yes
South 15.3 12 1.219 2.227 1.795 0.300 4.111 Yes
Situation 2
East 19.3 17.6 1.314 0.818 1.095 2.580 5.516 Yes
West 20.6 18.8 1.512 1.062 1.307 2.028 6.503 Yes
North 16.4 13.2 1.186 1.742 1.490 0.463 9.303 Yes
South 37.7 30.6 0.864 4.724 3.396 0.033 5.334 Yes
Situation 3
East 21.3 23.4 1.631 1.687 1.659 0.934 2.830 Yes
West 27.6 21.1 1.003 3.514 2.584 0.082 5.624 Yes
North 24.3 28.7 2.123 4.678 3.633 0.206 2.708 Yes
South 43.5 46.6 1.558 3.654 2.809 0.182 2.468 Yes
Situation 4
East 38.4 27.1 1.768 3.478 2.759 0.258 9.158 Yes
West 65.3 28.5 1.846 16.430 11.691 0.013 7.039 Yes
North 28.4 28.2 5.778 3.034 4.615 3.626 0.097 No
South 72.5 51.2 1.639 5.238 3.881 0.098 12.27 Yes

H & W Intersection
Situation 1
East 35.7 37.6 2.922 0.771 3.840 14.37 1.092 No
West 48.9 36.2 4.283 5.220 4.775 0.673 5.940 Yes
North 36.2 37.3 3.709 4.817 4.299 0.593 0.559 No
South 24.7 42.8 2.619 17.332 12.395 0.023 3.269 Yes
Situation 2
East 58.4 65.5 2.786 4.917 3.996 0.321 4.553 Yes
West 68.9 79.7 2.491 4.789 3.817 0.271 8.057 Yes
North 68.3 38.0 4.769 6.156 5.506 0.600 13.91 Yes
South 56.3 67.2 3.354 6.404 5.112 0.274 7.415 Yes
Situation 3
East 95.4 85.8 3.041 4.528 3.857 0.451 5.556 Yes
West 72.4 94.5 3.059 4.672 3.949 0.429 12.49 Yes
North 76.4 61.8 3.102 4.423 3.820 0.492 8.528 Yes
South 67.3 87.0 3.102 5.229 4.299 0.352 10.26 Yes
Situation 4
East 100.3 97.1 2.743 5.381 4.271 0.260 1.685 Yes
West 92.5 104.5 2.981 4.596 3.873 0.421 6.940 Yes
North 187.5 214.4 2.922 5.567 4.446 0.276 13.52 Yes
South 80.4 109.8 3.240 4.028 3.656 0.647 18.01 Yes
24 Daniel(Jian) Sun et al. / Simulation Modelling Practice and Theory 37 (2013) 18–29

VISSIM, and the minimal time headway in CORSIM, so that the overall performance of simulation replicates the real situation
(i.e. within ±10% of the field measured values).
The simulated travel time for calibration purpose in both simulators were presented in Tables 1 and 2, respectively. In
CORSIM, the travel time can only be obtained by link, and consequently, travel time for each link was obtained and accumu-
lated along the route. For the default setting (minimal time headway = 1 s), the average route travel time is 477.48 s, with a
relative error of 18.38% compared to field travel time (585 s). By increasing the value to 1.5 s, the overall route travel time
increased to 569.76 s, with the relative error reduced to 2.61%. However, when the value increased to 2 s, the overall route
travel time becomes 665.37 s, with the relative error increased again (13.47%). Consequently, the correct minimal time head-
way was deemed as 1.5 s.
The average route travel time in VISSIM was comparably easy to obtain. The travel times for the road segment included in
Zhoujiazui Road, Xinjian Road and East Daming Road were configured and generated directly after simulation. When using
the default minimal space headway (0.5 m), the average travel time is 499.4 s, with a relative error of 14.64%. By increasing
the value to 1 m, the output total travel time becomes 579.9 s, which is very close to the field value, with a relative error of
only 0.87%. However, when the minimal space headway was set to 2 m, the output total travel time comes to 686.8 s, with a
relative error around 17.4%. Consequently, the minimal space headway was chosen as 1 m.
Occasionally, multiple parameters may have to be calibrated simultaneously. To this end, a regressive-orthogonal design
method is generally adopted for large scale experiments [23]. Traffic flow, capacity, travel time, delay and queue length were
generally included as dependant variables, while the calibration parameters were selected as independence variables, so that
linear regression can be used to estimate the coefficients of the solution. As the calibration criteria (within ±10% of the field
measured average travel time) in this study based on minimal space headway (for VISSIM) and minimal time headway (for
CORSIM) were already met, no further complicated calibration experiment was carried out.

4. Analyses of simulation results

This section analyzes and compares the simulation results from CORSIM and VISSIM. Sensitivity analyses were conducted
to study the simulation results with four levels of input traffic flow, as situations defined in Section 3.1. The performance of
simulation was compared from perspectives of software usability, average control delay, average queuing length, and cross-
sectional traffic volume. Studies on software usability compared entire simulation efforts on network edit, model calibration,
and simulation output indices, while comparisons of the rest three indices were performed quantitatively based on simula-
tion output values.
Considering the last three indices are mainly for urban signal intersections, two typical intersections were chosen from
the study area: Xinjian Road & Tangshan Road Intersection (X & T Intersection), and Wusong Road & Haining Road Intersec-
tion (W & H Intersection). The X & T Intersection lies interior of the study area, with small area and comparative low through-
put (and low saturated), while the W & H Intersection is one major exit/entry of the study area. Both Wusong Road and
Haining Road are major arterials with high traffic volumes, and the intersection also has several main transportation infra-
structures, such as Shanghai Railway Station, and Suzhou River in the vicinity, thus is much larger scaled, with comparable
larger throughput traffic (high saturated as well).

4.1. Comparison analysis of software usability

The comparison of usability is mainly from three aspects, including network construction/edibility, model calibration ef-
fort required, and availability of simulation output MOEs.

4.1.1. Network construction/edibility

4.1.1.1. Network construction/edibility. In VISSIM, the network construction starts from link building, and then connected to
each other. Consequently, the layout of links should follow strictly on the base map. However network construction in COR-
SIM is started from node building, and then connects the nodes with new generated links, which to some extent, reduces the
network editing efforts.

4.1.1.2. Phasing setting. The difference of phasing setting is rather large. VISSIM 4.3 sets the signal controller device first, and
then creates phasing and timing within each controller, and finally deploys the signal controller to the selected link. Signal
phasing in CORSIM is comparable simpler and more straightforward, and all configuration operations are conducted within
the intersection node properties, with a clear phasing selection procedure.

4.1.1.3. Bus station and route setting. In VISSIM, the editing of bus station and route only needs to select route start and end
points, and then add the related route information on the network. However, in CORSIM, the route editing requires to input
all nodes and links passed by. The deployment becomes much more complicated for areas with many public transportation
routes covered.
 Daniel(Jian) Sun et al. / Simulation Modelling Practice and Theory 37 (2013) 18–29 25

Xinjian Road & Tangshan Road Intersection Wusong Road & Haining Road Intersection
60 160
140

Average Delagy (s)

Average Delagy (s)

50
120
40 100
30 80
60
20
40
10 20
0 0
1 2 3 4 1 2 3 4
Situation Situation
VISSIM CORSIM VISSIM CORSIM

Fig. 3. Trends of average delay for CORSIM and VISSIM.

4.1.2. Network calibration effort

The calibration efforts of the two simulators were explained in Section 3.3. The basic procedures are similar, except for the
key calibration parameter selection. In fact, the calibration procedures of almost all mainstream micro-simulation packages,
including PARAMICS, AIMSUN, etc. do not differ too much.

4.1.3. Measures of simulation output

As to the output MOEs of simulation results, both VISSIM and CORSIM include almost all indices that micro-simulation
can output. Moreover, VISSIM provides the GUI animation, along with the simulation procedure, and then collects the pre-
configured indices at the end of simulation. On the contrary, CORSIM generate simulation results first, and then output var-
ious MOEs. The animation procedure of CORSIM simulation can be generated by TRAFVU module. Moreover, the overall sim-
ulation speed of CORSIM is a little faster than that of VISSIM.

4.2. Comparison analysis of average control delay

Delay is one of the most important indices to measure the effectiveness and LOS of urban signal intersections. Various
definitions were provided for the travel time delays of signalized intersections, such as stop delay, control delay etc., in
which only control delay fully accounts for any slow down caused by intersection signal control. Control delay is the vehicle
travel loss time caused by the signal control, calculated as the difference between the intersection pass-by time with a nor-
mal speed and real travel pass-by time [25]. It is defined as summation of stop delay and acceleration/deceleration (start up
loss time) in HCM 2000 [25], and is a more accurate measurement of vehicular delay caused by intersection control.
In VISSIM, the delay is obtained whenever a vehicle is detected by a travel time detector despite of vehicle types. The con-
trol delay in VISSIM is measured from one or multiple travel times, using real travel time to minus ideal travel time (with no
other vehicles and signal control), calculated as the average value of all vehicle delays occurred within one or multiple road
segments or intersections [24].
Starting from Version 5.0, CORSIM reports the control delay (sec/veh) as a new MOE, which is comprised of the following
three components [20]:

 Stopped delay (Ds): the time lost whereas a vehicle is stopped in the queue waiting for green or waiting for its leader to
move forward;
 Delay incurred while a vehicle is decelerating to stop at the stop bar or the end of the queue (Dd); and
 Delay incurred while a vehicle is accelerating to gain its full operating speed after the signal indication turns green (Da).

Consequently, control delay in CORSIM is always smaller than the delay given in Webster equation, because the normal
operating speed of the vehicle is smaller than free flow speed especially when traffic volume is heavy [20].
Fig. 2 shows the comparison of average control delays from X & T Intersection and W & H Intersection, as well as the field
data for validation. It was found that the CORSIM simulation result is closer to the field data in X & T Intersection, which is
comparably lower congested. However, at high congested W & H Intersection (see Fig. 2b), simulation results from VISSIM
are more reliable. Especially, in Situation 4, the huge delay of north entry links is believed due to the reason that the link is
one of the main entries of the study area. In summary of the different situations, VISSIM tends to enlarge the delay of small
intersections (low congested), while CORSIM generally has larger delays in large and high saturated intersections. These
could be related closely to the various vehicle behavior characteristics (such as lane-changing duration, driver types, and
acceleration/deceleration), and the definition of driver behavior within the simulators, as different parameter settings affect
intersection output delays substantially [26,27].
For each situation, the weighted average delay was calculated across all entry links to generate the delay trends for var-
ious levels of traffic volumes, as shown in Fig. 3. It can be figured out that the increasing trend of VISSIM results is closer to
26 Daniel(Jian) Sun et al. / Simulation Modelling Practice and Theory 37 (2013) 18–29

Situation 1 Situation 2
10 12

Queuing Length (# of Veh.)

Queuing Length (# of Veh.)

8 10

8
6
VISSIM
VISSIM 6
Field Data
4 CORSIM
4 CORSIM

2
2

0 0
East West North South East West North South

Situation 3 Situation 4
16 18
Queuing Length (# of Veh.)

Queuing Length (# of Veh.)

14 16

12 14
12
10
VISSIM 10
8 VISSIM
Field Data 8
6 CORSIM
CORSIM 6
4 4
2 2
0 0
East West North South East West North South
(a)
Situation 1 Situation 2
20 25
Queuing Length (# of Veh.)
Queuing Length (# of Veh.)

18
16 20
14
12 15
VISSIM
10 VISSIM
Field Data
8 CORSIM 10
CORSIM
6
4 5
2
0 0
East West North South East West North South

Situation 3 Situation 4
30 80
Queuing Length (# of Veh.)

Queuing Length (# of Veh.)

70
25
60
20
50
VISSIM
15 40 VISSIM
Field Data
30 CORSIM
10 CORSIM
20
5
10
0 0
East West North South East West North South
(b)
Fig. 4. Average queue length in, (a) X & T Intersection, (b) W & H Intersection.

that of the traffic demand than that of CORSIM results. Consequently, it is recommended to use CORSIM in studying delays of
small scale intersections, while VISSIM has more advantages on investigating delays occurred in large intersections.
A two-sided T test was performed to investigate whether the delays from two models are statistically different. First, the
H0 and H1 hypotheses were provided as assume that for given level of confidence a = 0.10 (at 90% confidence level):
 Daniel(Jian) Sun et al. / Simulation Modelling Practice and Theory 37 (2013) 18–29 27

Situation 1 Situation 2
200 450
180 400
Traffic Volume (vhp)

Traffic Volume (vph)

160 350
140 300
120
250 VISSIM
100 VISSIM
200 Field Data
80 CORSIM
150 CORSIM
60
40 100
20 50
0 0
East West North South East West North South

Situation 3 Situation 4
600 600

500 500
Traffic Volume (vph)

Traffic Volume (vph)

400 400
VISSIM
300 300 VISSIM
Field Data
CORSIM
200 CORSIM 200

100 100

0 0
East West North South East West North South
(a)

Situation 1 Situation 2
800 1400

700 1200
Traffic Volume (vph)
Traffic Volume (vhp)

600 1000
500
800 VISSIM
400 VISSIM
600 Field Data
300 CORSIM
CORSIM
400
200
100 200

0 0
East West North South East West North South

Situation 3 Situation 4
1600 1600
1400 1400
Traffic Volume (vph)

Traffic Volume (vph)

1200 1200
1000 1000
VISSIM
800 800 VISSIM
Field Data
600 600 CORSIM
CORSIM
400 400
200 200
0 0
East West North South East West North South
(b)
Fig. 5. Cross-sectional traffic volume for, (a) X & T Intersection, (b) W & H Intersection.

H0. the delays from two models are not statistically different;

H1. the delays from two models are statistically different.

The critical T value is calculated l1a ¼ l0:95 ¼ 1:645, that is if the T-test value is larger than 1.645, we should reject H0,
2
and accept the alternative hypothesis that the delays from two software are statistically different. A detailed statistical com-
parison was performed as shown in Table 2. The statistical results indicated that in all cases except for particular entry lanes
28 Daniel(Jian) Sun et al. / Simulation Modelling Practice and Theory 37 (2013) 18–29

(e.g. X & T Intersection, Situation 1 Eastbound traffic, Situation 4 Northbound traffic; H & W Intersection, Situation 1 East-
bound traffic and Northbound traffic), the null hypothesis H0 was rejected, indicating the delays produced by the two sim-
ulation software are statistical different.

4.3. Comparison analysis of average queuing length

This section compares the field measured average queuing length of different situations with the corresponding simula-
tion outputs for the target intersections, thus to evaluates the performances of both simulation packages.
In VISSIM, the queue at intersections is defined from the queue counter location on the entry link to the last vehicle in
queuing state. If queue counters were placed in each lane of entry links, then all queuing vehicles can be recorded, so that
the average queue length can be calculated. In CORSIM, the queue lengths were recorded from the start of simulation, until
the end of each run. The simulated average queue lengths for different situations, along with the field measured value were
compared in Fig. 4.
It can be easily figured out that simulation results from both VISSIM and CORSIM have large difference from the field val-
ues under real traffic demand situation (Situations 2 and 3). The trend of queue length from VISSIM is comparably closer to
the field values. Given the queue length measurement is affected largely by individual intersection properties and instant
traffic flow, rather large fluctuations were found. By studying simulation results from all four situations of traffic volumes,
the queue lengths from VISSIM in total are closer to the field measured values. Consequently, the authors recommend using
VISSIM to simulate average queue length.

4.4. Comparison analysis of cross-sectional traffic volume

The field cross-sectional traffic volumes were counted manually during the data survey. In a similar way, VISSIM simu-
lation obtained the cross-sectional volume of each entry link by setting data collection points at each corresponding link. In
CORSIM, the dispatched vehicle number of each link was recorded, and consequently the cross-sectional traffic volume was
easily obtained by retrieving the vehicle numbers at each entry link. Fig. 5 presents the output cross-sectional traffic volume
for each situation within two intersections.
It was found that under both non-peak and peak traffic conditions (Situations 2 and 3), the output volumes of two sim-
ulation software were close to each other. Further investigations on volume of each direction indicate that the outputs of
CORSIM from both intersections are closer to the field measured values. This may be due to the reason that CORSIM treats
each link as a unit in recording global volume, and consequently has more reliable results. However, in VISSIM the measure-
ment was carried out through placing traffic counters manually, which may introduces errors because of inappropriate plac-
ing location or devoid of traffic counters. Consequently, the authors recommend using CORSIM to obtain cross-sectional
traffic volume for urban intersections.
In terms of level of congestions, differences between cross-sectional volumes under Situations 3 and 4 are small. This may
be due to the reason that both situations have traffic volume close to saturated condition, and when the traffic demand at-
tains intersection capacity, the throughout during unit period does not change much. For low congested Situations 1 and 2,
the trends of outputs of CORSIM are closer to the trends of traffic demand variation.

5. Conclusions and recommendations

In this research, we studied the two popular micro-simulation packages, VISSIM and CORSIM, and particularly compared
simulation performance of the two micro-simulation software (CORSIM and VISSIM) based on the urban network in North
Bund area, Hongkou District, Shanghai, China. Conclusions were drawn in terms of four selected key indices as follows:

1. For the software usability, the network editing and signal configuration in CORSIM are easier. Due to different simulation
mechanism, simulation results of CORISM can be output directly and fast, while VISSIM provides friendly configuration
interface with a separated output file for simulation results. For calibration effort, both software provide multiple calibra-
tion parameters to enable the simulated network replicating the real situation. In summary, operations in CORSIM are
comparably more convenient, while VISSIM provides versatile indices for direct output and is believed to be more appro-
priate for the beginners.
2. For the intersection average control delay, VISSIM is more appropriate for large intersections with high throughput traffic,
while CORSIM is good at modeling unsaturated intersections. For example, in W & H Intersection, simulation results of
VISSIM are closer to the field measured values, and also change closely related to the input traffic demands. CORSIM is
more accurate in replicating small scale intersections (intersected by 2 s arterials), which is mainly caused by the prin-
ciple of simulation models and default parameter settings.
3. For the average queue length, both software packages do not perform well, which may be caused by the large fluctuations
existed in the field situation. Comparably, VISSIM has closer simulation results to the real situation. However, it was
believed this index can only be used as a reference index in the comparative analysis study.
4. For the cross-sectional traffic volume, both simulation outputs are close to the field data, while CORSIM is slightly better.
 Daniel(Jian) Sun et al. / Simulation Modelling Practice and Theory 37 (2013) 18–29 29

Based on the analyses in terms of different key indices, under different levels of congestion, and for different scales of
intersections, it was concluded each simulator has its particular advantage in replicating real traffic. The main reason lies
in the embedded simulation models and default parameter configuration, including driver behavior setting, traffic environ-
ment setting, and vehicle type, etc. Consequently, researchers should choose their appropriate simulation tools based on
intersection type and level of saturation. In particular, the calibration effort should be closely related to the simulation objec-
tives, namely the four indices mentioned above. Although the results are promising, additional experiments should be con-
ducted to improve model performance.
First, the model calibrations in this study were performed based on urban arterial empirical data using one single param-
eter (minimal space headway for CORSIM, and minimal time headway for VISSIM). As a follow-up study, the calibration can
be extended to large scale multiple parameters calibration, in which the orthogonal design is generally adopted. A regression
equation may be constructed using the calibration parameters as independent variables, while various attributes, such as
volume, capacity, travel time, delay or queue length as dependant variables. Second, in addition to the selected key indices
(e.g. delay, queue length and cross-sectional volume), other MOEs, such as total stops time, network average speed may be
included. Finally, the main comparisons in this study were focused on urban intersections. Other facilities, such as unsignal-
ized intersections or urban streets may also be investigated, which would form an important extension to the current
research.

Acknowledgement

This research was sponsored in part by the National Natural Science Foundation of China (No. 71101109), the Shanghai
Pujiang Program (No. 12PJ1404600), and the Youth Fund of State Key Laboratory of Ocean Engineering (No. GKZD010059-
29). The authors would like to express their appreciation to Mr. Yizheng Wu from School of Transportation Engineering, Bei-
jing Jiaotong University, China, for his valuable effort and assistance in the data collection and analysis during this study.

References

[1] Y. Zhang, L.E. Owen, J.E. Clark, Multi-regime Approach for Microscopic Traffic Simulation, Transportation Research Record 1631, Transportation
Research Board, Washington, DC, 2001, pp. 103–112.
[2] M. Fellendorf, P. Vortisch, Microscopic traffic flow simulator VISSIM, Fundamentals of Traffic Simulation-International Series in Operations Research &
Management Science 145 (2010) 63–93.
[3] P. Hidas, A functional evaluation of the AIMSUN PARAMICS, and VISSIM micro-simulation models, Road and Transport Research 14 (4) (2005) 45–59.
[4] D.J. Sun, L. Elefteriadou, Research and Implementation of Lane-changing Model Based on Driver Behavior, Transportation Research Record 2161,
Transportation Research Board, Washington, DC, 2010, pp. 1–10.
[5] J. Barcelo, J. Casas, J.L. Ferrer, D. Garcia, Modeling advanced transport telemetric applications with microscopic simulations: the case of Aimsun2, in:
The 10th European Simulation Symposium Proceedings, Nottingham, England, 1998, pp. 362–367.
[6] G.D.B. Cameron, G.L.D. Duncan, PARAMICS – parallel microscopic simulation of road traffic, The Journal of Supercomputing 10 (1996) 25–53.
[7] L. Bloomberg, J. Dale, Comparison of VISSIM and CORSIM Traffic Simulation Models on a Congested Network, Transportation Research Record 1727,
Transportation Research Board, Washington, DC, 2000, pp. 52–60.
[8] N.T. Ratrout, A comparative analysis of currently used microscopic and macroscopic traffic simulation software, The Arabian Journal for Science and
Engineering 34 (1B) (2009) 121–133.
[9] Q. Yang, H.N. Koutsopoulos, A microscopic traffic simulation for evaluation of dynamic traffic management systems, Transportation Research Part C 4
(2) (1996) 113–129.
[10] S. Panwai, H. Dia, Comparative evaluation of microscopic car-following behavior, IEEE Transactions on Intelligent Transportation Systems 6 (3) (2005)
314–325.
[11] A. Kesting, M. Treiber, D. Helbing, General Lane-Changing Model MOBIL for Car-Following Models, Transportation Research Record 1999,
Transportation Research Board, Washington DC, 2007, pp. 86–94.
[12] L.A. Pipes, An operational analysis of traffic dynamics, Journal of Applied Physics 24 (1953) 274–281.
[13] G. Newell, Nonlinear effects in the dynamics of car following, Operations Research 9 (1961) 209–229.
[14] D. Gazis, R. Herman, R. Rothery, Nonlinear follow-the-leader models of traffic flow, Operations Research 9 (1961) 545–567.
[15] R. Wiedemann, Simulation des Verkehrsflusses, Schriftenreihe des Instituts für Verkehrswesen, Heft 8, Universität (TU) Karlsruhe, Germany, 1974.
[16] PTV, VISSIM User Manual (5.0 ed.), Planung Transport Verkehr AG, Karlsruhe, Germany, 2007.
[17] J. Rorbech, Multilane Traffic Flow Process: Evaluation of Queuing and Lane Changing Patterns, Transportation Research Record 596, Transportation
Research Board, Washington, DC, 1976, pp. 22–29.
[18] M. Brackstone, M. McDonald, J. Wu, Lane changing on the motorway: factors effecting its occurrence, and their implications, in: Proc. of the 9th Int.
Conf. on Road Transport Information and Control, London, UK, 1998, pp. 160–164.
[19] D.J. Sun, A. Kondyli, Modeling vehicle interactions during lane changes on arterial streets, Computer-Aid Civil and Infrastructure Engineering 25 (8)
(2010) 557–571.
[20] FHWA, CORSIM User Manual (1.04 ed.), Federal Highway Administration, U.S. Department of Transportation, McLean, Virginia, 1998.
[21] M.M. Abbas, B. Higgs, Z. Adam, A.M. Flintsch, Comparison of car-following models when calibrated to individual drivers using naturalistic data, in: TRB
90th Annual Meeting Compendium of Papers (CDROM), Transportation Research Board, Washington, DC, 2011.
[22] H. Yang, Q.J. Gan, W.-L. Jin, Calibration of a family of car-following models with retarded linear regression methods, in: TRB 90th Annual Meeting
Compendium of Papers (CDROM), Transportation Research Board, Washington, DC, 2011.
[23] B. Park, J.D. Schneeberger, Microscopic Simulation Model Calibration and Validation: A Case Study of VISSIM for a Coordinated Actuated Signal System,
Transportation Research Record 1856, Transportation Research Board, Washington, DC, 2003, pp. 185–192.
[24] B. Park, J. Won, I. Yun, Application of Microscopic Simulation Model Calibration and Validation Procedure: Case Study of Coordinated Actuated Signal
System, Transportation Research Record 1978, Transportation Research Board, Washington, DC, 2006, pp. 113–122.
[25] HCM, Highway Capacity Manual, Transportation Research Board, Washington, DC, 2000.
[26] J. Sun, L. Zhang, Vehicle actuation based short term traffic flow prediction model for signalized intersections, Journal of Central South University
(English version) 19 (1) (2012) 287–298.
[27] A. Kondyli, I. Soria, A. Duret, L. Elefteriadou, Sensitivity analysis of CORSIM with respect to the process of freeway flow breakdown at bottleneck
locations, Simulation Modeling Practice and Theory 22 (2012) 197–206.