Comparison of

INTEGRATION, TSIS/CORSIM and WATSim in Replicating Volumes and Speeds

on Three Small Networks

YUHAO WANG M.S. Candidate

and

PANOS D. PREVEDOUROS, Ph.D. Associate Professor

Department of Civil Engineering University of Hawaii at Manoa

2540 Dole Street, 383 Honolulu, HI 96822

tel: 808-956-9698 fax: 808-956-5014

e-mail: pdp@hawaii.edu

Accepted for Presentation at the 1998 Annual Meeting of the Transportation Research Board

and Publication in the Transportation Research Record

July 23, 1997 In revised form: November 5, 1997

2

INTRODUCTION

TSIS/CORSIM and WATSim are microscopic, integrated, simulation models.

INTEGRATION, on the other hand, models only the aggregate speed-volume interactions of traffic

and not the details of a vehicle's lane-changing and car-following behavior; thus, it is commonly

classified as a mesoscopic integrated simulation model. INTEGRATION is a routing-oriented

model for mixed networks (1,2); vehicles’ trip origins, destinations and departure times are

specified external to the model. INTEGRATION has been applied to several projects (e.g., 3,4,5),

and the model itself is under continuous improvement. Version 2.0 (6) was used in our tests.

TSIS/CORSIM (7) is a combination of two microscopic models, NETSIM and FRESIM.

Within the earlier integrated traffic simulation system (TRAF), the freeway/urban street system,

ABSTRACT

INTEGRATION, TSIS/CORSIM, and WATSim can simulate traffic operations on arterials and

freeways in an integrated fashion. This paper presents comparisons using three small networks in

Honolulu for which detailed and simultaneous flow conditions are known from surveillance tapes.

The models produced reasonable and comparable results on most of the tested network links. Of

the three software, only INTEGRATION can simulate U-turns, but it also is least able to model

complex signal operations. TSIS/CORSIM is best at replicating lane-changing behavior, but its

percentile input for off-ramps is both inconvenient and inaccurate. WATSim needed the least

calibration for producing good results, but its animation is inferior and its universal car-following

parameters are undesirable. The experiments also revealed that in no case did the default

parameters offer satisfactory results.

3

simulated with the combination of NETSIM and FRESIM, were composite rather than integrated

networks. A Windows version of TSIS (Traffic Software Integrated System) was developed to

provide an integrated, user-friendly, graphical user interface and environment for executing

CORSIM. The authors participated in the TSIS/CORSIM beta-testing and this paper includes

parts of the tests undertaken (the results are with the version released to the public in June 1997).

WATSim (Wide Area Traffic Simulation) (8), developed by KLD Associates, extends

NETSIM’s applicability to freeway and ramp operations. Features include those in TRAF-

NETSIM plus HOV configurations, light rail vehicles, toll plazas, path tracing, ramp metering,

and real time simulation and animation (8).

This paper presents comparisons of these models based on results from three small

networks. Simulated throughputs were compared with actual volumes and speeds.

DATA DESCRIPTION

Selected specifications for the data are given below.

(1) Freeway volumes were counted from videotapes by using AUTOSCOPE and by manually

checking selected ramps and mainline segments. All freeway data used in the analyses are from

simultaneous and contemporaneous tapes.

(2) Average freeway speed on selected lanes of cross-sections under camera surveillance were

estimated with AUTOSCOPE.

(3) Intersection volumes were obtained in the field.

(4) Signal controls for all intersections are actuated. However, they all were treated as pretimed

by using peak period averages, because the parameters of actuated controllers were not known and

INTEGRATION cannot simulate actuated signals.

4

(5) Since traffic flow changes with space, several 150m link increments upstream of the weaving

sections were modeled instead of a single long freeway mainline link to compare the three models'

results over space.

(6) O-D data are needed to run the INTEGRATION model. The O-D matrices were obtained

from the field volumes and the network geometry.

(7) A 15-minute period (7:30-7:45 AM) was simulated after proper initialization.

CASE STUDY 1: LILIHA ON-RAMP NETWORK

This case examines the models’ ability to simulate a congested on-ramp merge section and

the effect of platooned ramp traffic (Figure 1). Platooned flow on the ramp is caused by a

signalized intersection at the beginning of the ramp.

1A. INTEGRATION

Two 15-minute time periods were used to simulate the 15-minute period traffic flow (7:30-

7:45 AM) in the network; the first was for initialization and the second for drawing the simulation

results. Typical values of the four important model parameters for running INTEGRATION are

listed below.

FREEWAY RAMP STREET

Saturation flow (vphgpl) 2,200 1,900 1,900

Free-flow speed (km/h) 105 65 50

Speed at capacity (km/h) 55 30 25

Jam density (v/km) 125 125 125

The initial simulation throughputs and error percentages are shown in Column 1, Table 1a.

The result was obtained by treating the intersection as a single origin node. Compared with the

5

actual volumes, the simulated throughput of the on-ramp link was much smaller than the actual

volume. The throughputs on the merge link and the downstream mainline link were also small due

to the low on-ramp volume.

A unique feature of INTEGRATION is that the lane alignment and lane connectivity at

any merge or diverge segment is computed endogenously. The results from the previous simulation

suggested that an incorrect lane allocation was defined by the model. Thus, the use of a lane-

striping file (which enables the user to describe lane connectivity) was required for ramp merge

sections. Column 2, Table 1a, shows the simulation result using the lane-striping file in which all

lanes were declared to be of through movement type (code 010). The simulated throughputs on the

merge link and on-ramp link improved.

The signalized intersection was introduced in the next run. Unlike the uniform vehicle

departures from the single origin node, the signal gives rise to platoons on the ramp. The results

from this simulation run are shown in Column 3, Table 1a. The simulated throughputs improved

further, thereby demonstrating the effect of platooning on merging and the value of integrated

simulation.

1B. CORSIM

The network was divided into two parts; the surface street part simulated by NETSIM,

and the freeway part simulated by FRESIM. Simulation results were obtained after initialization

was completed and the network reached equilibrium. The FRESIM result obtained from default

car-following and lane-changing parameters is shown in Column 1, Table 1b. The simulated

volumes on mainline segments were considerably smaller than the actual volumes.

The FRESIM lane capacity is determined by an array of ten car-following parameters.

Given that car-following (headway) differs among drivers, FRESIM defines 10 driver types (each

represented by one the 10 numbers in the array) and uses a decile distribution to represent headway

6

variability. Payne (9) used the 1994 edition of FRESIM to find that the default car-following array

[15 to 6] corresponds to a lane capacity of 2,350 vphpl, and the array [12 to 3] corresponds to

2,750 vphpl, for 3-lane freeway segments. A similar experiment was conducted with the latest

TSIS/CORSIM (7). The freeway default capacity (2,400 vphpl) was similar to Payne's result, but

a higher capacity (3,100 vphpl) was achieved with the car-following array [12 to 3]. The

simulated volumes were comparable to the actual volumes (Column 2, Table 1b) when a 3,100

vphpl capacity was used.

Numerous experiments using different lane-changing parameters were then made. The best

result is listed in Column 3, Table 1b, which was obtained by using a lane change duration of 2.5

sec. instead of the 3.0 sec default. Ten independent replications with different random seeds were

conducted as discussed later.

The intersection signal and the Liliha on-ramp platoon effect on the simulated result was

also examined (Column 4, Table 1b). All freeway parameters remained unchanged, but the

intersection was treated as a single node (only the FRESIM network was modeled). The resultant

throughputs from this run decreased, and the simulated speeds on both the mainline segments and

the on-ramp were lower than those from the integrated network.

1C. WATSim

The Liliha on-ramp network was tested with and without the upstream signalization (Table

1c). The simulated ramp volumes from the two runs (Column 1: signal absent; Column 2: signal

present) were almost the same as the actual ramp volume. The intersection signal and platoon

simulation increased the throughputs on mainline links, but the simulated volumes were still much

lower than the actual volumes.

Freeway capacity can be directly specified in WATSim; it can vary between 1,000 vphpl

and 2,500 vphpl. Column 3, Table 1c, shows that increasing the lane capacity from 2,200 vphpl to

7

2,500 vphpl had no effect on the mainline throughputs. Default lane-changing parameters were

also modified to examine their effect on the mainline throughputs, but no significant improvements

were obtained. The WATSim model developer suggested that different acceleration lane lengths be

tried to obtain the observed mainline throughputs. By changing the acceleration lane length from

65m to 115m and using the 2,300 vphpl capacity, the simulated mainline throughputs became close

to the actual mainline volumes (Column 4, Table 1c).

1D. Comparison of MOEs

Table 1d lists selected MOEs which can be directly compared among the three models.

The MOEs were extracted from the best simulation runs. The three models produced similar

outputs (Figure 1): the speed and density gradually change and reach their respective minimum and

maximum values on the merge link. In WATSim, the speed gradually decreases and the density

gradually increases beginning 300m (Link 5) upstream of the merge link. In CORSIM, this

distance is 450m (Link 4) and frequent lane changes occur during animation from the busy right-

most lane to the less congested left lanes. This was close to what was observed with the

surveillance camera. The speed and density of INTEGRATION began to change only 150m (Link

6) upstream of the merge link. Few lane-changing maneuvers on upstream mainline links were

observed with the INTEGRATION animation which explains the relatively higher simulated

speeds on the upstream links.

Figure 2 shows the three models' detector speed curves for each lane of Link 6, on which

vehicle speed was measured with AUTOSCOPE. The INTEGRATION and WATSim speed

curves are similar to each other. The CORSIM speed on the right-lane exhibited the same trend,

but the speeds on the left- and middle-lane did not change much. The AUTOSCOPE speed,

available only for the right-lane, was 31.4 km/h and the station position was about 45m upstream

of the merge node. Therefore, both CORSIM and WATSim lane speeds are close to the

8

AUTOSCOPE measurement at this specified location, and the INTEGRATION lane speed at the

same location (about 56 km/h) is high.

All CORSIM and WATSim simulation results discussed above were obtained by using the

default random seeds. Ten independent replications with different random seeds also were made

for this as well as the other two case studies. The volume and almost all speed averages differed

by 1.5% or less from those obtained from the default random seed. Sample size estimation based

on estimates of variance showed that the required sample for 90% confidence is 1. For this reason,

as well as the lack of space, only the default random seed results are shown. However, results

from the ten replications (CORSIM and WATSim only) for all case studies can be found in (10).

1E. Lessons Learned

1. The upstream signalized intersection and the traffic platoon on the on-ramp clearly

affect the merge simulation results. All three models are capable of simulating this type of

integrated freeway/surface street traffic network.

2. The use of the lane-striping file for ramp merge simulation is not documented well in the

INTEGRATION manual. To simulate ramp merge sections, the lane connectivity should be

explicitly specified by using the lane-striping file.

3. The default lane capacity of CORSIM is about 2,400 vphpl, but under the default

setting the actual link volumes could not be duplicated. It is rather incredible that the 3,100 vphpl

capacity must be used to simulate the Liliha on-ramp merge, as the last beta version we tested

produced good results with a 2,650 vphpl capacity. The intralink lane-changing logic of the

evaluated CORSIM version (7) has been greatly enhanced in order to realistically duplicate the

intralink lane-changing maneuvers. The enhancement of intralink lane-changing logic is positive,

but more lane-changing turbulence is introduced in the current version than in the previous version.

9

As a result, the default car-following parameters are no longer suitable, and a higher lane capacity

must be used for realistic results.

4. WATSim's ability to directly accept the user-specified freeway capacity is welcome. It

was observed that mainline vehicles have difficulty passing through a merge segment with a short

acceleration length, in which case the length of the acceleration lane must be elongated to obtain

field observed throughputs.

CASE STUDY 2: PALI HWY. OFF-RAMP NETWORK

This network was used to examine the three models' ability to simulate a freeway

divergence situation (Figure 3). The Pali Hwy. off-ramp has enough capacity to accommodate the

vehicles diverging from the freeway mainline with no spillback from this off-ramp observed in the

morning peak period.

2A. INTEGRATION

INTEGRATION’s manual (6) indicates that, for a freeway divergence segment such as

Link 6 in this network, the lanes on the divergence link must be striped to permit the required

straight through and diverging movements even though the model's internal logic would likely

compute identical lane allocations. The result is shown in Column 1, Table 2a. The four

macroscopic traffic flow parameters were the same as for Case Study 1.

The simulated volumes in Column 1, Table 2a, were very close to the actual volumes.

However, there were some problems with the animated displays and the MOE output file. On the

animation screen, no lane-changing maneuvers were observed on the divergence segment (Link 6)

and its upstream mainline links, and no vehicles actually took the right-most diverging lane onto the

off-ramp. All diverging vehicles jumped to the off-ramp from the three through-movement lanes,

10

including the left-most lane. The simulated travel speeds increased rather than decreased when

vehicles reached the divergence segment.

The next simulation experiment used a critical angle (α, in Figure 3) greater than 45o.

Column 2, Table 2a, lists the corresponding throughputs. The animation screen showed that the

vehicles destined to the off-ramp took the deceleration lane and many lane-changing maneuvers

occurred on the mainline segments in order to switch to the off-ramp.

The simulated throughputs are considerably lower than the volumes observed on the

mainline and off-ramp links. Increasing freeway capacity to 2,400 vphpl resulted in simulated

volumes which are close to the actual volumes (Column 3, Table 2a).

2B. CORSIM

Free-flow speeds of 105 km/h for freeway mainline segments and 65 km/h for the off-

ramp, default car-following sensitivity factors, and default lane-changing parameters were

employed in the initial simulation run. The simulated throughputs were much lower than the actual

volumes (Column 1, Table 2b). Throughputs with a capacity of 3,100 vphpl are listed in Column

2, Table 2b. They increased, but the error should be reduced further.

The CORSIM lane-changing parameters were modified to 2.5 sec. for lane changing

duration (3.0 sec. default) and 1,350m for activating distance (750m default). The simulated

throughputs (Column 3, Table 2b) improved.

2C. WATSim

The capacity parameters used in the initial simulation run were 2,200 vphpl for freeway

links and 1,900 vphpl for the off-ramp link. The result of the initial simulation run is shown in

Column 1, Table 2c. All simulated throughputs were much lower than the actual volumes.

11

Increasing only the freeway capacity did not improve the simulated results; Column 2,

Table 2c, lists the result using a 2,500 vphpl capacity. Of the various lane-changing parameters,

only the distance to activate lane-changing maneuvers has a significant effect on the simulation

result. The default distance in the WATSim manual is 135m, but output displays 45m. Both

values are too short for freeway diverge simulation. Column 3, Table 2c, shows the result obtained

by using a 300m activating distance and freeway capacity of 2,300 vphpl, the best combination of

capacity and distance from numerous simulation experiments.

2D. Comparison of MOEs

Table 2d lists selected MOEs obtained from the best simulation runs of the three models.

The models produced comparable MOEs on almost all the links. Figure 3 illustrates the gradually

changing trends of travel speeds and densities along the freeway mainline. These trends conform to

the actual traffic flow situation.

The effect of the distance parameter on MOEs is also reflected in Table 2d. To obtain

throughputs close to the actual volumes, a distance of 1,350m was used in CORSIM to activate the

mandatory lane changes. In WATSim, a 300m activating distance was used. Therefore, diverging

vehicles in CORSIM began changing lanes much farther away from the divergence segment than in

WATSim, resulting in lower travel speeds on these early upstream mainline links such as Links 3

and 4. Unlike CORSIM and WATSim in which the same activating distance is applied to

diverging vehicles on all freeway lanes, INTEGRATION designates staggered activating distances

for vehicles on different freeway lanes based on the minimum number of required lane changes to

complete off-ramp diverging, and on the increasing propensity to complete lane changes as the

diverging vehicle approaches the divergence segment. For this simulation, the activating distance

is about 350m for the middle lane of the 3-lane mainline section, and 750m for the left-most lane.

12

As a result, the INTEGRATION speeds on Links 3 and 4 are between those obtained from

CORSIM and WATSim.

2E. Lessons Learned

1. INTEGRATION can internally compute the lane alignment for freeway diverge

sections, but an exit angle greater than 45o is required if the diverge segments are set in the lane-

striping file.

2. The distance to activate the mandatory lane changes has a significant effect on the

freeway off-ramp diverge simulation. The default distances in both CORSIM and WATSim are

too short for the tested network.

3. The models eventually produced throughputs comparable to real volumes, and their

MOEs were also comparable. The animation mechanisms, however, are different. In

INTEGRATION, some diverging vehicles may come to a complete stop for several seconds at the

divergence position if they cannot find an acceptable headway gap to switch to the off-ramp.

CORSIM lets vehicles which cannot complete lane-changing maneuvers to the off-ramp pass

through the divergence node to the downstream mainline and produces missed vehicle statistics in

the output file. WATSim employs an old lane-changing logic; i.e., vehicles can change lanes only

when entering a new link, therefore, no intralink lane-changing maneuvers were observed on the

freeway links.

CASE STUDY 3: LUNALILO/PUNCHBOWL NETWORK

The network, shown in Figure 4a, consists of two parts. The upper part is a freeway

weaving section including the Lunalilo St. on-ramp and the Vineyard Blvd. off-ramp, and the lower

part is a signalized intersection (Punchbowl St. and Vineyard Blvd.) at which the Vineyard Blvd.

off-ramp terminates. There are two special features in this network. The first is the lane

13

restriction line (the bold line on Links 6 and 7) on the freeway weaving section which prohibits lane

changes from the two left-most lanes to the off-ramp while permitting the crossing of the line from

the right side to the left. The second feature is the U-turn movement allowed at the intersection

from westbound Vineyard (Link 11) to eastbound Vineyard (Link 15). The objective of this

network configuration is to examine the three models' ability to simulate a complex freeway

weaving section together with a complex signalized intersection.

3A. INTEGRATION

The intersection is controlled by an actuated signal. INTEGRATION cannot simulate

actuated signalization, so a pretimed signalization plan with average cycle and phase lengths was

employed (Figure 4a). There is a problem in simulating Phase 3 in which the left-turn movement is

not permitted. INTEGRATION cannot prohibit the left-turn movement in this phase. In addition,

Link 11 is served in three phases, but up to two phases for each link can be modeled.

INTEGRATION internally determines the proper lane alignment between an upstream

link and its downstream link. This is true for a 3-lane link which becomes a 2-lane downstream

link. However a lane alignment problem was found for a 4-lane link (Link 13) or a 5-lane link

(Link 11) which become a 3-lane downstream link. The left-most lane on the downstream, 3-lane

link (Link 15 or 16) is not connected to any through movement lane on the corresponding upstream

link (Link 13 or 11).

To solve the signalization, lane alignment and U-turn issues, a number of link

manipulations were made. Space limitations preclude their inclusion but they are detailed in (10).

Column 1, Table 3a, shows the result for this network. Macroscopic flow parameters

were as in Case Study 1. No special treatment was specified for the lane restriction “barrier” on

the weaving section as INTEGRATION does not provide a direct method to deal with such a

barrier. Compared with the actual volumes, the simulated throughputs on the freeway weaving

14

section and the related links (ramp links and downstream street link) were very low because

weaving vehicles could not complete lane changes smoothly.

Reducing the duration of weaving or lane-changing maneuvers can increase the simulation

outflows (11). INTEGRATION (6) does not provide an entry to modify the weaving maneuver

duration. However, it was found that the jam density has significant effects on lane switching time

and weaving throughputs. Column 2, Table 3a, lists the simulated result of increasing the jam

density from 125 to 140 v/km/l. An improvement was realized, but errors remained large.

An attempt was made to simulate the lane restriction barrier on the freeway weaving

section. A lane-striping file was used in combination with vehicle types assigned to different O-D

pairs. In the O-D file, the vehicles traveling from the freeway origin to the street destinations (i.e.,

via the off-ramp) were assigned a specific vehicle type. In the lane-striping file, a lane restriction

code for this specific vehicle type was assigned to some lanes on freeway links, as shown in Figure

4b. Such a lane restriction code forced the off-ramp destined vehicles to switch lanes to the right-

most lane before entering the weaving section. The simulation result from using the lane-striping

file is shown in Column 3, Table 3a. All simulated throughputs on the freeway mainline and ramp

links were greatly improved and became close to the actual volumes.

3B. CORSIM

The four-phase, pretimed signalization can easily be simulated by CORSIM, and no lane

alignment problem between upstream and downstream links arose. However, as NETSIM cannot

simulate the U-turn vehicles which discharge during the same signal phase as the left-turn vehicles,

U-turning was modeled as left-turning. For the weaving restriction, a lane barrier can be defined in

FRESIM which prevents lane-changing maneuvers between affected lanes in a pair. This

restriction affects both lanes separated by the barrier; but on the subject weaving section, vehicles

are allowed to cross the barrier from the right lane to the left.

15

Column 1, Table 3b, lists the results of the initial CORSIM simulation in which the

default FRESIM car-following sensitivity factors and no barrier were used. The simulated

throughputs on freeway weaving and ramp segments were considerably lower than the actual

volumes.

The car-following parameters for the freeway flow were then modified (array [13 to 4]) for

2,900 vphpl capacity. The simulated volumes on all freeway links increased (Column 2, Table

3b). The on-ramp volume was still low, but a higher-than-actual number of vehicles used the off-

ramp. Since the lane restriction was not modeled, some on-ramp vehicles could not complete lane-

changing to switch into the mainline and went to the off-ramp. Further increasing the freeway

capacity did not improve the results.

Although the FRESIM lane barrier is not identical to the actual lane restriction line on the

weaving section, its effect on the simulation result was examined (Column 3, Table 3b). Compared

with Column 2, the volumes on the weaving section (Link 6 and 7) increased due to the increase of

the on-ramp volume. However, the off-ramp volume decreased and a higher-than-actual number of

vehicles exited the downstream mainline (Link 8) because some mainline vehicles destined for the

off-ramp were prevented by the FRESIM lane barrier from changing to the desired off-ramp.

Different lane change parameters were also tried but had little effect on forcing the off-ramp

destined mainline vehicles to switch to the right-most lane before entering the weaving section.

Thus, the FRESIM lane barrier is not suitable for the actual lane restriction on the subject weaving

section.

The actual volume percentage for the off-ramp movement was 25% (1648/6720) and the

fraction of the downstream mainline volume was 75%. FRESIM can internally compute the trip

interchange fractions based on the turn movement percentages if the actual weaving and non-

weaving volumes (as shown in Figure 4a) are not user-defined. To obtain simulated volumes

which are as close as possible to actual volumes, the actual weaving exchanges were not used in

16

the next run (Column 4, Table 3b). Compared with Column 2, simulated volumes on all freeway

links, except for the off-ramp, increased considerably. Given the 25% off-ramp turn movement,

only 1,311 vph off-ramp volume was obtained from the internally computed weaving O-D table

which also explained the low simulated off-ramp volume of 1,376 vph.

Column 5, Table 3b, indicates the simulation result after distorting the turn movements. A

30% turn percentage for the off-ramp vehicles was used resulting in a 1,596 vph off-ramp volume

from the internally computed O-D trip table. The 2,900 vphpl capacity, 2.0 second lane change

duration, and 1,350m lane change activating distance were employed in this run. The simulated

throughputs of the off-ramp and all other links were close to the actual volumes. Details on user-

defined and internally computed weaving O-D table analyses are included in (10).

17

3C. WATSim

WATSim also cannot simulate the U-turn movement at the Punchbowl/Vineyard

intersection and the lane restriction on the freeway weaving section, hence, the U-turn movement

was modified to a left-turn movement. The initial WATSim simulation used a 2,200 vphpl

freeway capacity, default lane-changing parameters, user-defined freeway weaving interchanges,

and no specification for the lane restriction barrier (Column 1, Table 3c). The simulated

throughputs on the on-ramp and the two weaving links were about 300 vph below the actual

volumes.

The distance over which drivers decide to perform the lane-changing maneuver is an

important parameter. A distance of 135m was explicitly defined and all other parameters remained

unchanged (Column 2, Table 3c). The simulated throughputs on weaving links, on-ramp, and Link

11 increased. This simulation run showed that the default value of this distance parameter is not

135m but 45m as indicated in the output. Due to the lack of a setting for direct simulation of the

lane restriction line on the weaving section, there was no significant improvement when higher

freeway capacities and longer distances for activating lane-changing maneuvers were used.

Turn movement percentages must be input in WATSim for a divergence link (Record Type

21); interchange O-D data can also be defined (Record Type 95/96) if detailed weaving and non-

weaving volumes are known. In previous simulation runs, both 21 and 95/96 were used. Records

95/96 were not used in the final run, such that, for the divergence segment (Link 7), only the

destination volumes or turn movement percentages (off-ramp and downstream mainline) were

specified regardless of the origins (on-ramp or upstream mainline). With the weaving and non-

weaving volumes no longer differentiated, a reduction of traffic flow turbulence on the weaving

resulted. The simulated volumes from this run were close to the actual volumes section (Column 3,

Table 3c).

3D. Comparison of MOEs

18

Tables 3d and 3e list the simulated MOEs on the freeway segments and intersection links.

All three models produced comparable MOEs on the on-ramp and off-ramp links. The similar

trends of MOEs changing along the freeway mainline are illustrated in Figure 5. However, there

are considerable MOE differences on the weaving segments (Link 6 and 7). It was observed from

the surveillance videotape that most weaving vehicles complete the lane-changing maneuvers on the

4-lane weaving segment (Link 6). The minimum speed and the maximum density should therefore

be reached on this link. The CORSIM and WATSim MOEs and CORSIM animation replicated

this field observation (WATSim animation does not display detailed intralink lane changes). In

INTEGRATION animation, however, there were many weaving vehicles which completed lane

changes on the 5-lane weaving segment (Link 7), resulting in the minimum mainline speed on this

link.

Table 3f displays the surveillance detector speeds for each lane on the 5-lane weaving

segment (Link 7). The actual speeds were measured by AUTOSCOPE on the surveillance

videotape. WATSim lane speeds are very close to the actual speeds. The CORSIM speeds are

somewhat lower than, but still comparable to, the AUTOSCOPE lane speeds. The

INTEGRATION lane speeds are clearly low.

The simulated speeds on all intersection links were low, which is consistent with the high

percentage of vehicles experiencing stops at the intersection. Because the U-turn vehicles were

approximated as left-turns in CORSIM and WATSim, their resultant speeds on Link 11 were

somewhat lower than INTEGRATION’s. Generally, the MOEs from CORSIM and WATSim

were closer to each other than to INTEGRATION’s.

3E. Lessons Learned

19

1. INTEGRATION can simulate the U-turn movement, but its ability to simulate complex

signalization is limited. In this case study, two links approaching the same downstream node (the

intersection) were modeled to separately simulate the left-turn and the through movements. This

modification solved several problems simultaneously: three signal phases can serve a link, left-

turns can be prohibited during the through movement phase, and lane alignment from a 4- or 5-lane

link to a 3-lane link is acceptable.

2. The main simulation problem for CORSIM and WATSim was the U-turn movement at

the intersection. The two models do not permit inputs for representing U-turn vehicles which

discharge in the same phase as the left-turn vehicles.

3. There are no explicit ways in all the three models to simulate the lane restriction barrier

on the subject freeway weaving section.

CONCLUSIONS

INTEGRATION, TSIS/CORSIM, and WATSim were applied to three heavily loaded

traffic networks for which exact volumes and speeds (on specific lanes and locations) were known.

The models produced reasonable and comparable simulated results on most of the tested network

links. The experiments also revealed that the main limitation of these models is the large number

of parameters that need to be modified in order to replicate the real traffic conditions. In no case

did the default parameters offer satisfactory results. Parameter calibration can be tedious and

time consuming. It also may raise concerns regarding the validity of “what if” and future scenario

analyses that involve major changes to network characteristics.

Specific strengths and weaknesses of the three software examined are as follows.

INTEGRATION

• The only model which can simulate the U-turn movement among these models.

20

• The optional lane-striping file enhances flexibility in simulating a variety of traffic operations.

• It has a limited ability to simulate signalized intersections.

• The lane-changing mechanism may not reflect (Honolulu) driver behavior and is not user-

adjustable.

TSIS/CORSIM

• It has the most realistic lane-changing maneuvers.

• Oddly, freeway capacities as high as 2,800-3,000 vphpl were used to duplicate the real traffic

conditions.

• As in NCHRP 385 (12), we conclude that the percentile input for off-ramps is inconvenient

and causes difficulty in replicating field conditions.

• Payne et. al’s (9) useful paper on FRESIM capacity settings must be updated because it does

not apply to the released TSIS/CORSIM.

WATSim

• It required the least modification to default parameters to achieve good results.

• Its animation is relatively inferior.

• The globally (both the surface street and freeway) applied NETSIM car-following and lane-

changing parameters is an undesirable feature that is likely to compromise accuracy.

ACKNOWLEDGMENT

Prepared in cooperation with the State of Hawaii Department of Transportation,

Highways Division and U.S. Department of Transportation, Federal Highway

Administration. The contents of this paper reflect the views of the authors, who are

responsible for the facts and accuracy of the data presented herein. The contents do not

21

necessarily reflect the official views or policies of the State of Hawaii DOT or the FHWA.

This paper does not constitute a standard, specification or regulation.

REFERENCES

1. Van Aerde, M., and S. Yager. Dynamic Integrated Freeway/Traffic Signal Networks: Problems

and proposed Solutions. Transportation Research A, Vol. 22A, No. 6, 1988, pp. 435-443.

2. Van Aerde, M., and S. Yager. Dynamic Integrated Freeway/Traffic Signal Networks: A

routing-Based Modeling Approach. Transportation Research A, Vol. 22A, No. 6, 1988, pp. 445-

453.

3. Hellinga, B., and M. Van Aerde. An Overview of a Simulation Study of the Highway 401

Freeway Traffic Management System. Canadian Journal of Civil Engineering, Vol. 21, 1994,

pp. 439-454.

4. Marcus, C.T., and D. Krechmer. The Use of Simulation Models on the Central Artery/Third

Harbor Tunnel Project. In 1995 Vehicle Navigation & Information Systems Conference

Proceedings, pp. 280-285.

5. Gardes, Y., and A.D. May. Simulation of IVHS on the Smart Corridor Using the

INTEGRATION Model: Initial Investigation. PATH Research Report, UCB-ITS-PRR-93-3,

1993.

22

6. Van Aerde and the Transportation Systems Research Group. INTEGRATION - Release 2,

User's Guide. December, 1995.

7. KAMAN Science Corporation. CORSIM User's Guide Version 1.03. June, 1997.

8. KLD Associates, Inc. WATSim Model: User Guide. April, 1996.

9. Payne, H.J., S. Thompson, and G-L. Chang. Calibration of FRESIM to Achieve Desired

Capabilities. Presented at 76th Annual Meeting of the Transportation Research Board,

Washington, D.C., 1997.

10. Wang, Y. Evaluation of Integrated Traffic Simulation Models: CORSIM, INTEGRATION,

and WATSim. Report No. UHM/CE/97-07, Department of Civil Engineering, University of

Hawaii at Manoa, 1997.

11. Van Aerde, M., M. Baker, and J. Stewart. Weaving Capacity Sensitivity analysis Using the

INTEGRATION Model. Presented at 75th Annual Meeting of the Transportation Board,

Washington, D.C., 1996.

12. Roess, R.P. and J.M. Ulerio. Comparison of the 1994 Highway Capacity Manual’s Ramp

Analysis Procedures and the FRESIM Model. NCHRP 385, TRB, National Research Council,

Washington, D.C., 1997.

23

FIGURE 1. Liliha St. on-ramp network, and speed, density results.

Link 1 Link 2 Link 3 Link 4 Link 5 Link 6 Link 7 Link 8

Liliha on-rampH-1 Freeway

5004 vph

1140 vph

850 vph

982 vph

Lilih

a S

t.

Φ1: 50 sec.

Φ2: 20 sec.

Φ3: 60 sec.

55% 45%

7%47%46%

59% 41%

Volume percentage

1326 vph

150 m 150 m. 150 m 150 m 150 m 150 m 65 m 290 m

90 mN

0

10

20

30

40

50

60

70

80

90

100

L i n k 3 L i n k 4 L i n k 5 L i n k 6 L i n k 7( m e r g e )

L i n k 8

Sp

eed

(km

/h)

IN T E G R A T IO NC O R S IMW A T S IM

0

10

20

30

40

50

60

70

80

90

100

L i n k 3 L i n k 4 L i n k 5 L i n k 6 L i n k 7( m e r g e )

L i n k 8

Den

sity

(ve

h/k

m/ln

) IN T E G R A T IO NC O R S IMW A T S IM

24

FIGURE 2. Lane speed on Link 6 of the Liliha St. on-ramp network. a. INTEGRATION speed on each lane

10

20

30

40

50

60

70

80

90

150 m 120 m 90 m 60 m 30 m 0 m(merge)

Detector Posit ion

Sp

eed

(km

/h)

left lane

middle lane

right lane

b. CORSIM speed on each lane

10

20

30

40

50

60

70

80

90

150 m 120 m 90 m 60 m 30 m 0 m(merge)

Detector Posit ion

Sp

eed

(km

/h)

left lane

middle lane

right lane

c. WATSIM speed on each lane

10

20

30

40

50

60

70

80

90

150 m 120 m 90 m 60 m 30 m 0 m(merge)

Detector Posit ion

Sp

eed

(km

/h)

left lane

middle lane

right lane

25

FIGURE 3. Pali Hwy. off-ramp network, and speed, density results.

αH-1 Freeway

Pali Hwy. off-ramp

Link 1 Link 2 Link 3 Link 4 Link 5 Link 6 Link 7150 m 150 m 150 m 150 m 150 m 130 m 300 m

235 m

6328 vph

908 vph

5420 vph

N

30

40

50

60

70

80

90

100

Link 2 Link 3 Link 4 Link 5 Link 6(diverge)

Link 7

Sp

eed

(km

/h)

INTEGRATIONCORSIMW ATSIM

10

15

20

25

30

35

40

45

50

55

Link 2 Link 3 Link 4 Link 5 Link 6(diverge)

Link 7

Den

sity

(ve

h/k

m/ln

)

INTEGRATIONCORSIMW ATSIM

26

FIGURE 4a. Lunalilo/Punchbowl network.

10% 1648 vph

5116 vph

Freeway Interchange

90%

29%71%

Link 1Link 2Link 3Link 4Link 5Link 6Link 7Link 8

H-1 Freeway

Lunalilo on-ramp

Vineyard off-ramp1604 vph

1648 vph

5116 vph

150 m 62 m 206 m 150 m 150 m 150 m 150 m

Vineyard Blvd.

Link 13

Link 11

Link

12

Link

14

Link 15

Link 16

Pun

chbo

wl S

t.

1945 vph

1408

vph

592

vph

1648 vph

200 v

ph

2% 98%

8%68%

24%5% 90% 5%

2%64%34%

Turn percentage

90 m

90 m

60 m

300

m

φ1: 20sec

φ2: 20sec

φ3: 60sec

φ4: 60sec

(=1604x10%+5116x29%)1604 vph

N

150 m

FIGURE 4b. Lane restriction (patterned) on the freeway mainline.

Link 7 Link 6 Link 5 Link 4 Link 3 Link 2Link 8

Lunalilo on-ramp

Vineyard off-ramp

27

FIGURE 5. Speed and density along the freeway mainline (Lunalilo/Punchbowl network).

30

40

50

60

70

80

90

100

Link 8 Link 7 (weave) Link 6 (weave) Link 5 Link 4 Link 3 Link 2 Link 1

Sp

eed

(km

/h)

INTEGRATION

CORSIM

WATSIM

10

15

20

25

30

35

40

45

50

Link 8 Link 7 (weave) Link 6 (weave) Link 5 Link 4 Link 3 Link 2 Link 1

Den

sity

(ve

h/k

m/ln

)

INTEGRATION

CORSIM

WATSIM

28

TABLE 1. Results of the Liliha St. on-ramp network.

a. INTEGRATION Results

Simulation run (1) (2) (3)

Link

Actual volume (vph)

Sim. volume (vph) %

Sim. volume (vph) %

Sim. volume (vph) %

Link 6 (upstream) 5004 5060 1.1 5040 0.7 5016 0.2

Link 7 (merge) 6328 5752 -9.1 6284 -0.7 6320 -0.1

Link 8 (downstream) 6328 5744 -9.2 6232 -1.5 6272 -0.9

On-ramp link 1324 708 -46.5 1268 -4.2 1328 0.3

b. CORSIM Results

Simulation run (1) (2) (3) (4)

Link

Actual volume (vph)

Sim. volume (vph) %

Sim. volume (vph) %

Sim. volume (vph) %

Sim. volume (vph) %

Link 6 (upstream) 5004 4712 -5.8 4921 -1.7 4973 -0.6 4957 -0.9Link 7 (merge) 6328 6009 -5.0 6187 -2.2 6314 -0.2 6258 -1.1Link 8 (downstream) 6328 6017 -4.9 6195 -2.1 6306 -0.3 6274 -0.8On-ramp link 1324 1301 -1.7 1274 -3.8 1333 0.7 1305 -1.4

c. WATSIM Results

Simulation run (1) (2) (3) (4)

Link

Actual volume (vph)

Sim. volume (vph) %

Sim. volume (vph) %

Sim. volume (vph) %

Sim. volume (vph) %

Link 6 (upstream) 5004 3948 -21.1 4213 -15.8 4217 -15.7 5020 0.3Link 7 (merge) 6328 5276 -16.6 5550 -12.3 5550 -12.3 6305 -0.4Link 8 (downstream) 6328 5272 -16.7 5546 -12.4 5554 -12.2 6313 -0.2On-ramp link 1324 1328 0.3 1333 0.7 1368 3.3 1289 -2.6

d. Comparison of MOEs

Link 3 Link 4 Link 5 Link 6 Link 7 Link 8 On-ramp(merge)

Speed (km/h) INTEGRATION 91.0 91.0 91.0 68.0 33.0 68.0 23.0

CORSIM 97.3 61.9 36.2 25.0 23.7 82.7 16.6WATSIM 86.7 85.0 77.8 40.8 35.5 74.7 51.7

Density (veh/km/ln)INTEGRATION 19.8 19.8 19.8 26.4 53.8 34.4 55.0

CORSIM 17.2 26.1 42.5 61.6 83.8 25.4 73.8WATSIM 20.4 20.1 22.4 41.6 45.1 28.4 30.6

29

TABLE 2. Results of the Pali Hwy. off-ramp network.

a. INTEGRATION Results

Simulation run (1) (2) (3)

Link

Actual Volume (vph)

Sim. Volume (vph) %

Sim. Volume (vph) %

Sim. Volume (vph) %

Link 6 (diverge) 6328 6336 0.1 5876 -7.1 6280 -0.8Link 7 (downstream) 5420 5420 0.0 5060 -6.6 5404 -0.3Off-ramp 908 892 -1.8 824 -9.3 888 -2.2

b. CORSIM Results

Simulation run (1) (2) (3)

Link

Actual Volume (vph)

Sim. Volume (vph) %

Sim. Volume (vph) %

Sim. Volume (vph) %

Link 6 (Diverge) 6328 5780 -8.7 6260 -1.1 6340 0.2Link 7 (downstream) 5420 4992 -7.9 5308 -2.1 5468 0.9Off-ramp 908 764 -15.9 884 -2.6 880 -3.1

c. WATSIM Results

Simulation run (1) (2) (3)

Link

Actual Volume (vph)

Sim. Volume (vph) %

Sim. Volume (vph) %

Sim. Volume (vph) %

Link 6 (diverge) 6328 5964 -5.8 6080 -3.9 6316 -0.2Link 7 (downstream) 5420 5172 -4.6 5256 -3.0 5452 0.6Off-ramp 908 796 -12.3 800 -11.9 864 -4.8

d. Comparison of MOEs

Link 2 Link 3 Link 4 Link 5 Link 6 Link 7 Off-ramp(diverge)

Speed (km/h) INTEGRATION 78.1 60.0 60.0 49.0 60.0 84.0 53.0

CORSIM 71.7 55.4 45.6 42.2 58.7 91.7 65.3WATSIM 78.1 73.1 63.2 54.9 61.0 83.7 61.1

Density (veh/km/ln)INTEGRATION 30.8 37.4 39.6 44.0 30.0 22.0 12.6

CORSIM 29.1 37.2 44.9 49.6 35.9 19.8 13.3WATSIM 27.2 29.6 33.9 39.3 26.4 21.8 15.7

30

TABLE 3. Results of the Pali Hwy. off-ramp network.

a. INTEGRATION Results

Simulation Run (1) (2) (3)

Actual Volume (vph)

Sim. Volume (vph) %

Sim. Volume (vph) %

Sim. Volume (vph) %

Freeway Link 5 (upstream) 5116 3918 -23.4 4327 -15.4 5103 -0.3Link 6 (weave) 6720 5441 -19.0 5745 -14.5 6708 -0.2Link 7 (weave) 6720 5456 -18.8 5760 -14.3 6705 -0.2Link 8 (downstream) 5072 4121 -18.8 4342 -14.4 5047 -0.5On-ramp link 1604 1473 -8.2 1413 -11.9 1605 0.1Off-ramp link 1648 1323 -19.7 1398 -15.2 1648 0.0

StreetLink 11 1848 1485 -19.6 1586 -14.2 1844 -0.2Link 12 592 592 0.0 592 0.0 592 0.0Link 13 1945 1810 -6.9 1953 0.4 1953 0.4Link 14 1408 1353 -3.9 1372 -2.6 1372 -2.6

b. CORSIM Results

Simulation Run (1) (2) (3) (4) (5)

Actual Volume (vph)

Sim. Volume (vph) %

Sim. Volume (vph) %

Sim. Volume (vph) %

Sim. Volume (vph) %

Sim. Volume (vph) %

Freeway Link 5 (upstream) 5116 4770 -6.8 5055 -1.2 5108 -0.2 5156 0.8 5108 -0.2Link 6 (weave) 6720 6094 -9.3 6435 -4.2 6686 -0.5 6761 0.6 6686 -0.5Link 7 (weave) 6720 6090 -9.4 6431 -4.3 6694 -0.4 6758 0.6 6690 -0.4Link 8 (downstream) 5072 4598 -9.4 4703 -7.3 5355 5.6 5385 6.2 5040 -0.6On-ramp link 1604 1320 -17.7 1361 -15.1 1601 -0.2 1575 -1.8 1586 -1.1Off-ramp link 1648 1500 -9.0 1748 6.0 1339 -18.8 1376 -16.5 1658 0.6

Street Link 11 1848 1698 -8.1 1950 5.5 1548 -16.2 1590 -14.0 1867 1.0Link 12 592 588 -0.7 596 0.7 585 -1.2 588 -0.7 588 -0.7Link 13 1945 1942 -0.2 1938 -0.4 1931 -0.7 1923 -1.1 1920 -1.3Link 14 1408 1376 -2.3 1376 -2.3 1387 -1.5 1376 -2.3 1383 -1.8

c. WATSIM Results

Simulation Run (1) (2) (3)

Actual Volume (vph)

Sim. Volume (vph) %

Sim. Volume (vph) %

Sim. Volume (vph) %

Freeway Link 5 (upstream) 5116 5130 0.3 5130 0.3 5141 0.5Link 6 (weave) 6720 6412 -4.6 6491 -3.4 6690 -0.4Link 7 (weave) 6720 6416 -4.5 6495 -3.3 6690 -0.4Link 8 (downstream) 5072 4811 -5.1 4863 -4.1 5073 0.0On-ramp link 1604 1316 -18.0 1368 -14.7 1571 -2.1Off-ramp link 1648 1616 -1.9 1635 -0.8 1623 -1.5

StreetLink 11 1848 1788 -3.2 1830 -1.0 1845 -0.2Link 12 592 592 0.0 592 0.0 596 0.7Link 13 1945 1938 -0.4 1946 0.1 1953 0.4Link 14 1408 1365 -3.1 1368 -2.8 1376 -2.3

31

TABLE 3. (continued)

d. Freeway MOEs

Link 8 Link 7 Link 6 Link 5 Link 4 Link 3 Link 2 Link 1 On-Ramp Off-Ramp(weave) (weave)

Speed (km/h)INTEGRATION 78.0 38.0 62.0 68.0 78.1 78.0 91.0 91.0 44.0 54.0

CORSIM 84.2 65.1 34.7 60.6 86.9 89.0 89.0 89.0 21.8 62.9WATSIM 81.0 70.1 52.5 72.8 82.9 85.4 88.2 89.6 35.2 60.8

Density (veh/km/ln)INTEGRATION 24.2 36.3 27.6 26.4 24.2 24.2 22.0 19.8 36.7 14.7

CORSIM 19.8 25.8 47.8 28.2 19.7 19.2 19.3 19.2 73.5 12.8WATSIM 21.8 19.9 31.9 24.1 21.6 20.8 20.2 17.9 46.5 15.2

e. Street MOEs

link 11 link 12 link 13 link 14Speed (km/h)

INTEGRATION 8.6 5.0 5.8 9.9CORSIM 7.5 4.3 6.2 11.5WATSIM 6.1 4.6 5.6 8.6

Density (veh/km/ln)INTEGRATION 44.0 49.5 71.5 67.6

CORSIM 49.5 71.8 78.4 60.4WATSIM 54.1 66.8 81.4 78.4

Total Time (sec/veh)INTEGRATION 40.6 39.0 48.2 105.0

CORSIM 43.3 50.4 51.9 94.1WATSIM 53.7 46.4 57.9 125.7

Avg. Queue (veh)INTEGRATION 18.0 6.0 25.0 37.0

CORSIM 18.0 6.0 21.0 28.0WATSIM 22.0 7.0 23.0 42.0

f. Lane Speed (km/h) on Link 7

Left-through Lane

Middle-through Lane

Right-through lane

Left-exit Lane

Right-exit Lane

Link Average

AUTOSCOPE x 80.0 x 72.0 x xINTEGRATION 41.1 31.2 33.0 34.9 35.7 38.1CORSIM 69.3 60.8 54.9 50.6 53.6 65.1WATSIM 91.4 80.0 67.4 72.5 77.4 70.1