A Comparison of the VISSIM and CORSIM Traffic Simulation Models On A Congested Network

Traffic simulation packages like CORSIM and VISSIM are frequently used as tools for analyzing traffic, since they are an effective approach for quantifying the benefits and limitations of different alternatives. There may be those who are cautious or skeptical about the application of a complex program to make a critical design decision. This concern is often appropriate, as many models are unproven or there is little information available about their accuracy. As these simulation models become easier to use, it may be practical to use more than one model on some studies. The two-model approach was applied in this study as a means of making the analysis more reliable, and the results more defensible. The results proved the consistency and reasonableness of the simulation tools, and provided everyone involved with confidence about the analysis. This study also illustrated the value in using a range of performance measures, and a sensitivity analysis. More generally, it proved the value in providing as much comparative information as possible before making a design decision. The results were generally consistent, and the end product was a set of clear, defensible, and well-supported conclusions. While the experience gained through the application of CORSIM and VISSIM in this study was in some ways unique to the study area, this experience can provide insight to other transportation professionals charged with selecting and applying these simulation models to similar networks. To that end, some of the characteristics of both models are contrasted in this paper. Bloomberg and Dale page 2 INTRODUCTION AND PURPOSE This paper summarizes the findings of a comprehensive traffic operations analysis conducted using two simulation models, CORSIM and VISSIM. Traffic simulation packages like CORSIM and VISSIM are frequently used as tools for analyzing traffic. A common example is the analysis of multiple alternatives for a roadway design. Simulation modeling is an effective approach for quantifying the benefits and limitations of different alternatives. Typically, a single software package (e.g., CORSIM) is selected for these types of analyses. The specific model selected will depend on the circumstances (e.g., the type of facility, and the experience of the staff assigned to apply the simulation model). Regardless of the model selected, most researchers and practitioners working with simulation models will develop a calibrated base model of existing conditions, extend the model to include the design alternatives (generally using projected future year traffic demands), and then make conclusions based on the modeling results. One drawback of this approach is that there may be concerns about the accuracy of the simulation modeling tool. As these simulation models become easier to use, it may be practical to use more than one model on some studies. As an example, the authors recently completed work on a study where both CORSIM and VISSIM were applied to a study of design alternatives in Seattle, WA. The two-model approach was selected as a means of making the analysis more reliable, and the results more defensible. The purpose of this paper is twofold: • To provide detailed information about recent findings from using these two popular traffic simulation models (CORSIM and VISSIM). Comments are directed largely at the users of these two models. • To illustrate how practitioners and researchers can and should use multiple simulation modeling packages in their studies. It is the authors’ hope that others will use some of these approaches to improve their modeling processes. The paper begins with a brief discussion of the CORSIM and VISSIM models, and a description of the project where the models were applied. Then, the analysis approach and results are outlined. The findings are summarized in two ways: a technical comparison of the two models, and some more general comparisons believed to be of value to simulation model users. The final section provides some concluding remarks about the models and the process, recommendations to users, and suggested areas of additional research. Bloomberg and Dale page 3 TRAFFIC SIMULATION MODELS The CORSIM Model CORSIM (1,2,3,4) is a microscopic simulation model designed for the analysis of freeways, urban streets, and corridors or networks. The model includes two predecessor models: FRESIM and NETSIM. FRESIM is a microscopic model of freeway traffic, and NETSIM is a model of urban street traffic. CORSIM’s capabilities include simulating different intersection controls (e.g., actuated and pre-time signals); almost any surface geometry including number of lanes and turn pockets; and a wide range of traffic flow conditions. CORSIM is based on a link-node network model. The links represent the roadway segments while the nodes mark a change in the roadway, an intersection, or entry points. CORSIM was developed and is maintained by the Federal Highway Administration (FHWA). It is run within a software environment called the Traffic Software Integrated System (TSIS), which provides an integrated, Windows-based interface and environment for executing the model. A key element of TSIS is the TRAFVU output processor, which allows the analyst to view the network graphically and assess its performance using animation. Version 4.2 of TSIS was used for this study. The VISSIM Model VISSIM (5,6,7) is a microscopic, time step and behavior based simulation model developed to analyze the full range of functionally classified roadways and public transportation operations. VISSIM can model integrated roadway networks found in a typical corridor as well as various modes consisting of general purpose traffic, buses, light rail, heavy rail, trucks, pedestrians, and bicyclists. The model was developed at the University of Karlsruhe, Germany during the early 1970s. Commercial distribution of VISSIM began in 1993 by PTV Transworld AG, who continues to distribute and maintain VISSIM today. VISSIM version 2.91 was used in this study. The model consists of two primary components: (1) simulator and (2) signal state generator (SSG). The simulator generates traffic and is where the user graphically builds the network. The user begins by importing an aerial photo or schematic drawing of the study area into the simulator. Next, the user begins “drawing” the network and applying attributes (e.g., lane widths, speed zones, priority rules, etc.). Although links are used in the simulator, VISSIM does not have a traditional node structure. The lack of nodes provides the user with the flexibility to control traffic operations (e.g., yield conditions) and vehicle paths within an intersection or interchange. Bloomberg and Dale page 4 The SSG is separate from the simulator. It is where the signal control logic resides. Here, the user has the ability to define the signal control logic and thus emulate any type of control logic found in a signal controller manufacturer’s firmware. The SSG permits the user to analyze the impacts of signal operations including, but not limited to: fixed time, actuated, adaptive, transit signal priority, and ramp metering. It is important to note that fixed time control can be implemented in the simulator. The SSG reads detector information from the simulator every time step. Based on the detector information, the SSG decides the status of the signal display during the subsequent time step. PROJECT APPLICATION CORSIM and VISSIM were applied as part of the State Route (SR) 519 (S. Royal Brougham Way) alternative analyses conducted for the Washington State Department of Transportation (WSDOT) in 1999. Figure 1 illustrates the future project study area, which is located in downtown Seattle, just south of the Central Business District (CBD). SR 519 is an east-west arterial that begins at the ferry dock (just west of the study area) and continues to the east. In its current alignment, SR 519 is a two-way, 4-lane (with left-turn bays), at-grade roadway that ends at 4 Avenue. In the future, the western terminus of I-90 will become the east end of SR 519. SR 519 and nearby roadways are often highly congested during the peak periods. WSDOT identified the need for physical improvements to the roadway network to eliminate an at-grade railroad crossing, to improve overall traffic performance, and to accommodate future growth. Initially, many alternatives were generated using different combinations of these options. Ultimately, six were retained for more focused analysis using the simulation models: A: A three-lane alignment on SR 519, prohibiting access to SR 519 from the 4 Avenue ramp, and no reversible lanes on Atlantic Street. B: A two-lane alignment on SR 519, prohibiting access to SR 519 from the 4 Avenue ramp, and no reversible lanes on Atlantic Street. G: A three-lane alignment on SR 519, prohibiting access to SR 519 from the 4 Avenue ramp, with reversible lanes on Atlantic Street. H: A two-lane alignment on SR 519, prohibiting access to SR 519 from the 4 Avenue ramp, with reversible lanes on Atlantic Street. I: A three-lane alignment on SR 519, prohibiting access to the Seahawks’ garage from I-90, and no reversible lanes on Atlantic Street. Bloomberg and Dale page 5 J: A two-lane alignment on SR 519, prohibiting access to the Seahawks’ garage from I-90, and no reversible lanes on Atlantic Street. These six design alternatives were analyzed using forecasted 2020 demands during the PM peak period when an event (baseball game) was scheduled. Volumes were adjusted for changes in demand and turning probabilities as required by each alternative. ANALYSIS APPROACH The simulation modeling described in this paper was undertaken to help quantify the benefits and impacts of different alternatives. Traffic operations analysis, using simulation modeling, was identified as the best approach for assessing the traffic performance impacts of the various alternatives. Approaches for analyzing the system as a series of intersections were also considered. For example, the Highway Capacity Manual (HCM) (8) procedures could have been used to estimate delays at signalized intersections. However, the procedures do not adequately capture the system impacts of long queues and severely oversaturated conditions. Therefore, simulation modeling was identified as the most appropriate tool. This section describes how the simulation models were developed and run to support the analysis. Inputs and Coding The first step in the process was to compile the input data needed for the two models. These included supply (geometric components), demand (traffic), and control (traffic signal timing). The input data required for the two models were quite similar, although both models include other features that were not used here. Once the inputs were reviewed, and a consistent set of data was developed, the study network and design alternatives were coded in both VISSIM and CORSIM. The analysts developing the two models frequently communicated to ensure that the same supply, demand, and control assumptions were used in both models. However, since one goal for this study was to compare the results of VISSIM and CORSIM, the two models were developed independently to allow a fair comparison between the two sets of results. Testing and Validation Once the models were completed, the two models were tested and validated. On-screen animation and model outputs were reviewed for reasonableness and coding accuracy. A particular concern was driver behavior in the congested networks. Occasionally, CORSIM driver behavior was observed to be unrealistic (e.g., blocking traffic in the right lane to jump a long queue in the left turn lanes). In other Bloomberg and Dale page 6 cases, vehicles got "stuck" for periods of time. In these cases, changes to CORSIM input parameters (e.g., lane alignments, node locations, and driver behavior parameters) were needed. Then, model outputs were reviewed with WSDOT staff and others familiar with the network. This step served two purposes. First, it provided independent checks of the network and coding to ensure that the two models were accurate and consistent. Second, it provided a professional assessment of the validity of the future traffic demand projections and network performance. During this validation exercise, several improvements to the networks of both models were identified and executed. Model Runs Once the validation was complete, a series of model runs and analyses were conducted to analyze the traffic network. To allow for the most robust analysis possible, special care was taken to consider any reason for error or inconsistencies in the results: • Ten (10) runs were made for each scenario and model. Since both VISSIM and CORSIM are stochastic (random) models, there may be minor differences in the results depending on the random number seed. Averaging the results from multiple runs addressed this issue. The results presented in later sections are the averages for ten runs. • Different measures of effectiveness were used. A qualitative assessment of each scenario was made by observing traffic (using the on-screen animation provided by both models). Then, comparisons of travel time on specific routes (illustrated in Figure 2) were made. Finally, systemwide measures of effectiveness (e.g., delay, speed) were assessed. • A sensitivity analysis was conducted, where demands were increased and decreased by 10%. In other words, factors of 1.10 and 0.90 were applied to all demand inputs, and results were analyzed. A total of 180 runs were made with each model. There were six design alternatives, and ten model runs were conducted for each. With the sensitivity analysis, each set of alternatives was analyzed three times with varying demand assumptions. Outputs from the two models were extracted into spreadsheets, so direct comparisons could be made between the two models and among the various design alternatives. Findings from the comparisons of the alternatives were communicated to WSDOT. For this paper, however, the more relevant findings are the comparison between the models and the lessons learned in the process. These findings are provided in the following section. Bloomberg and Dale page 7 FINDINGS This section is divided into three parts. The first part is a general comparison of the two models. The second is a quantitative comparison of the results from VISSIM and CORSIM. The last part is a summary of recommendations for other modelers, based on the successes (and obstacles) in this work. General Comparison of the Two Models There are occasional studies that provide a comprehensive summary of model families and individual packages (2), but direct comparisons of applications of specific models are difficult to find in the literature; references (3), (9), and (10) are examples. Much is learned about these models with direct comparisons, and it is hoped that the findings documented here will help to advance the knowledge of both models. In general, CORSIM and VISSIM have similar structures and capabilities. There are, however, some distinct differences that are noteworthy: (1) network coding process, (2) car-following logic, (3) gap acceptance, (4) modeling of signals, (5) animation features, and (6) output data. Network Coding Structure The network coding process is different between the two models. CORSIM uses a link-node structure. The user defines the location and attributes of nodes (e.g., intersections), and the nodes are connected with links. The user then assigns attributes (e.g., turn movement percentages speeds, lane configurations, traffic control devices, etc.) to the links and nodes. VISSIM networks, on the other hand, eliminate the use of nodes and rely on links and connectors to build a network. Networks are graphically built over a background map (i.e., aerial or base map) of the study area thus allowing the user to match the network geometry (e.g., curvature) to field conditions. Links are used to define the width and number of lanes for a given roadway segment. The connectors are then used to connect the links at intersections enabling the user to control vehicle paths in an intersection. Next, traffic compositions are defined by the user and attributes assigned to the links and connectors. Car-Following Logic The car-following model in CORSIM sets a desired amount of headway for individual drivers (there are ten user-definable driver types) corresponding to a specific amount of headway. Within the constraints of traffic control devices and other system elements, vehicles seek to maintain a minimum car-following distance while not exceeding their maximum speed. Bloomberg and Dale page 8 CORSIM uses an interval-based simulation approach, moving every vehicle (represented as a distinct object) and updating each traffic signal every second. When a vehicle is moved, its position (both lateral and longitudinal) on the link and its relationship to other vehicles nearby are recalculated, as are its speed, acceleration, and status. VISSIM uses the psycho-physical driver behavior model developed by Wiedemann (5) in 1974. The basic concept of this model is that the driver of a faster moving vehicle starts to decelerate as he reaches his individual perception threshold to a slower moving vehicle. Since he cannot exactly determine the speed of that vehicle, his speed will fall below that vehicle’s speed until he starts to slightly accelerate again after reaching another perception threshold. This results in an iterative process of acceleration and deceleration. VISSIM, like CORSIM, uses an interval-based simulation approach. VISSIM simulates traffic flow by moving “driver-vehicle units” through a network. Stochastic distributions are used to replicate individual driver-vehicle unit behavior and dynamic headway. Every driver with his specific behavior characteristics is assigned to a specific vehicle. Gap Acceptance The ten driver types in CORSIM are assigned variable gap acceptance parameters for permissive leftturns, right turn on red, and other gap acceptance situations. Each gap acceptance decision is independent; made by an individual driver considering the current available gap and a personal gap acceptance value. Gap acceptances in VISSIM is user-definable and location specific. Therefore, gap acceptance can vary from one point to another with a particular network based on the type of operations being simulated (e.g., permitted left turns, right turns on red, U-turns, and all-way stop control). Gap acceptance can also be varied by vehicle type. VISSIM provides an unlimited number of user-definable vehicle types. Signal Control Both CORSIM and VISSIM can model nearly any realistic signal control system. Both models also can interface with external control devices either through interfaces to physical traffic signal controllers or control algorithms residing in external software programs. However, the models differ somewhat in how actuated control is coded. CORSIM uses the “card” structure required for all of its input data, while VISSIM provides an external signal state generator that allows the user to custom define the control algorithm to be modeled. Bloomberg and Dale page 9