EVALUATION OF SIMULATION MODELS FOR CONGESTED DALLAS FREEWAYS

Recent and continuing growth has resulted in increased demand for travel on urban freeways throughout Texas. Many freeways are operating under congested conditions throughout much of the day. However, proposed operational improvements for congested freeways are difficult to evaluate or to simulate accurately because under congestion there is increased effect of vehicle interactions and impact of design elements on traffic flow. There have been improvements in recent years in traffic models that offer promise in the simulation of congested freeways or freeway elements including ramps, merges, weaving areas, and basic freeway sections. The more promising models include simulation of vehicle interactions, lane changing, car following and vehicle rerouting, accommodate origin-destination information, and in general try to model driver behavior. However, there is no universally accepted model for simulating congested conditions and thus arises the need for a project to determine which models produce the best results under different congested conditions. The objectives of this project are to select appropriate models for simulating congested freeways, test the calibration and validation performance of those models using data collected on Dallas freeways, and provide recommendations on the use of the best model for congested freeways in Texas. The CORSIM (FRESIM component), FREQ, and INTEGRATION models were selected for project based on the literature review. Three different freeway sections with bottlenecks that caused recurrent congestion were selected to test and evaluate model performance. Before and after operational data (i.e., speeds and volumes) at each of the sites were used in the attempt to calibrate and validate the chosen models. The research team determined that all of the models performed relatively well for uncongested conditions; however, the performance became sporadic and mostly unreliable for congested conditions. None of the models was successfully calibrated and validated for all of the test sites. The CORSIM model had the best overall performance in this project, but it was effective only on the simplest test site.