Traffic Flow Modeling to Improve Traffic State Prediction

Abstract

In this research, the relationship between microscopic car-following models and macroscopic models has been explored and it was found that, based on the traditional assumption that traffic density is the reverse of space headway under steady-state homogeneous traffic conditions, most of the existing macroscopic speed-density relations can be derived from microscopic car-following models. The traditional assumption does not hold under non-homogeneous traffic and different macroscopic traffic models can be derived from different headway-density. The research also investigated the compatibility between the macroscopic and microscopic simulation. The microscopic simulation model, VISSIM, was calibrated and validated on an urban freeway. The VISSIM outputs were compared with the predicted traffic speed, density and flow from the second-order macroscopic model, METANET. Three levels of traffic demands and seven different time step lengths in macroscopic simulation were applied to evaluate the compatibility of the two models. It was concluded that, in macroscopic simulation, there exists an optimum time step length. Under moderate to heavy traffic demands, the predicted traffic states from the macroscopic simulation are consistent with the outputs from the microscopic simulation, and under stop-and-go traffic states, a significant difference exists between the two models. In addition, the impact of merging and weaving from freeway ramps on the performance of macroscopic simulation models was experimentally investigated. Several merging and weaving formulations in speed dynamics were evaluated and their contributions to the predicted traffic speed were quantitatively analyzed. Analysis of variances were carried out on the prediction errors from different models and it was concluded that, for the given formulation, the impact of merging and weaving terms on the prediction accuracy was not statistically significant, merging and weaving terms can be omitted in macroscopic simulation models. Finally, several improvements on the macroscopic simulation models were proposed. The improved models were applied to two freeways and compared with outputs from the original model, using both simulation data as well as field measured data from two freeways. It was concluded that the models with the proposed improvements have obviously better performance than the original model, especially in congested traffic conditions.