Using Particle Filtering as a Tool for the Comparison of Car-Following Models

The Car-following models and their application in microsimulation is of great importance in a broad range of transport applications, ranging from highway design to Intelligent Transport Systems. However, one of the fundamental challenges in the application of these models is the lack of conclusive and reliable studies on the comparison of car-following models. In spite of fundamental differences in the mathematical structures of these models, none of the many calibration-based studies in the literature so far has been able to deliver a reliable account of the strengths and weaknesses of different car-following models in replicating driving behavior, as it is often found that different carfollowing models have only marginal differences in terms of cumulative errors and different carfollowing models describe different trajectories best. In this paper a method based on parameter tracking and particle filtering is proposed to gain a more accurate insight into fundamental differences of car-following models and their ability to consider driving behavior. The application of the proposed method on four car-following models and multiple trajectories shows that unlike the traditional comparison criterion, i.e. comparison based on cumulative errors, the proposed method provides a clear and consistent measure of performance.