Fusing Physics-Based and Data-Driven Models for Car-Following Modeling: A Particle Filter Approach

Abstract

Microscopic modeling of vehicle movements and interactions is pivotal in traffic flow theory. Physics-based car-following (CF) models using mathematical formulations can delineate driving behavior in various traffic conditions with decent interpretability. However, given predetermined mathematical forms, they might fail to characterize complex, highly nonlinear phenomena. Data-driven CF models naturally excel in this regard considering their flexible architectures, but their performance is subject to data quality, especially distribution bias. In this paper, we propose a novel physics-informed particle filter (PIPF) model that fuses and takes advantage of the two approaches. Utilizing the intelligent driver model as the physics-based model and the multioutput Gaussian process regression as the data-driven model, the PIPF model integrates and embeds both models into a particle filter framework, enhancing both model adaptability and accuracy. The performance of the proposed model is examined through both single vehicle and multivehicle numerical experiments using the NGSIM trajectory data set. Compared with physics-based and data-driven models alone, the PIPF model demonstrates a performance improvement in terms of the root mean square error of about 11.16% and 29.43% in scenarios characterized by sparse data and about 19.81% and 3.84% in scenarios with sufficient data. Compared to traditional particle filtering models, the number of particles to achieve optimal results is reduced by 20%, meaning less computational complexity.