Modeling car-following behaviors considering driver heterogeneity: A multi-regime stochastic framework

Car-following behavior exhibits stochastic characteristics influenced by the inherent randomness of human drivers. Stochastic models have been extensively developed to capture the probabilistic nature of car-following behavior dynamics. However, the time-varying nature driven by driver heterogeneity has not been adequately studied. To this end, this paper proposes a stochastic modeling framework that incorporates multi-regime car-following models with a Bayesian calibration approach to account for the driver heterogeneity in human car-following behaviors. More specifically, our framework employs a K-means clustering algorithm to categorize human drivers into three driving styles and leverages a hierarchical Dirichlet process-hidden semi-Markov model (HDP-HSMM) to segment car-following sequences into diverse driving regimes, thereby capturing driver heterogeneity. According to the segmented driving regimes, three distinct hierarchies of multi-regime Bayesian intelligent driver models (denoted pooled, hierarchical, and unpooled B-IDM) are developed to capture the time-varying nature of car-following behaviors across diverse driving regimes. These models are well-calibrated using a Bayesian approach with car-following trajectory data extracted from the Waymo open motion dataset and Lyft level-5 dataset. Deterministic and stochastic simulations are performed to evaluate the effectiveness of the proposed stochastic modeling framework. The experimental results demonstrate significant differences in car-following behaviors across various driving styles and driving regimes. The proposed framework effectively represents these heterogeneous car-following behaviors through the developed multi-regime hierarchical B-IDM with time-varying parameters. Additionally, the stochastic simulation achieves a more accurate representation than the deterministic simulation in replicating the dynamics of human car-following behaviors.