Abstract:

To address the inadequate characterization of the spatiotemporal evolution of microscopic interaction patterns in platoons following lane-changing insertions, this paper proposes a method for analyzing platoon evolution based on the identification of interaction intensity primitives. First, vehicle trajectory data from the affected platoon behind a lane-changing vehicle are extracted. By integrating safety potential field theory with considerations of conflict severity and spatiotemporal proximity, a quantitative model for vehicle interaction intensity is established. Second, a Hierarchical Dirichlet Process-Hidden Markov Model (HDP-HMM) is employed to automatically identify primitives within interaction intensity time series without pre-specifying the number of states. Subsequently, clustering algorithms are utilized to determine interaction intensity thresholds and define interaction patterns across different risk levels, followed by an analysis of the propagation and transition characteristics within the platoon. Finally, an empirical analysis is conducted using 1,763 lane-change events from the CitySim dataset. The results indicate that 16,550 interaction primitives were identified, from which 11 typical interaction patterns were extracted. The average duration of these primitives is 1.90 seconds, consistent with human driver reaction times. In terms of evolutionary characteristics, the middle-risk maintain pattern has the highest proportion (31.29%), while the high-risk rising pattern has the lowest (4.48%). Significant backward risk propagation and self-transition stability are observed within the platoon, with the middle-risk maintain pattern exhibiting the highest self-transition frequency. The primitive-pattern framework developed in this study uncovers the evolution laws of vehicle interaction intensity under lane-changing perturbations, providing theoretical support for autonomous driving decision-making, real-time traffic risk warnings, and the calibration of microscopic simulation models.

Key words: intelligent transportation, lane-changing vehicles, interaction strength, interaction primitives, HDP-HMM, spatiotemporal evolution