Attention-switching car-following behavior modeling under variable speed limits: Explaining structural homogeneity...

Drivers in variable speed limit (VSL) environments continuously balance following the leader and complying with the posted speed limit, yet existing car-following models treat the desired speed as fixed and cannot represent this state-dependent attention allocation. Moreover, whether VSL induces structural changes in car-following behavior or primarily triggers situational adaptation remains underexplored. This study addresses both challenges using wide-area trajectory data from the Shanxi Wuyu Freeway under three VSL conditions (60, 80, and 100 km/h) at Level of Service A-B. A dual-dimension behavioral homogeneity framework reveals that car-following behavior remains structurally homogeneous across scenarios with no discrete driver subtypes. On this basis, an Attention-Switching Car-Following Model (AS-CFM) is proposed that introduces a continuous attention weight λ as a function of time headway (THW), simultaneously controlling a dynamic desired speed and an adaptive headway within an Intelligent Driver Model (IDM)-type acceleration framework, enabling smooth transitions between leader-following and speed-limit compliance. Calibration on 1136 events shows a 4.3 % per-event RMSE reduction relative to the IDM, while leave-one-scenario-out cross-validation demonstrates a 15.1 % average reduction in prediction error relative to the IDM under unseen VSL scenarios. Within-event attention dynamics further indicate that the observed variability is primarily situational rather than dispositional, with λ dynamically buffering scenario-sensitive parameters during close-following conditions. These findings suggest that a unified attention-switching framework offers a parsimonious and transferable approach to car-following modeling across VSL regimes, providing a mechanistic account of how drivers allocate attention under dynamic speed regulation and a more robust behavioral foundation for simulation-based VSL evaluation.