关键词: 双向双车道道路；半幅封闭作业区；强化学习；混行交通流；ICV

Abstract: The partial closure of bidirectional two-lane work zones often creates bottleneck delays, reducing road capacity. In order to improve the traffic efficiency of the two-way double-lane half-closed operation area, alleviate the vehicle delay on the road section, and give full play to the control advantages of intelligent connected vehicles (ICV), a speed guidance model is designed based on the improved TD3 algorithm, which enables ICV to pass through the signal port at high speed, with low delay and no stop. First, a signal timing method for bidirectional two-lane work zones is designed using the Webster model. Second, technical improvements to the TD3 algorithm are implemented: hybrid weighted averaging is introduced to mitigate Q-value overestimation bias, and a StepLR scheduler is integrated to optimize learning rate sensitivity. Third, the model’s state pace, action space, and reward function are systematically formulated. Numerical simulations demonstrated that the proposed model effectively guides most ICVs to arrive at intersections during green phases and execute alternating lane maneuvers under mixed traffic flows with varying ICV market penetration rates and traffic volumes. This work provides a novel framework for optimizing traffic operations in dual-lane work zones under mixed ICV-human driving conditions.

Key words: bidirectional two-lane road; alternatively passing work zone; reinforcement learning; mixed autonomy; ICV