A Novel Variable Speed Limit Control for Freeway Work Zone Based on Deep Reinforcement Learning

The reduction of available lanes in freeway work zone areas often induces bottlenecks, which have a negative impact on travel time, safety, and the environment. In this paper, a more effective framework for variable speed limit control based on deep deterministic policy gradient (DDPG) is developed for solving the problem, which can impose dynamic and distinct speed limits across lanes. To achieve higher learning efficiency, this paper improves the DDPG algorithm from two aspects: balancing exploration and exploitation and improving the efficiency of parameter updating. According to simulation results of open-source software Simulation of Urban Mobility (SUMO), the proposed DRL-based DVSL controller is able to improve the safety, efficiency, and environment-friendliness of the bottleneck section in freeway work zone.