Signalized Intersection Eco-driving Strategy Based on Deep Reinforcement Learning
Eco-driving in a connected and autonomous driving environment has great potential to improve traffic efficiency,energy saving,and emission reduction.This paper proposes a prosocial eco-driving strategy based on the deep reinforcement learning algorithm that optimizes the longitudinal manipulation and lateral decision-making of the connected and automated vehicle(CAV).The state space is divided into the local variables related to dynamic vehicle characteristics and the global variables associated with signalized intersection to ensure adequate interaction between the CAV and the roadway environment.The designed reward function integrates the vehicle driving requirements,synergy with signals and global energy saving incentives.In addition,this study developed a typical urban road intersection scenario to train the model.The results show that the proposed strategy can achieve eco-driving of the CAV in collaboration with the signal and output lateral control to ensure the vehicle travels to the target lane.In addition,simulations in a dynamic traffic environment reveal that the proposed method can improve the capacity at the intersection by about 17.90％and reduce the traffic system's fuel consumption and pollutant emissions by approximately 8.76％through the control of multiple CAVs to guide the human-driven vehicles.
intelligent transportationeco-drivingdeep reinforcement learningconnected and autonomous vehiclesignalized intersection
万方数据
维普
