With the popularization of mixed traffic flow environments between Connected Automated Vehicles(CAVs)and Human-driven Vehicles(HDVs),the current strategies for rapid and prioritized passage of high-priority vehicles such as emergency vehicles are difficult to play an effective role in mixed traffic flow scenarios.To achieve fast,efficient,and low disturbance traffic for high-priority vehicles in mixed traffic flow scenarios,a decision method based on multi-agent reinforcement learning is proposed.Firstly,using SUMO traffic simulation software,a reinforcement learning model is constructed,using CAVs in front of and on the side of high-priority vehicles as intelligent agents participating in reinforcement learning.Secondly,the Proximal Policy Optimization(PPO)is used to train the model using multiple continuous action spaces based on the number of agents;By adjusting the longitudinal speed of autonomous connected vehicles and sparse the longitudinal spacing between basic units of mixed traffic flow,provide lane changing and overtaking space for high-priority vehicles.Finally,three different longitudinal spacing scenarios were used to validate the trained model.The results indicate that this method is suitable for mixed traffic flow scenarios and scenarios where all vehicles are CAVs;In Scenario 2,compared to the existing lane pre-clearance strategy,the passing time and distance of high-priority vehicles are reduced by 17.39%and 5.09%,respectively;The overall number of lane changes has decreased by 75%,significantly reducing the impact on traffic and demonstrating significant performance in mixed traffic flow scenarios.
关键词
智能交通/混合交通流/强化学习/协同换道/车队编组/行为决策
Key words
Intelligent Transportation/Mixed Traffic Flow/Reinforcement Learning/Collaborative Lane Changing/Fleet Formation/Decision Making
维普
万方数据