关键词: 智能驾驶, 匝道混行汇流, 深度 Ｑ学习, 平均道路时空利用率, 驾驶风格

Abstract: Vehicle confluence at a highway on-ramp area is a difficult problem for the decision system of intelligent vehicles,among which the mixed confluence of human-driven vehicles and intelligent vehicles has been one of the most complicated cases.In order to improve the vehicle traffic efficiency in the confluence area and reduce the emissions of pollutants,by analyzing the influence of vehicle lane distribution on traffic efficiency,this paper innovates an intelligent vehicle merging model based on Deep Q-Network (DQN) algorithm,and improves the optimization objective function of the algorithm according to the average road space-time utilization.At the same time,according to the real data set,the driving style of vehicles is defined and the mixed traffic simulation scene is established.Finally,the experimental results show that,under three different traffic flow conditions,compared with the Intelligent Driver Model (IDM) model,the DQN-based main lane vehicle change model improves the overall traffic efficiency of the ramp confluent area by 23.10% on average.Fuel consumption per vehicle reduces by an average of 12.7%.The emission of all kinds of polluting gas per vehicle reduces by about 10% to 20%.