Transit Priority Signal Control Strategy Based On Deep Reinforcement Learning

With the accelerating of urbanization,urban traffic congestion has becoming a more and more serious problem,which can be alleviated effectively by the priority development of public transit with the features of low-emission and large-capacity.Chinese government has declared that strengthening the urban public transportation management is an important part of urban development in recent years.In order to improve the comprehensive service of the urban public transportation system and alleviate the pressure of urban traffic congestion and resources and environment,the priority development of urban public transit needs to be promoted.Furthermore,the development and integration of Artificial Intelligence(AI)and Internet of Vehicles(Io V)technology has promoted the transfer process from traditional transportation to intelligent transportation,and also provides a more flexible way for the priority development of public transit.In this paper,the optimization control strategy of transit priority signal based on deep reinforcement learning is studied.By taking the closed-loop control system composed of intersections and traffic signal controllers as the research object,an optimization control strategy of transit priority signal is proposed based on the deep reinforcement learning method.Then the proposed approach is applied to the real traffic scenes.Meanwhile,the Simulation of Urban Mobility(SUMO)traffic simulation platform is used to build the experimental environment.Simulation results based on actual traffic scenarios have shown that the method proposed in this paper can effectively achieve the public transport priority.The main contents of this paper are as follows:1.The research goal of the priority of public transportation signals is proposed by throughing the evaluation and analysis of the operation efficiency of public transportation vehicles at typical intersections,Then the control model and algorithm of transit priority are deeply studied to provide a theoretical basis for constructing an optimization control strategy of transit priority signals.2.An optimization control strategy of transit priority signals for single intersections based on improved deep reinforcement learning is proposed.Firstly,the time boundary constraints of the transit priority signal control problem are designed by considering safety,overall benefit and energy saving,and the optimization control model of the transit priority signal is constructed.Secondly,the structure of the Deep Q Network(DQN)algorithm is adjusted and improved in order to increase the training speed of algorithm.Finally,the traffic system simulation software SUMO is used for simulation experiments,and the experimental results are compared with the timing control scheme obtained by the Webster method.The results have shown that the optimization signal control strategy proposed in this paper can reduce the average delay of public transportation vehicles and all vehicles by 30% and 15.38%,and also increase the traffic capacity by 16.67% and 14.29% respectively.In addition,in order to further verify the signal control effect of the proposed method,a single intersection in Fuzhou is used as an example traffic scene.The experimental results have verified that the proposed method in this paper can effectively reduce the average delay of public transit.3.An optimization control strategy of transit priority signal of multi-agent regional multiintersection based on the improved deep reinforcement learning is proposed.With the basis of the optimal control strategy of transit priority signals at single intersections,the interaction relationship of the traffic states between adjacent intersections in the regional multi-intersections is analyzed,and the control strategy of collaborative signal of multi-agent deep reinforcement learning for regional multi-intersections is constructed.Comparing with the original timing signal control methods,the proposed method decreases the average delay by 23.89%,and improves the traffic capacity of transit at regional multi-intersection by 22.19%.At the same time,the traffic capacity and average delay of other vehicles at some intersections are also performed well,which shows that the algorithm in this paper can effectively achieve public transport signal priority at regional multi-intersection.