Research On Regional Signal Control Method Based On Deep Reinforcement Learning

The rapid development of social economy has laid the foundation for the process of urbanization.However,with the continuous acceleration of urbanization,the number of motor vehicles is also growing rapidly,and traffic congestion has gradually become an important factor restricting economic development.As the main place of traffic congestion,intersection correlation characteristics and signal control have always been the focus of academic research.The traditional signal control algorithm is difficult to adapt to the increasing traffic flow.Since the 21 st century,the rise of deep learning and reinforcement learning technology has brought a new direction to the research of signal control.In this thesis,a new sub area division algorithm is proposed,and based on this,an area signal control algorithm based on deep reinforcement learning is proposed.The main research is as follows:This thesis explores and studies the role of sub district division in traffic control.Firstly,based on the flow correlation degree,cycle correlation degree and link traffic flow density correlation degree,the weight between intersections is determined by entropy method.Secondly,in order to adapt to different scale regional networks,this thesis proposes a heuristic sub area partition algorithm LDPSO,which introduces Levi path into the traditional PSO algorithm and applies it to the partition of traffic network.In this thesis,the results of the partition algorithm are used as the basis of regional signal control.Finally,this thesis selects urban regional networks of different sizes as the experimental object,and simulates them under light traffic flow and heavy traffic flow respectively.The experimental results not only prove that LDPSO division algorithm can improve the capacity of different traffic networks,but also prove that this algorithm is superior to other comparison algorithms.Aiming at the problem that the existing regional signal control algorithm does not solve the information interaction between intersections,this thesis proposes an SPMAAC regional signal control algorithm combined with sub area division.Firstly,this thesis uses LDPSO partition algorithm to determine the area range of signal lamp control.Then,aiming at the problem of explosive growth of state space when MAAC algorithm is applied to regional signal control,this thesis improves it: the intersection agent only observes the information of the agents in the learning sub area,and focuses on the neighbor agents directly connected to it.Finally,the simulation results show that the algorithm can effectively improve the traffic capacity of the traffic network,whether in the regional road network of new first tier cities and third tier cities,or in the case of light traffic flow and heavy traffic flow.There are 21 figures,12 tables and 74 references.