Research On Adaptive Control Of Traffic Signal Based On Deep Reinforcement Learning

With the implementation of my country’s urbanization strategy,more and more people use private cars as a means of transportation.Using artificial intelligence and big data related knowledge to deal with traffic problems has become a research hotspot.Traffic congestion affects the sustainable development of the city,and it brings huge economic losses to the city.Therefore,it is necessary to greatly improve the traffic efficiency.How to alleviate traffic congestion,for which the development of intelligent transportation systems is the best choice to solve the problem.Due to the inaccurate description of the traffic state in the traditional traffic system,it still stays in the alternate phase of a single traffic light at a fixed time.The current solution has the problem of long algorithm processing time and poor control effect.With the development of artificial intelligence,people pay more and more attention to the application of reinforcement learning to traffic intersections.This dissertation mainly optimizes the traffic signal control problem,and uses deep reinforcement learning algorithm to control the intersection.The main work is divided into the following two parts:1.parameter settings are performed on the traffic simulation software SUMO,and various functional modules are set to simulate different traffic junctions.Through the interface of SUMO simulation software and Python language,and through this interface,users can easily call various functional modules.SUMO is free and open source software,and has a high degree of visualization,which greatly saves the cost of the experiment,and the demonstration of the experimental effect is very convenient.Through this platform,the reinforcement learning algorithm is designed and optimized,and the credibility of the simulation experiment is greatly improved.2.in view of the problems of long algorithm convergence time,poor control effect,and unreasonable reward function of adaptive traffic control system,the article proposes a new solution,using DDPG algorithm to control traffic intersections.By comparing the fixed timing method and the adaptive control method,it is known that the adaptive method has better traffic performance and realizes the adaptive control of the traffic signal at the intersection.This algorithm can adapt to most traffic conditions,and has better performance and stability than other algorithms.Aiming at the low training efficiency and slow convergence speed of the DDPG algorithm,the reward function is improved to obtain the improved DDPG algorithm,which is simulated and experimented on the simulation software.The experimental results show that the improved DDPG algorithm has greatly improved the training effect of the DDPG algorithm,and this method has obvious advantages over the traditional method.In general,thisarticle explores both the simulation software and the algorithm.The scheme proposed in this article can effectively solve the intersection control problem.