Research On Intelligent Control Method Of Intersection Signal Light

With the rapid development of China’s economy,more and more people buy cars for convenient travel,and the increasing number of cars has brought serious congestion to the road traffic.Aiming at the problem of low traffic efficiency under the timing control of traffic lights at single intersection,in this thesis an intelligent control method is proposed based on deep reinforcement learning to control traffic lights by timely detecting vehicle flow,average vehicle speed and queue length of vehicles at the intersection,so as to alleviate traffic congestion and improve traffic efficiency at the intersection.The main work is as follows:(1)The video image shot by the intersection surveillance camera is taken as the input image of YOLOX target detection algorithm.The convolutional neural network is used to extract the image features.The target detection algorithm outputs the vehicle target position information in the image.The output information of YOLOX target detection algorithm is taken as the input of DeepSORT tracking algorithm.The vehicle flow,vehicle speed and lane queue length are calculated statistically by using the data obtained from detection and tracking.The improved target detection algorithm can reduce the situation of missing and misdetecting target vehicles and improves the accuracy of detecting traffic status information.(2)The bidirectional five-lane traffic simulation model of single-intersection is built in VISSIM software,which not only provides traffic state information for the agent to learn,but also provides a simulation environment for the agent to control traffic lights.(3)The deep reinforcement learning algorithm is used to establish a traffic signal control algorithm.The traffic flow,the average speed of vehicles passing the intersection and the average queue length of vehicles are taken as the state input of the agent.The reward value is calculated by the traffic flow,the average speed of vehicles passing the intersection,the average queue length of vehicles and the travel delay time,and the value brought by the action is calculated through the neural network.When calculating action value,action Q value will be overestimated by DQN,and long-term accumulated experience will bring large errors.Therefore,Double DQN and Dueling DQN are introduced to improve the algorithm,and the improved method can alleviate the problem of overestimating Q value.(4)The control algorithm is programed in Python language,and the experiment is completed through com interface and VISSIM co-simulation.The experimental results show that the traffic light control method based on deep reinforcement learning is superior to the traditional control method in single intersection,and can reduce the congestion of single intersection and improve the traffic efficiency of vehicles.