Techincal Research On Intelligent Traffic Signal Timing Optimization Based On 3D Convolution Deep Reinforcement Learning

At present,China’s urbanization construction continues to advance,a large number of people settle in the city,the car ownership continues to increase,and the phenomenon of urban traffic congestion is worsening,which has brought serious adverse effects on the development and quality of China’s social economy and residents’ living standards,and has also caused adverse effects and great restrictions on the further promotion of urbanization construction.Traffic congestion is related to many factors,such as route guidance,signal control,traffic infrastructure,car ownership and so on.This paper mainly focuses on the topic of traffic signal control.It combines deep learning with clustering algorithm to construct an adaptive dynamic decision-making control system of traffic signals,so as to better solve the congestion problem in the traffic network with high saturation.First of all,considering the complexity of the traffic network,there are many factors that affect the control of traffic lights.We can not only consider the historical traffic data of a single time segment.Therefore,this study adds the historical traffic data of different time segments.At the same time,for the fact that the structure of the traffic network is unbalanced and the overall scale is large,K-means clustering algorithm is used to cluster the traffic network The whole road network is divided into several sub regions(intersection).In addition,considering the outstanding performance of deep learning technology in the field of image perception,this paper studies the introduction of 3D convolution neural network model to simulate the characteristics of the traffic network in the space-time dimension,and input the spatiotemporal dynamic road network information data which is standardized.In this paper,the intersection is regarded as the intelligent subject of deep learning.The deep reinforcement learning algorithm is used to analyze the road network control action of the intersection.Through continuous feedback learning,the dynamic control system of intersection signal light which can deeply strengthen and accumulate experience is constructed.The intelligent control of intersection signal light is carried out,and the signal strategy is adjusted according to the road network congestion High traffic capacity can effectively solve the problem of congestion.This research is based on SUMO platform to simulate urban traffic network.We have two evaluation indexes,one is the average detention time of vehicles in the road network,the other is the number of vehicles in the road network,and then compare the proposed strategy with the traffic light timing strategy based on deep reinforcement learning.According to the experimental results,the improved traffic light timing strategy based on 3D convolution depth reinforcement learning is better than the above strategy in the case of different road network vehicle congestion,and can more effectively alleviate the situation of traffic congestion.