Research And Implementation Of Traffic Control Optimization Method Based On Deep Reinforcement Learning

With the rapid development of China’s economy and the continuous improvement of people’s living standards,the number of urban motor vehicles has grown rapidly,and traffic congestion has become an increasingly common problem in modern cities.As an indispensable place for vehicles,pedestrians to gather,turn,and evacuate,intersections are the bottleneck of urban capacity and the main management object of traffic signal control.The traditional model-based traffic signal control method often lacks real-time signal control decisions when faced with complex and time-varying traffic flows.At the same time,the traffic parameters used by it are difficult to fully reflect the current urban traffic state.With the development of next-generation high-tech technologies such as big data and artificial intelligence,new models and methods are constantly emerging,and have achieved obvious results in actual traffic control applications.This thesis mainly discusses the traffic control optimization method based on deep reinforcement learning.Taking urban isolated intersection and multi-intersection groups as the research object,a set of traffic signal control method oriented to practical application is proposed by using deep learning and reinforcement learning methods,and the algorithm module is implemented by software based on the general platform of a company’s traffic algorithm.The simulation results based on the measured traffic data show that the proposed method can effectively improve the signal control effect.The main work and contribution are as follows:1.Proposed a single intersection signal control method based on deep reinforcement learning.The two-dimensional matrix is adopted to transform the traffic state at the intersection,and the high-level abstract representation of the traffic state characteristics is carried out through deep learning,so as to realize the accurate perception of the traffic state.On this basis,the adaptive traffic signal control strategy is implemented by means of reinforcement learning,which can effectively reduce the queue length of vehicles at a single intersection in the time-varying traffic flow environment.2.Proposed a multi-junction signal control method based on deep reinforcement learning.On the basis of single intersection signal control method,deep reinforcement learning is extended to signal control of intersection group.According to the traffic characteristics and control requirements of the intersection group,the state space,action space and reward value functions in reinforcement learning are designed,and a signal control optimization method is proposed,which makes it suitable for multi-intersection signal control scenarios,to improve the average speed of vehicles in the intersection group and reduce the number of vehicles in the intersection group.3.According to the main research results of this thesis,an intelligent algorithm module is implemented.Based on a commercial general traffic algorithm platform,the intelligent algorithm module is designed and implemented by analyzing the platform algorithm structure.Finally,the research content of this paper is summarized and the shortcomings in the study are presented,as well as the prospects for future work.