Urban Adaptive Traffic Signal Control Approach Based On Multi-agent Reinforcement Learning

When urban social economic development reaches a certain level,urban roads tend to have different degrees of congestion.In order to overcome the limitations of existing urban road signal control methods,this paper proposes an optimal search-type reinforcement based on Markov decision process.The learned urban road adaptive signal control method is built and the simulation platform is built to verify the effectiveness of the proposed adaptive signal control method.This paper focuses on the application of agent technology to urban road adaptive signal control methods.The main work involves the following aspects: First,fully investigate the limitations of existing signal control methods,and consider the traffic pressure at the intersection of neighbors.The state of the environment in which the agent is located is defined,and the action selection strategy is proposed.The change value of the parking number is used as the reward value obtained by the transition between the agent state action pairs;then,combined with the simulation modeling capability of the VISSIM traffic simulation software,the VB system Graphical user interface and rapid application development capabilities,as well as the powerful matrix computing power and drawing capabilities of MATLAB system,using the generation and calling of COM components to build an interactive simulation platform based on VISSIM-VB-MATLAB,the adaptive signal control method proposed in this paper.The effectiveness is verified by simulation.In addition,part of the iterative learning process is extracted as the observation object,and the action control schemes of each intersection in the learning process,that is,the actions selected by each agent,and the Q table of each agent are visually processed.Finally,the intersection is performed.Total number of stops and total travel delays The running efficiency evaluation index is the control group with the number of intersection stops and the travel delay detection data in the fixed-cycle signal control mode.The number of intersection stop times and travel time detector data in the adaptive signal control mode is used as the experimental group,and SPSS statistics are used.The analysis software performs a paired sample t test.The experimental data show that the operating efficiency of the intersection in the adaptive signal control mode is significantly higher than that in the fixed-cycle signal control mode.Compared with the fixed-cycle signal control mode,the adaptive signal control method proposed in this paper can reduce the crossover.11.3%-51.8% of the number of stops and 13.1%-48.7% of the delay.The above research results show that the proposed urban adaptive traffic signal control method basedon multi-agent reinforcement learning is an effective signal control method,which can significantly improve the operational efficiency of each intersection in the region,and can compensate for the fixed-cycle signal control.The model can not adapt to the shortage of dynamic traffic demand,and can replace the widely used fixed-cycle signal control method to effectively control the traffic flow at urban road intersections.