Research And Design Of Traffic Signal Control Based On Multi-agent Coordination In Edge Computing Environment

With the development of the economy,the number of vehicles has maintained a continuous growth,and traffic has become congested.Traffic signal control is an important method to improve the traffic capacity of the road network.Because traditional traffic signal control methods cannot be adjusted with changes in traffic flow,adaptive control based on intelligent algorithms is currently the main research direction.However,the traffic flow affects each other.Besides,the overall performance of the road network cannot be represented only by the single intersection,and the more intersections involved,the worse the control effect.In addition,traffic signal control has the characteristics of real-time data collection and high delay requirements,so it is not suitable to deploy its calculation process in the cloud.To solve these problems,this paper divides the road network into multiple sub-areas in the edge computing scene,designs the intersection cooperation mechanism and combines multi-agent reinforcement learning to realize the traffic signals control in each sub-area.The main contents include the following:This paper proposes a layered architecture of the edge computing in traffic signal control scene.The network entities are identified and the static relationships and cooperation modes between the entities are designed.A road network partition method combining the temporal dynamics and spatial dynamics of traffic flow is proposed.This paper consider the classification quality and signal control range to determine the optimal number of sub-areas in the road network,and the road network is divided by the clustering algorithm.Besides,this paper proposes an algorithm combining fuzzy control and reinforcement learning to implement single intersection traffic signal control based on the edge nodes.The complicated and changeable traffic flow information is optimized through fuzzy decision,and the traffic signal control problem is abstracted into a Markov decision process.Reinforcement learning is used to solve the optimization process to obtain a dynamic signal timing scheme.An algorithm combining collaborative computing and multi-agent reinforcement learning is proposed to implement multi-intersection traffic signal control.The entities implement information cooperation through a reasonable multi-agent interaction model,perform collaborative calculations based on static physical characteristics and dynamic traffic flow characteristics,and achieve adaptive signal control of multiple intersections in the sub-area with reinforcement learning.