Research On Decision-making Method Of Urban Traffic Congestion Dispersal Under Emergency Based On Reinforcement Learning

The road emergency,represented by traffic accidents,is one of the important factors causing urban traffic congestion.Due to the contingency and unpredictability of traffic congestion caused by emergencies,it is difficult for traffic managers to take measures to alleviate congestion in advance.Therefore,it is particularly important to take control measures in time after the emergency.Timely and effective control measures can reduce the spread of traffic congestion,accelerate the dissipation of congestion,and reduce the impact range,time and degree of traffic emergencies.Research on the decision-making method for alleviating urban road congestion can effectively prevent and alleviate the urban road network traffic congestion and improve the intelligent management level of urban traffic.Based on the cellular automata model,a dynamic traffic flow simulation model of urban road network is constructed.Reinforcement learning technology is used to optimize the decision-making of traffic congestion management,and the corresponding simulation platform is developed,the decision support system of congestion dispersal under emergency is preliminarily designed.First of all,this paper improves the vehicle update rules for road segments and intersections on the basis of the traffic flow model of the preceding vehicle speed and the safety distance,constructs a dynamic traffic flow simulation model of urban road network based on cellular automata,establishes the evaluation index system of road network traffic operation state including basic,characteristic and comprehensive indicators,which realizes the visual and quantitative evaluation of the operation status of the urban transportation system.Secondly,a traffic flow reinforcement learning model of urban road network is established by taking the road section,intersection and vehicle as the agent,taking the average travel speed of the road section as the state,taking the control measures such as vehicle prohibition as the action,and taking the degree of congestion change,the time required and the number of passing vehicles as the basis of rewards and punishments.Q-learning algorithm is used to solve the model,and the optimal mapping relationship between traffic flow state and congestion evacuation strategy under emergency is established,which realizes the adaptive learning of congestion evacuation measures.The simulation results show that the method can effectively alleviate traffic congestion according to different situations.Finally,based on the dynamic traffic flow simulation model of urban road network and reinforcement learning technology,the decision support system of congestion dispersal under emergency is preliminarily designed,which realizes the close combination of scientific theory and engineering practice,and provides theoretical basis and decision support for congestion control under emergency.