Research And Implementation Of Regional Road Traffic Collaboration Based On Multi-agent

At present,the rapid growth of the number of motor vehicles in my country has intensified the problem of traffic congestion.The intersection controlled by signal lights is the main place of traffic congestion.The traditional timing traffic signal control method cannot adapt to the dynamically changing traffic environment.With the development of artificial intelligence technology,it has become an inevitable choice to establish an intelligent traffic signal control system and improve the traffic efficiency of urban road vehicles.Firstly,a deep reinforcement learning agent model for single intersection signal control is constructed,and elements such as state space,action space,and reward function in the model are designed.Specifically,the number of vehicles on the entrance road of the intersection,the average speed,and the signal phase are used to construct the traffic state vector,and the classic four-phase structure of the intersection is used to construct the action space,and the delay time,queue length,and number of stops are used as indicators for judging the effectiveness of signal control.The reward function uses the deep reinforcement learning algorithm DQN to train the decision-making neural network in the model.In order to overcome the problem that it is difficult to obtain complete traffic state information in actual traffic scenarios,the LSTM network is introduced into the network structure of DQN.In order to overcome the low efficiency of agent exploration,For the problem of slow convergence speed of the algorithm,it is proposed to introduce a noise network into the DQN network.In order to verify the performance of the algorithm,a traffic signal simulation platform based on SUMO is built,including the design of the simulation scene and the design of the simulation environment.By simulating and comparing other methods in different traffic scenarios,the results show that the optimized algorithm can better coordinate the various phases of the intersection,and reduce the delay time,queue length and parking times of vehicles.Secondly,a multi-agent reinforcement learning model of regional signal cooperative control is constructed,and the state,action,and reward elements in the model are designed.Specifically,the local traffic state vector is constructed by the number of vehicles,average speed,and signal phase of each intersection.The local traffic state of all intersections constitutes the global traffic state.The action space of each intersection is a classic four-phase.Time,queuing length,and parking times are used to design local reward functions.The local rewards of all intersections constitute the global reward.In order to overcome the problem of dimension explosion in the traditional multi-intersection signal control method,this paper introduces a learning mode of centralized training and distributed execution.And use the multi-agent reinforcement learning algorithm Qmix under the cooperative relationship to train the decision-making neural network in the model.In order to solve the problem that Qmix is difficult to accurately express the contribution of the local value of each intersection in the area to the global value,the local value and the global value are compared.The relationship between values is studied and analyzed,and the attention weight in the attention mechanism is used to dynamically learn the relationship between local values and global values,thus reflecting the importance of each intersection to the overall situation.By simulating and comparing other six control methods in different traffic scenarios,the results show that the optimized algorithm can better cooperate with all intersections,reduce the delay time,queue length and parking times of vehicles in the entire area,and effectively Improve the traffic efficiency of vehicles.Finally,with the help of Pycharm,Qt designer and other tools,the traffic signal collaboration software is realized.The software integrates the algorithm studied in this paper under the graphical interface,and the actual control effect of the algorithm can be tested directly through the operation interface.The results show that the traffic signal control algorithm studied in this paper can effectively improve the traffic congestion and achieve the expected research goals.