Research On Traffic Signal Control Based On Deep Reinforcement Learning

With the development of society and the improvement of living standards,people’s experience and requirements for transportation have become more and more demanding;Intelligent Transportation System(ITS)plays a key role in alleviating traffic congestion.Technologies such as deep learning and reinforcement learning provide a possibility to get a more efficient and convenient traffic signal control methods.This paper explores and applies the deep reinforcement learning algorithm model,takes the closed-loop system composed of vehicles and traffic signal controllers at the intersection as the main research object,completes the real-time adjustment of vehicles in the prescribed simulation environment,and improves the timeliness of the vehicles.The contents of this dissertation are as follows.(1)Traffic signal control analysis of Q network model based on fully connected neural network.In this model,first we analyze the factors that affect the results of the fully connected neural network;Then we focus on the influence of the number of neurons and the number of hidden layers in the structural parameters on the simulation results.Through related simulations,it is concluded that the number of neurons will promote the final simulation results;Finally,we make a comparative analysis between the simulation results of the number of neurons and the number of hidden layers and the simulation results of the number of neurons alone.The simulation results show that the simulation results of neurons alone are better than the simulation results generated by the combined effects.(2)Traffic signal control analysis of Q network model based on convolutional neural network.In this model,first we analyze the factors that affect the results of the convolutional neural network;Then we focus on the influence of the convolution kernel size,the number of convolution layers in the structural parameters,and the batch size,learning rate in the operating parameters on this simulation,and it is verified through related simulations that the size of the convolution kernel will have an impact on the final simulation result;Then through simulations,the simulation results of the combined effect of the size of the convolution kernel and the number of layers of the convolution layer are compared with the simulation results of the size of the convolution kernel alone,and the simulation results of the convolution kernel alone are better than the simulation results produced by the combined effect.Finally,the simulation verified that the optimization of the operating parameters has a certain promotion effect on the simulation results,and compared the simulation results of the operating parameters alone and the operating parameters and the convolution kernel.The analysis found that the results under the combined effect are better than the results under the single effect.Under the limited simulation index,we analysis and compare the above two models,and get the better model under different indexes.