Research On Route Guidance Algorithms Based On Deep Reinforcement Learning

With the steady development of China's industrialization and science and technology,too many vehicles bring many problems such as air pollution and traffic congestion to many cities.In addition,according to the nature of urban traffic,urban traffic is affected by many uncontrollable factors such as accidents and weather.These factors make it difficult to induce vehicles in the traffic network.In a complex dynamic road network environment,reinforcement learning-based path induction is an effective solution.This paper analyzes the characteristics of urban traffic path guidance and the shortcomings of path guidance algorithms based on reinforcement learning,and proposes a traffic path guidance algorithm based on deep reinforcement learning to improve traffic congestion and improve the efficiency of traffic path guidance in order to achieve optimal Effects of traditional reinforcement learning path induction:First,through the analysis of intensive study for dynamic route guidance in the whole traffic network environment state as a reinforcement learning state of high dimension difficult,combined with the ability of high-dimensional complex state and the depth of the fitting of reinforcement learning status value function,is presented based on the depth of the reinforcement learning route guidance model to realize the vehicle route guidance.Secondly,in order to solve the overestimation and stability problems of deep reinforcement learning network models,the ability of Double DQN to decouple optimal action selection and state value calculation is used to solve the overestimation problem of traditional DQN.The output Q value of the network is obtained by adding the value function and the advantage function to improve the stability of neural network training.Therefore,a deep reinforcement learning path induction model based on a hybrid network is designed,and the Boltzmann probability selection strategy is used to realize the traffic road Vehicles in the network perform route guidance.Finally,by comparing with the reinforcement learning path induction algorithm,based on the number of vehicles in the road network and the average driving time of the vehicle as the evaluation criteria,the validity of the deep reinforcement learning path induction model proposed in this paper is verified,which can further improve the effect of route guidance on vehicles in the traffic network.