Traffic Choice Behavior And Driverless Simulation Based On Reinforcement Learning

With the rapid growth of the global economy,the number of cars on the urban roads is increasing year by year,the traffic congestion situation is only increasing,the situation of driving safety is becoming increasingly severe,the traffic congestion is accompanied by the occurrence of traffic accidents,saving the cost of congestion and driving safety need more and more attention.How to save travel time on the road has become a hot research topic.Intelligent transportation promotes the emergence of driverless vehicles.In order to ensure the smooth and safe traffic system and the safety of driving determine whether the vehicle can reach the destination smoothly,this paper collects and analyzes the travel choice data by constructing a virtual platform of travel choice behavior,and proposes a reinforcement learning model to study the travel choice behavior,and then plans the path of the driverless vehicle by improving the DDPG algorithm,so as to achieve the goal of slow down The purpose of solving road congestion.(1)In view of the lack of the actual data of the travel choice behavior,this paper uses Django based on the web framework to develop an experimental platform for the selection of the departure time of urban commuting.After modifying the experimental parameters by fine-tuning method,the travel choice experiments under the conditions of path selection and charge management are implemented.Using this platform to simulate the travel choice behavior,and collect relevant data to build data set.The most significant advantage of the platform is that it not only breaks the traditional PC implementation,but also can use mobile devices such as pad or mobile phone to carry out experiments using We Chat applet.(2)With the increase of means of transportation,traffic congestion has become the most typical urban disease,and people's commuting time is gradually longer.In order to better solve this problem,based on Vickery's bottleneck model,this paper carried out the departure time selection experiment under the condition of the change of the road capacity at the bottleneck.Through the analysis of the experimental data on the time selection behavior and results of the subjects,the main task is to It is necessary to analyze the equilibrium characteristics of global choice behavior and the influence of individual choice behavior on global equilibrium.The results show that the model has strong adaptability and can reproduce the whole selection behavior under various experimental conditions.(3)With the rapid development of the economic era,the traffic flow on the road is increasing gradually,which leads to frequent traffic accidents and road congestion.The path selection is particularly important.However,in the complex environment,it is very difficultly to find the difference between the optimal action and the suboptimal action using DDPG algorithm when the action dimension is high.To solve this problem,we proposes an Optimization Evaluation(OE-DDPG)algorithm based on the DDPG algorithm.By improving the loss function and adding the Dropout mechanism,the difference between the TD?error value of the optimal action and that of the suboptimal action can be widened.Applying the present algorithm,we simulated the learning process of driverless vehicles in three different environments.The results show that our algorithm can effectively improve the convergence speed and robustness,and can well carry out path planning.