Research On Virtual Link Driving Behavior Agent Based On Deep Reinforcement Learning And UBI Evaluation

As an important part of the intelligent transportation system,intelligent vehicles can help reduce traffic accidents,ease traffic congestion,and improve road traffic efficiency.With the increasing number of cars in China,the incidence of traffic accidents has increased to a certain extent.Research shows that 80%～90% of traffic accidents are caused by human driving factors,including but not limited to the driving behavior of drivers,travel time and road weather environment.Therefore,for different road sections and different traffic conditions,it is of great significance to obtain driving behavior norms through DRL,collect driving behaviors through personal terminals,evaluate and analyze them,and help drivers cultivate good driving habits and build intelligent vehicles.Combined with UBI insurance to evaluate driving behavior factors,the research goal is to make decisions on intelligent vehicle driving behavior norms and evaluate driving behavior.The main contents include:(1)Aiming at the problems of overestimation and low training efficiency of traditional depth Q learning algorithm,an improved D3QN(Dueling Double Deep Q Network)algorithm based on traditional depth Q learning algorithm is proposed.D3 QN algorithm firstly updates the Q value by updating the Q network of parameters on the basis of traditional DQN algorithm,which solves the over estimation problem of traditional DQN algorithm.Then the output network structure is changed.By outputting dominant states and value functions respectively,the state value can be estimated more smoothly.The experimental results show that the improved D3 QN algorithm improves the training efficiency compared with the traditional algorithm.(2)Based on deep reinforcement learning,a simulation training environment UBI for driving environment rules of virtual link is constructed＿Robot。 Combined with the influencing factors of UBI insurance for driving behavior judgment and the actual road driving rules,a simulation training environment was built through the Open AI Gym interface of the reinforcement learning platform.Finally,the feasibility of the environment was verified through random strategies and traditional in-depth reinforcement learning strategies,providing a simulation training platform for the proposed in-depth reinforcement learning algorithm strategy.(3)The UBI intelligent driving behavior evaluation software system is designed.The system can collect the real driving behavior information of drivers through mobile phones and upload the data to the cloud platform for analysis,load the driving behavior agent model through the cloud platform,and give the corresponding recommended driving behavior actions in combination with the real-time driving status.Compare the online data and the driving behavior agent model in the same time dimension to the two driving behavior sequences,further introduce a reasonable scoring mechanism,use AHP-DTW algorithm to score and evaluate the driving behavior after driving,and apply the driving behavior agent training to the evaluation of actual driving behavior,so as to improve the driver’s driving behavior and create good travel habits.