Research On Intelligent And Adaptive Driving Strategy Oriented To The Internet Of Vehicles

With the development of artificial intelligence,automotive intelligence has become a new direction for the development of the automotive field and intelligent industry.At the same time,the Internet of Vehicles is an important part of future intelligent transportation,and vehicles can use the Internet of Vehicles technology to communicate with other road users and infrastructure.This thesis mainly studies the safe and efficient driving strategy planning of ego vehicle（EV）through self-perception,vehicle-to-vehicle communication and vehicle-road coordination in the context of the Internet of Vehicles.The EV refers to an autonomous vehicle on the road.The existing research mainly completes the driving strategy planning according to the real-time driving state of the EV and surrounding vehicles.This thesis not only focuses on the current driving state of the EV and surrounding vehicles,but also predicts the future driving trajectories of surrounding vehicles in advance.Finally,the trajectory prediction results are added to the deep reinforcement learning model to solve the driving strategy planning problem of the EV’s lane-changing behavior on urban roads.The work of this thesis mainly includes the following contents:Design a trajectory prediction model A-Bi LSTM based on driving style and lane change intention.A-Bi LSTM aims to predict the future driving trajectories of vehicles around the EV.Based on long short-term memory（LSTM）,this thesis selects bidirectional LSTM（Bi LSTM）as the network architecture,and combines the attention mechanism to build the model A-Bi LSTM*.The input of the model A-Bi LSTM* only includes the driving sequence data of the vehicle,while the input of the model A-Bi LSTM also includes the driving style and lane change intention of the vehicle.In terms of experiments,the performance of A-Bi LSTM proposed in this thesis is compared with LSTM and A-Bi LSTM*.The results show that the model A-Bi LSTM is more accurate and reliable in predicting the vehicle trajectory,and its increased prediction time is within an acceptable range.Design a driving strategy planning model and algorithm P-DDQN based on deep reinforcement learning.P-DDQN aims to plan a safe and efficient driving strategy for the EV.Aiming at the lane changing scene of the urban two-lane highway environment,this thesis proposes a driving strategy planning method based on deep reinforcement learning.This method uses the above-mentioned trajectory prediction model to predict the future driving trajectories of surrounding vehicles,and then adds the trajectory prediction results to the state space of the agent,so that the observation information of the driving environment is richer and the driving vision is wider,so as to ensure the selection of safer and more efficient action.Finally,using highway-env,an autonomous driving simulation tool,the P-DDQN algorithm proposed in this thesis is compared with DDQN and DQN algorithms.The experimental results show that the P-DDQN can improve the average lane-changing pass rate and average driving speed of the EV in the lane-changing scene.