| The emergence of intelligent connected vehicles can improve people’s transportation efficiency,reduce environmental pollution and reduce the probability of accidents.In recent years,with the breakthrough of artificial intelligence,computer,communication and other technologies,the technology of intelligent connected vehicles has developed to a large extent,and it is a hot topic for scholars in various fields.However,there are still many problems that need to be solved in order to fully realize the safe and efficient driving of intelligent connected vehicles on the road.In the composition system of the intelligent connected vehicle,the decision-making system is the command center of the vehicle.A good decision-making method is very important for the vehicle.It is very challenging to develop a better decisionmaking algorithm.Signalized intersections are the main scenarios for intelligent connected vehicles to drive.The efficiency of vehicles at signalized intersections is greatly affected by their driving behavior on the upstream section.Therefore,the behavior decision method of the vehicle on the upstream section is of great significance to pass the signal intersection efficiently.The existing behavior decision-making research scenes are mainly concentrated in the interior of the signalized intersection,most of which are speed control of a few vehicles,and very few decision-making models for the upstream section of the signalized intersection.And there are fewer lane selection models,most of which are lane-changing decision-making or lanechanging execution models for known target lanes.Rare lane selection models do not consider the efficiency impact between multiple vehicles.Based on this,this paper specifically studies the behavioral decision-making problem of intelligent connected vehicles in the upstream section of signalized intersections.The behavior decision model of vehicles on the section before the signalized intersection and the lane selection decision model of multiple vehicles in the upstream section are established.And two models are simulated and analyzed.First,this paper analyzes the influencing factors that affect the decision-making of intelligent connected vehicles in front of signalized intersections,and then establishes a decision-making framework for the behavior of vehicles.Future driving states of intelligent connected vehicles and preceding vehicles are derived through the car following model.The condition of passing the intersection stop line is judged when the intelligent connected vehicle is the leader or the follower,and the lane change is add to the driving style to seek higher traffic efficiency.Secondly,this paper analyzes the lane selection process of multiple vehicles in the upstream section of the signalized intersection and factors that affect the lane selection decision.Considering the path demand,the driving efficiency and the impact of lane change,the lane selection Markov decision process is established,and the DQN reinforcement learning method based on multi-vehicles network sharing is used to optimize the lane selection strategy.The research and analysis results show that: the behavior decision model of vehicle on the road section before the signalized intersection can effectively predict states of preceding vehicles based on the signal countdown time,vehicle positions and speed states obtained by the Internet of Vehicles.Intelligent connected vehicles can make accurate judgments of the condition to pass through the intersection and make accurate choices between passing,stopping,and changing lane.The decision-making model for lane selection can help vehicles complete their driving tasks on the road section,vehicles can choose the optimal one between multiple lanes that satisfies the balance between itself and the system’s interest.Moreover,the model can improve multiple vehicles’ driving speed on the upstream section of the signalized intersection,and it can also reduce the length of queue before the signalized intersection,the efficiency of multi-vehicles is optimized on the whole. |
