Lane-Changing Method Of Autonomous Vehicles Based On Game Theory

Lane-changing is one of the most common driving behaviors,which is also one of the main reasons of traffic jams and traffic safety accidents.The research on lane changing system of automatic vehicles is of great significance to improve lane-changing safety and relieve traffic pressure.In this thesis,the lane-changing decision model and trajectory planning model of automatic vehicles in the scene of free flow lane-changing are taken as the research objects.Combined with machine learning,a game decision model of lane-changing is proposed,which fully considers the real-time interaction behavior between lane-changing vehicles and surrounding vehicles.Based on the multi-objective optimization algorithm,a polynomial lane-changing trajectory planning model with safety,handling stability and ride comfort as optimization objectives is established.Firstly,the error analysis of trajectory data in NSGIM data set is carried out,and the data is cleaned by symmetric exponenitial moving average algorithm,which provides data support for the training and testing of the model in this thesis.Then,the game interaction behavior between the autonomous vehicle and the surrounding vehicles in the process of the free-flow lane-changing is analyzed,and a non-cooperative lane change game model under complete information is established.In order to obtain a more accurate game profit matrix,the relative distance,relative speed,acceleration,traffic flow,average speed and other features between vehicles on the current lane and the target lane are extracted from the NSGIM dataset as the input of the support vector machines classifier.Taking the classification probability of the support vector machines output as the game profit,the game profit matrix is established and the game Nash equilibrium is solved to obtain the optimal strategy of each vehicle in the scene of vehicle lane changing.Then,a quintic polynomial lane changing trajectory planning model and a cubic polynomial lane changing speed planning model are established in the Frenet coordinate system.Taking the ordinate and vehicle speed at the end of lane change as variables,a series of lane change trajectory clusters are generated and collision safety detection is carried out.Taking the longitudinal acceleration derivative,the yaw rate and the sideslip angle as the optimization objectives,the trajectory selection problem is understood as a multi-objective optimization problem.NSGA-II algorithm is used to seek the Pareto solution from the trajectory cluster to determine the final lane changing trajectory.Finally,the game-decision-making process is displayed and the simulation experiments are carried out to verify the trajectory planning results under the successful and failed changing conditions in Car Sim,Pre Scan and Simulink.