Research On Algorithms Of Lane-Changing Decision And Trajectory Planning For Smart Cars

With the development of computer technology and artificial intelligence technology,research on intelligent driving cars has attracted the attention of major universities and enterprises,and has become an important research content for scholars in related fields.Studies have shown that,among the causes of traffic accidents,lanechanging is the most common cause,so the importance of studying smart car active lane-changing systems is self-evident.The lane-changing decision and lane-changing trajectory planning play an important role in the smart car lane-changing system.The lane-changing decision module receives the traffic information from the sensing module and makes a reasonable lane-changing decision.After receiving the lanechanging information,the planning module plans the lane-changing trajectory,which provides a follow-up trajectory for the control module.Aiming at the lane-changing scene of smart cars,this paper designs a lane-changing decision algorithm and lane-changing trajectory planning algorithm for smart cars,which is based on the national key research and development project "N2/N3 pure electric commercial vehicle power platform key technology research and vehicle application".The lane-changing decision algorithm is based on a large number of realcar lane-changing data,and is developed through the XGBoost algorithm training in machine learning.The lane-changing trajectory planning algorithm is designed using a fifth-order polynomial function curve.This article first introduces the principles of the adopted lane-changing decision algorithm and lane-changing trajectory planning algorithm.Secondly,this article extracts lane-changing trajectory fragments based on the NGSIM data set collected by the US Federal Highway Administration.The NSGIM data set mainly includes vehicle number,vehicle global position information,vehicle speed information,vehicle acceleration information,time information,and so on.The data in the NGSIM data set is filtered by the symmetric exponential moving average filtering algorithm(s EMA).Based on the original information in NGSIM,this paper extracts the vehicle’s own driving state information and surrounding traffic vehicle driving state information,intercepts the lane-changing fragment set for vehicles with lane-changing behavior,and uses the clustering method of Kmeans to extract the effective lane-changing trajectory set,using a hybrid algorithm to mark the starting point of the lane change.Again,based on the above-extracted lane-changing trajectory set,this paper designs the XGBoost lane-changing decision algorithm based on Bayesian optimization.First,this paper analyzes the correlation between lane-changing decision variables and the contribution of lane-changing decision-making variables to lane-changing decisionmaking behavior,thereby determining effective lane-changing decision variables.Then,a decision-making simulation experiment was established,the XGBoost lane-changing decision model was trained based on the effective lane-changing decision variables,and the hyperparameters of the decision-making model were optimized by Bayes.On this basis,this paper designs a comparative verification experiment to compare this lane change decision model designed with other models,and verify that the lane change decision model designed in this article has improved prediction accuracy,recall rate and other indicators.Finally,the characteristic variables of the lane change decision model designed in this paper are optimized,which simplifies the complexity of the model and improves the generalization ability of the model.This paper also designs a lane-changing reference trajectory planning algorithm based on multi-objective optimization,according to the lane-changing instructions generated by the decision model.This paper uses a fifth-degree polynomial to plan the lane-changing trajectory set,combined with the lane-changing collision detection method,and comprehensively considers the comfort of the lane-changing process,the efficiency of the lane-changing,the impact of the lane-changing on the traffic,and the risk degree of the lane-changing process,to design the value function for evaluating lane changing trajectory.In order to verify the planning effect,this paper carried out a simulation experiment of lane-changing trajectory planning,established simulation working conditions,planned the set of lane-changing trajectories according to the working conditions,and selected the optimal lane-changing trajectory with the minimum value of the value function.Besides,this paper designed an optimization experiment to optimize the weight of each indicator in the value function.After optimization,the value of each indicator and the value function of the lane-changing trajectory are all small.