Research On Lane Changing Fusion Decision Of Passenger Cars Considering Driving Style

As China’s economy develops rapidly and the number of cars owned increases with each passing year,the road traffic safety environment faces enormous challenges.;The percentage of lane-changing traffic accidents caused by driver decision-making errors remains high every year.As the most complex stage in the driving process,lane change behavior should not only comprehensively consider the road and traffic flow information around the vehicle,but also the influence of driver style factors should not be ignored;Compared with the single-structure machine learning model,the vehicle lane change decision fusion of multiple machine learning algorithms can both enhance the model’s generalization ability and raise the precision rate of the vehicle lane change decision task;Therefore,introducing driving style factors into the vehicle fusion lane change decision model has important practical significance for improving road traffic safety.(1)NGSIM vehicle trajectory reconstruction.The cause of the error in the track of NGSIM data is analyzed;The discrete wavelet decomposition method is used to identify the outliers in the vehicle track data;Combined with the cubic Lagrange interpolation method,the identified outliers are interpolated and corrected;Based on the symmetrical exponential moving average algorithm(s EMA),the noise in the corrected track data is smoothed to complete the vehicle track reconstruction.(2)Analysis and sifting of lane change decision characteristics.The lane change behavior is divided into stages and the two key moments in the lane change process are identified;Determine the lane change type studied in this paper and eliminate the tracks such as forced lane change in the tracks;Analyze the influencing factors of lane change decision process and select reasonable decision characteristic variables;Finally,through the correlation analysis of each decision feature variable and the importance of the decision feature to the lane change behavior,the redundancy in the decision feature variable is removed,and the important decision features are screened out.(3)Clustering analysis of driving style based on lane changing characteristics.Using a fixed time window to extract trajectory fragments from the moment of intention to change lanes,select and solve parameters such as acceleration and impact that can characterize driving style;Using principal component analysis to reduce the dimensionality of driving style parameters;Based on K-means clustering analysis,driving style characteristics are divided,and driving style types are determined and labeled through parameters such as acceleration and impact.(4)Research on lane change decision model considering driving style.The vehicle lane change fusion decision model is constructed by combining the extreme gradient lifting tree(XGBoost)and the logical regression(LR);The decision feature samples are divided into training sets and test sets based on the cross-validation method of ten fold;The optimal value of super parameters in XGBoost model is determined by Bayesian optimization algorithm;Through model simulation comparison,the accuracy of the lane change decision-making task based on the driving style fusion decision-making model established in this article has been improved by about 4%,compared to other single structure models,which has been improved by about 2%.