The Machine Learning Method For Predicting Lane Changing Risk And Trajectory Of Intelligent Vehicle

Driver technology,driving habits,psychological state and traffic safety awareness are directly related to the safety of vehicle lane change,which is the main cause of traffic accidents.Intelligent vehicles with multiple high-performance sensors and intelligent control systems can effectively reduce the risk of vehicle lane change,but the lane change risk and lane change trajectory prediction method is complex,prediction accuracy and prediction timeliness is difficult to meet the requirements of high-speed vehicle driving conditions,for this reason,this paper carried out the relevant research,the main content is as follows:First,the I-80 highway in the NGSIM trajectory dataset is selected as the research object,and the Symmetric Exponential Moving Average(SEMA)method is used to pre-process this dataset,so as to lay the foundation for subsequent model training and validation.According to the characteristics of the dataset,the analysis was carried out with the free lane changing behavior of vehicles on the main roads.Secondly,according to the analysis of vehicle lane change process,four situations of collision between vehicles were summarized,and the Time Headway(THW)in the process of vehicle lane change was selected as the lane change risk index;in order to ensure the validity of the dataset,the second screening of the dataset was conducted,and the Pearson correlation coefficient method was used to correlate the seven selected lane change risk characteristics parameters;The XGBoost lane change risk prediction model was established using the above seven lane change risk characteristics parameters,and the data set was divided and normalized to meet the model training requirements.The results show that the XGBoost lane change risk prediction model has better performance in predicting lane change risk indicators,and the root mean square error of its prediction results is 0.2804,and the prediction accuracy and training speed of the model are better than those of the Random Forest algorithm,and the prediction performance of the seven feature parameter prediction model is better than that of the two feature parameter prediction model.The prediction performance of the seven-feature parameter prediction model is also better than the two-feature parameter prediction model.Finally,the CNN-LSTM lane change trajectory prediction model was established by combining the Convolutional Neural Network(CNN)and Long Short-Term Memory Network(LSTM),and the vehicle trajectory sequence prediction data was extracted to compare and The results show that the single time step input,five time step input and ten time step input used in the CNN-LSTM lane change trajectory prediction model can accomplish the prediction task,and their RMSE values are 0.3537,0.2479 and 0.0584 respectively,and It was found that the more historical trajectory sequence inputs,the higher the prediction accuracy;the prediction accuracy of the established CNN-LSTM model was higher than that of the LSTM lane change trajectory prediction model.