Research On Vehicle Lane Change Trajectory Prediction Considering Lane Change Intention And Style

At present,the related technologies of smart transportation are developing rapidly,which can effectively prevent traffic accidents and improve driving safety.The lane changing behavior of drivers is one of the main vehicle behaviors that lead to traffic accidents,and accurate prediction of vehicle lane changing trajectories can effectively reduce traffic risks.The lane changing behavior of vehicles is not only influenced by the driving status of the vehicle,but also by the driver’s intention and style of lane changing.The accuracy of trajectory prediction using only vehicle driving status data is relatively low.To address such issues,it is necessary to consider various factors that affect vehicle travel for lane change trajectory prediction research.In this paper,the prediction model of vehicle lane change trajectory considering lane change intention and style is established based on Long Short Memory Neural Network(LSTM),and the impact of lane change intention and style on the prediction accuracy of vehicle lane change trajectory is studied.The main research contents are as follows:(1)Establish a lane change trajectory prediction model using vehicle driving status data.Based on the LSTM model and utilizing vehicle coordinates,vehicle speed,and acceleration data,a PV-LSTM model is constructed to compare the prediction effects of different hidden layer node numbers and select a network structure with higher accuracy for subsequent research.Based on the historical data of the vehicle,different time domain lane change trajectories were predicted.Within the 5-second prediction time domain,the RMSE and FDE values of the model prediction results were 0.96 m and 1.42 m.(2)Study the impact of lane changing intention on the accuracy of vehicle lane changing trajectory prediction.The RF-LSTM model is established by adding an intention recognition module based on random forest algorithm to the PV-LSTM model.This module can transform the driver’s intention recognition into a binary classification problem after training using vehicle trajectory data,and the recognition accuracy of lane changing intention reaches 93.89%.Compared with the PV-LSTM model,the RF-LSTM model with the addition of lane change intention recognition module has an average prediction accuracy improvement of 19.07% in the 5-second prediction time domain,and an improvement of18.71% in the endpoint prediction accuracy.(3)Study the impact of lane changing intention and style on the accuracy of vehicle lane changing trajectory prediction.On the basis of the RF-LSTM model,a style recognition module was added to establish the RFBP-LSTM model.This module is based on BP neural network and optimized using genetic algorithm.The trained and optimized style recognition module can recognize the cautious,ordinary,and aggressive styles of drivers,with an accuracy of 97.71%.Compared with the RF-LSTM model,the RFBP-LSTM model,which considers the intention and style of lane changing,has an average prediction accuracy improvement of 10.65% in the 5-second prediction time domain,and an improvement of12.47% in the endpoint prediction accuracy.