Research on Active Lane Changing Decision and Trajectory Planning Methods for Freeways

Autonomous lane change technology of intelligent vehicle is one of the current research hotspots, and its development is very important to improve the driving safety of intelligent vehicle. In this paper, an autonomous lane change decision model based on GSCV-LightGBM and a multi-constraint autonomous lane change trajectory planning model based on quintic polynomial are proposed to optimize the safety, comfort and traffic efficiency of autonomous vehicle active lane change. The automatic lane change decision model uses simple moving average filtering technology to remove outliers from the public NGSIM data set, and uses grid search cross-validation algorithm to optimize the parameters of LightGBM model. Compared with other machine learning algorithms, GSCV-LightGBM model has the highest accuracy and the shortest decision time. The trajectory planning model based on quintic polynomial multi-boundary constraints divides vehicle free lane change behavior into barrier-free lane change and free lane change with obstacles based on safety distance threshold. The local path planning evaluation function is set respectively, and the lane change trajectory is evaluated twice by using the urgency degree to ensure the safety, comfort and traffic efficiency of the planned trajectory. Finally, the model predictive control method was used to control the trajectory of the simulation vehicle following the lane change planning through CarSim-Simulink co-simulation to verify the reliability of the lane change decision model and the trajectory planning model. The results show that the planned trajectory conforms to the multi-constraint conditions and can effectively improve the comfort, lane change efficiency and safety of the autonomous vehicle.