Energy-efficient trajectory planning with curve splicing based on PSO-LSTM prediction

Abstract

Energy-efficient trajectory planning aims to optimize the economic performance for autonomous vehicles on the premise of ensuring driving safety, which excavate the energy saving potential and further improve the driving mileage. In this research, a curve splicing energy-efficient trajectory planning method based on surrounding vehicles trajectory prediction is presented. The long short-term memory (LSTM) neural network is adopted to construct the trajectory prediction model, and the hyperparameters of the LSTM are optimized by particle swarm optimization (PSO). To make the energy-efficient decision, the energy-efficient estimation model with motor MAP is developed by the correlation between vehicle driving energy consumption and motor efficiency, and the energy-efficient decision function was designed based on the average efficiency of behavior switching and the target behavior efficiency. Furthermore, a trajectory planning method with hierarchical planning of guide line and vehicle speed is presented based on B-spline curve and rolling dynamic programming (RDP). Via the traversal test, the dynamic adjustment of the guide line structure parameters is realized, and the RDP speed optimization objective function is designed with the goal of energy-efficiency. To precisely and rapidly control the EVs to track the reference trajectory, a model predictive control (MPC) with the goal of traceability was proposed. Eventually, the effectiveness of the energy-efficient trajectory planning algorithm is verified in the urban and the expressway condition respectively. The results show that the energy-efficient performance of the algorithm application is obvious in the expressway condition, and the average energy consumption improving rate is 11.11%.