A tolerant sequential correction predictive energy management strategy of hybrid electric vehicles with adaptive mesh discretization

Abstract

This paper establishes an adaptive correction predictive energy management strategy (EMS) to obtain optimal power distribution in the case of inaccurate prediction with excellent computational efficiency for parallel hybrid electric vehicle (HEV). Firstly, deep neural network (DNN) models are trained to forecast future speed sequence with different prediction horizons to prepare for online model predictive control (MPC) implementation. According to the degree of forecasting errors, a novel tolerant sequential correction algorithm as the solver of MPC strategy is selected in tuning mechanism instead of traditional dynamic programming (DP) to cope with the rough accumulated prediction. Besides, to compromise between fuel optimization and lower computation burden, statistical analysis of battery SOC variation distribution is gained from historical driving cycles' calculation through offline DP. Then, nearest neighbor interpolation model is fitted to generate optimal ranges of state variable in each step, which adaptive mesh discretization method is intelligently reducing state and control calculation grid's number. Numerical simulations demonstrate that the proposed tolerant sequential correction algorithm applied in MPC scheme with adaptive mesh discretization have yielded the favorable capability of the fuel economy in the consideration of inaccurately velocity prediction and excellent computational efficiency, which exhibits the practical adaptability in real driving routes.