A novel hierarchical predictive energy management strategy for plug-in hybrid electric bus combined with deep deterministic policy gradient

Abstract

Energy management is a crucial technology to improve the energy economy of the plug-in hybrid electric bus (PHEB). This article proposes a novel hierarchical predictive energy management strategy combined with the deep deterministic policy gradient (DDPG) algorithm for superior energy economy performance and fast state of charge (SOC) reference planning of PHEB. In the upper layer, real velocity data collected from a fixed bus route are used to train the DDPG algorithm and the well-trained neural networks are extracted to plan the SOC reference trajectory efficiently before departure. In the lower layer, deep neural network (DNN) is employed to predict the velocity in a short term and a model predictive control (MPC) optimal controller is designed to allocate power flows optimally by tracking the SOC reference trajectory accurately. Simulation results show that the proposed strategy with high-efficiency SOC reference planning improves the energy economy by 4.32% compared with DDPG, and the energy economy achieves 98.61% of the global optimal algorithm. More importantly, the robustness and adaptiveness are validated in the case of imprecise velocity prediction and inaccurate pre-known driving cycles. This article contributes to the energy economy improvement for PHEBs through MPC and DDPG methods.