Volume 3: Innovative Solutions for Energy Transitions: Part II

Deep Reinforcement Learning Based Energy Management of Hybrid Electric Vehicles with Expert Knowledge Renzong Lian, Jiankun Peng, Yuankai Wu, Huachun Tan, Hongwen He, Jingda Wu

Abstract

Reinforcement learning for energy management of hybrid electric vehicles has become a research hotspot. In this paper, a deep reinforcement learning (DRL) based energy management strategy (EMS) combined with expert knowledge is proposed, and an improved framework of deep deterministic policy gradient is adopted. In order to realize a reasonable tradeoff in the EMS, a multi-objective function of the fuel consumption and the battery charge-sustaining is established. In terms of action space of DRL, simplified action space, i.e. the optimal brake specific fuel consumption (BSFC) curve, is applied to the engine, thereby improving the sampling efficiency of DRL. The simulation results demonstrate that the expert knowledge can improve fuel economy and speed up convergence efficiency of the DRL based EMSs.

Keywords Energy management strategy, Hybrid electric vehicle, Weight assignment, Expert knowledge, Deep deterministic policy gradient.