Energy Management Strategy for Plug-in Hybrid Electric Bus based on Improved Deep Deterministic Policy Gradient Algorithm with Prioritized Replay

Abstract

Deep reinforcement learning-based energy management strategy (EMS) is a state-of-art technology for hybrid electric vehicles (HEVs). This paper proposes a novel EMS based on improved deep deterministic policy gradient (DDPG) algorithm with prioritized replay for a power-split plug-in hybrid electric bus (PHEB) to improve the fuel economy of PHEB as well as the learning efficiency of DDPG. Firstly, prioritized experience replay is incorporate into DDPG to use samples more efficiently. Secondly, a real-world speed profile collected from a fixed bus route rather than short-distance standard driving cycles is used to train the improved DDPG algorithm until it converges completely. The superiority of the proposed EMS in terms of learning efficiency and fuel economy is validated under another real-world speed profile which is different from the training dataset. Simulation results indicate that the proposed EMS improves fuel economy by 3.22% and learning efficiency is improved significantly compared with the DDPG-based EMS.