Adaptive Hierarchical Energy Management Strategy for Hybrid Electric Vehicles Based on Deep Reinforcement Learning

In this work, the energy management control of parallel hybrid electric vehicles (HEVs) are investigated. In order to improve the fuel efficiency, an adaptive hierarchical energy management strategy is proposed, which novelly integrates an equivalent consumption minimization strategy (ECMS) with a deep reinforcement learning (DRL) method of a twin delayed deep deterministic policy gradient (TD3) algorithm. Compared to existing methods, the proposed method demonstrates superior efficiency in power demand allocation and achieves improved performance in fuel economy, while effectively maintaining reasonable variations of the battery state-of-charge (SOC). The effectiveness of the proposed approach is verified through numerical simulations with different driving cycles and is benchmarked against the existing methods.