Adaptive hierarchical energy management strategy for fuel cell hybrid engineering vehicles based on deep reinforcement learning

Tianyu Li, Ming Li, Xindi Wang, Jianwei Cui, Jianyang Dong, Huiying Liu, Hai Xu,
Adaptive hierarchical energy management strategy for fuel cell hybrid engineering vehicles based on deep reinforcement learning,
International Journal of Hydrogen Energy,
Volume 168,
2025,
150985,
ISSN 0360-3199,
https://doi.org/10.1016/j.ijhydene.2025.150985.
(https://www.sciencedirect.com/science/article/pii/S0360319925039850)
Abstract: Fuel cell hybrid engineering vehicles have been known for their efficient energy use, making it crucial to formulate efficient energy management strategies. The present study proposes an adaptive hierarchical energy management system that integrates the equivalent consumption minimization strategy (ECMS) with deep deterministic policy gradient (DDPG), a leading data-driven reinforcement learning algorithm. Furthermore, a predictive model according to convolutional neural network combined with a gated recurrent unit and an attention (CNN-GRU-Attention) has been developed to accelerate the convergence of the proposed strategy. As shown by simulation comparisons with dynamic programming, DDPG, adaptive ECMS, and ECMS, this method attains a 94.8 % fuel efficiency of dynamic programming, surpassing the other three strategies. Hardware-in-the-loop tests indicate that DDPG-ECMS reduces equivalent hydrogen consumption by 17.4 %, 3.6 %, and 12.6 % compared to DPPG, adaptive ECMS, and ECMS, respectively, demonstrating significant fuel efficiency. This research establishes a basis for developing intelligent algorithms for energy management and improves energy conversion efficiency of hybrid energy storage systems.
Keywords: Energy management; Reinforcement learning; Fuel cell hybrid vehicles; Equivalent consumption minimization strategy