Online energy management strategy for fuel cell hybrid powertrain based on multi-objective constraint rules embedded in soft actor-critic learning

The fuel cell/battery hybrid powertrain offers a promising solution for fuel cell vehicles by integrating the high energy density of hydrogen fuel cells with the high-power density of batteries. However, real-time energy management of such a multi-source system faces challenges in simultaneously achieving economic efficiency, durability, and adaptability. To address this, this study proposes an online energy management strategy called MOCR-SAC. It incorporates multi-objective constraint rules (including hydrogen consumption, fuel cell degradation, battery degradation, fuel cell optimal efficiency deviation, and battery optimal state of charge deviation) within a Soft Actor-Critic reinforcement learning framework, enabling adaptive and intelligent power allocation. Evaluated on a 12-m fuel cell bus under standard Chinese driving cycles, MOCR-SAC reduces hydrogen consumption by at least 4.28 % and operating costs by 7.32 % compared to conventional SAC (without constraints or using single rules). It also outperforms other online reinforcement learning methods in component degradation, cost, battery SOC regulation, and hydrogen economy. Compared to the global optimum obtained by dynamic programming, its operating cost deviation remains within 4.50 %, while hydrogen consumption is 5.63 % lower. Under both deterministic and uncertain driving cycles, the total operating cost deviates by less than 10 %, demonstrating strong robustness and adaptability. The proposed strategy can be pre-trained offline and deployed online with minimal computational overhead, meeting the real-time requirements of vehicle energy management. In summary, MOCR-SAC significantly enhances the performance, efficiency, and durability of fuel cell hybrid powertrains, offering a practical and scalable solution for sustainable transportation.