Development of an Energy Management Strategy for Multi-Stack Fuel Cell Hybrid Electric Vehicle Using Deep Reinforcement Learning

The integration of multi-stack fuel cells (FCs) is generating considerable interest in electrified vehicles due to the heightened levels of flexibility and resilience they offer. Efficient power distribution in such a multi-source system, given the distinct energetic characteristics of power sources like FCs and batteries, is a critical concern. This paper proposes a model-free power distribution approach between FCs and batteries using a Deep Reinforcement Learning (DRL) algorithm known as deep Q-network (DQN). The proposed DRL-based EMS aims to fulfill requested power requirements while minimizing hydrogen consumption costs and FC degradation. Additionally, to ensure an adaptive updating model of DRL that accommodates degraded FCs, online identification of FCs is integrated into the system, providing continual updates to the EMS during operation. To evaluate the effectiveness of the proposed EMS, the results are compared to a sequential quadratic programming (SQP) EMS. The results indicate that the total operating cost of the DQN is 46% lower compared to SQP.