Intelligent energy management strategy for fuel cell hybrid vehicles utilizing deep reinforcement learning and driving condition recognition

Fuel cell hybrid vehicles (FCHVs) require efficient energy management strategies to improve fuel economy and ensure reliable operation under diverse driving conditions. In this study, an adaptive energy management strategy is developed using deep reinforcement learning. A gated recurrent unit model with speed prediction is employed to recognize driving conditions with 97 % accuracy. Based on the identified conditions, a deep deterministic policy gradient framework optimizes continuous power distribution between the fuel cell and battery systems. A tailored reward function is designed to reduce hydrogen consumption and stabilize the battery state of charge. Using real-world driving data for training and validation, the proposed strategy achieves a hydrogen consumption of 1401.38 g, representing a 5–8 % reduction compared with benchmark methods, while maintaining safe state-of-charge operation. These results demonstrate that the proposed method provides improved efficiency, adaptability, and robustness, highlighting its potential as a practical solution for fuel cell hybrid vehicles.