Energy Management Strategy for Hydrogen Fuel Cell Vehicles Based on LSTM Temperature Prediction and Adaptive ECMS

ABSTRACT

Conventional energy management strategies (EMSs) for fuel cell hybrid electric vehicles, such as the equivalent consumption minimization strategy (ECMS), typically neglect thermally induced irreversible degradation of fuel cells (e.g., membrane dehydration and catalyst aging) under high‐temperature operation. Moreover, temperature regulation is often treated as a passive response, making it difficult to simultaneously ensure fuel economy and thermal safety. To address this issue, this paper proposes an adaptive ECMS based on long short‐term memory (LSTM) temperature prediction (LSTM‐A‐ECMS). First, a high‐fidelity co‐simulation platform integrating electrochemical and thermal dynamics is developed. Then, an LSTM network is trained using multidimensional feature sets to predict future fuel cell temperature. The predicted temperature, together with the battery state‐of‐charge (SOC), is incorporated into the dynamic adjustment of the equivalent factor, enabling a transition from passive temperature feedback to predictive and coordinated optimization of economy and thermal safety. Under the CLTC driving cycle, compared with SOC‐based A‐ECMS and temperature‐feedback‐based T‐A‐ECMS, the proposed strategy reduces the peak fuel cell temperature to 81.91°C, achieving reductions of 2.6°C and 2.09°C, respectively. Meanwhile, the equivalent hydrogen consumption is reduced to 615.9 g/100 km, with improvements of approximately 0.9% and 0.4%. Under the WLTC cycle, the peak temperature is further reduced by 2.1°C and 1.0°C, while total hydrogen consumption decreases by about 1.9% and 1.7%. In addition, the final SOC remains within a reasonable range, indicating that the performance improvement is not achieved at the expense of excessive battery depletion. Hardware‐in‐the‐loop (HIL) experiments demonstrate that the proposed strategy achieves an average execution time of 3.4 ms per step, with key performance deviations below 0.5% compared to simulation results, confirming its real‐time capability and engineering feasibility. Overall, the proposed method effectively balances thermal safety and operational economy in fuel cell vehicles, providing a new perspective for prediction‐driven EMSs.