To enhance the reliability and maintenance efficiency of fuel cell systems during long‐term operation, this study proposes a bidirectional long short‐term memory–gated recurrent unit (DI‐Bi‐LSTM‐GRU) prediction model that combines calculus‐based feature enhancement with a transfer learning strategy. The model uses single‐channel voltage sequences as input and incorporates differential and integral features to characterize variation rates and cumulative degradation trends. A transfer mechanism initializes target domain parameters using prior degradation knowledge, enabling early‐stage degradation prediction with limited data. Experiments conducted across multiple operating conditions and fuel cell types demonstrate that the model achieves high prediction accuracy and robustness. It effectively captures and responds to transient behaviors, such as voltage recovery and sudden drops. Specifically, under the dynamic cycle‐9.4A dataset, root mean squared error is reduced by 55.49% compared to Bi‐LSTM‐GRU. On stable datasets, the model achieves RMSE of 0.00295 with only 150 h of training data, representing an 85% improvement over nontransfer models. Moreover, compared with current mainstream methods, the proposed framework shows significant superiority. These results demonstrate the method's potential for real‐time deployment and early health monitoring of fuel cells.
Yin et al. (Wed,) studied this question.