Sensitivity-enhanced fault diagnosis in proton exchange membrane fuel cell via hybrid time convolutional network and bidirectional gated recurrent unit with multi-head attention

The reliability and durability of proton exchange membrane fuel cell (PEMFC) systems are critical for their widespread applications, with fault diagnosis playing a key role in enhancing performance. This study develops a hybrid model integrating the temporal analysis capability of time convolutional network (TCN), the sequence memory advantage of bidirectional gated recurrent unit (BiGRU), and the feature attention mechanism of multi-head attention (MHA). The proposed TCN-BiGRU-MHA framework enables precise signal pattern recognition through multi-component synergy. To address the challenges of limited sample sizes and imbalanced category distributions, this study employs a comprehensive data-processing strategy incorporating additive noise, systematic data panning, and data scaling. This strategy aims to effectively enrich the dataset, enhance the diversity of data samples, and establish a more robust training foundation for the model. Experimental results validate the effectiveness of the proposed TCN-BiGRU-MHA model in diagnosing unknown fault types under small-sample conditions, with an overall accuracy of 99% and F1 score of 98%, thereby evidencing its superior generalization capability and fault diagnosis performance. Finally, a sensitivity analysis is conducted to evaluate the impact of various hyperparameters on model performance, further demonstrating the robustness and reliability of the TCN-BiGRU-MHA model for practical applications.