Fault diagnosis of proton exchange membrane fuel cell using multiple convolutional neural networks with multi-scale attention mechanism

To enhance the accuracy and robustness of fault diagnosis in proton exchange membrane fuel cells (PEMFCs), this study proposes a hybrid fault diagnosis model based on stacking ensemble learning. This model integrates multiple convolutional neural networks with a multi-scale attention mechanism (Stacking-MCNN-MSA). In the proposed model, MCNN is utilized to extract data features. A multi-head self-attention (MSA) mechanism is then applied to assign appropriate weights to these features. This process emphasizes critical information while suppressing noise. Subsequently, the MCNN-MSA component acts as the base learner. The predictions from the base learner are fed into the meta-learner to obtain the final fault diagnosis results. The research commences with data preprocessing, which involves crucial steps such as data noise reduction and data expansion. After that, the Stacking-MCNN-MSA model is constructed and evaluated through simulation experiments. Its performance is compared with that of alternative algorithms. The results demonstrate that the proposed model achieves high diagnostic accuracy under both original and noisy data conditions. Notably, after data expansion, the model attains a diagnostic accuracy of 98.67 %. These findings validate the effectiveness of the Stacking-MCNN-MSA model and provide a solid foundation for its practical application in PEMFC fault diagnosis.