The superior environmental performance of proton exchange membrane fuel cells (PEMFCs) has led to their widespread application in transportation, distributed power generation, and other fields. Accurate degradation prediction of PEMFCs is crucial for reducing costs and enhancing the reliability of equipment operation. However, capturing the degradation details of PEMFCs from measurement data with high-frequency noise is a challenging task. To address this, this paper proposes a hierarchical pure convolution data-driven model, ConvTimeNet, which focuses on local degradation modeling while capturing the multi-scale dependencies between degradation data. Specifically, this study introduces a deformable patch layer to perceive the local patterns of temporal dependence units. The extracted local patterns undergo multi-scale dependency analysis on the designed hierarchical pure convolution blocks. Consequently, both local patterns and multi-scale dependencies are effectively modeled, thereby enabling health status monitoring of PEMFCs. The proposed method is also validated in multi-step forecasting scenarios; for a 64-step horizon, the MAPE is only 0.203%.

PEMFC; data-driven; health status monitoring; convolutional neural network; degradation modeling