Voltage Degradation Prognostics for Commercial Proton Exchange Membrane Fuel Cell System Based on Transformer and its Variants

Abstract

Transformer and its variants demonstrate significant potential for predicting the performance degradation trends of proton exchange membrane fuel cells (PEMFCs), enabling accurate capture of degradation patterns and timely adjustments to control strategies to extend PEMFC lifespan. However, despite advancements in Transformer variants, their applicability for predicting PEMFC degradation trends remains unclear, particularly for the commercial high-power PEMFC, as existing research primarily focuses on small-scale laboratory stacks. This study addresses this gap by investigating a 60kW commercial PEMFC under two 1000-hour aging test modes with distinct hydrogen supply conditions (ambient temperature and low temperature). Using a characteristic current-based data extraction method, aging datasets were obtained for three characteristic currents. The single cell voltage was selected as the aging feature parameter to construct a Transformer model and four variants (Informer, Half-Transformer, Half-Informer, and Autoformer) for degradation prediction. Comparative analysis revealed that the Autoformer outperforms other models in aging voltage prediction accuracy. The robustness of Autoformer was further analyzed under multi-step ahead prediction, training set missing, and multivariate input conditions. Results demonstrate that the Autoformer maintains high prediction accuracy across diverse scenarios, with the cumulative distribution function of absolute prediction errors consistently within 5%. These findings highlight the Autoformer’s potential for integration into PEMFC control systems, offering promising applications in system health management to enhance practical value.

Keywords: Commercial fuel cell, two modes of aging test, Transformer variants, Voltage degradation prediction, Robustness