Transformer based lifetime interval prediction for dynamic operating proton exchange membrane fuel cells

Remaining useful life is crucial for proton exchange membrane fuel cell (PEMFC). However, the complex decay mechanism makes existing methods incapable of quantifying the PEMFC decay uncertainty. To address above issues, a hybrid interval prediction method (HIPM) is proposed. First, multi-feature fusion based on incremental empirical modal decomposition (IEMD) decomposes and reorganizes the nonlinear features of the PEMFC into multiscale degradation components. Second, the temporal Transformer effectively addresses the challenge of modeling long-term dependencies in PEMFC degradation prediction. Third, a novel interval prediction method precisely quantize the uncertainty of PEMFC degradation. Experimental results show HIPM achieves a root mean square error of 0.0047 with limited training data while accurately quantifying PEMFC degradation uncertainty across all conditions.