ITransformer-Based High-Predictive-Ratio Lifetime Evaluation Method for PEM Fuel Cells Under Dynamic Operating Conditions

Life assessment is critical for the testing and maintenance of Proton Exchange Membrane Fuel Cells (PEMFCs). However, insufficient training data in initial tests and the complex aging mechanisms of PEMFCs have made it difficult to accurately predict their long-term performance and lifespan. To address this issue, a high predictive ratio lifetime evaluation method based on multivariate health feature fusion is proposed in this paper. Firstly, a feature selection method based on local interpretability is proposed to improving the input dimension and interpretability of the forecasting method under limited data. Secondly, a multivariate health feature fusion method is proposed to decompose and reorganize the nonlinear indicators of PEMFCs into health indicators that characterize long-term decay, aiming to reduce cumulative errors in long-term predictions. Thirdly, a long-term prediction framework based on Itransformer is presented to effectively capture the decay correlations among multiple variables, and improve the generalization ability of long-term lifetime prediction. Experimental results under multiple operating conditions demonstrate that proposed method achieves a lifetime prediction error of only 1.0460% for 96-step decay predictions under 20% dynamic condition training data, with significantly better real-time performance than the compared method, thereby validating its application value in early PEMFC testing.