PEMFC Performance Prediction Based on Degradation Mechanism and Machine Learning

In transportation and portable applications, prognostic health management is essential to achieve long life and good performance of proton exchange membrane fuel cell (PEMFC). While the performance degradation mechanism of fuel cells (FCs) is complex and affected by multiple factors, achieving highly accurate health prediction remains a challenging problem. In this article, a long-term performance prediction method for FCs combining degradation mechanisms and machine-learning methods is proposed. First, the aging parameters characterizing the degradation of the catalyst, diffusion layer, and proton exchange membrane are estimated using the extended Kalman filter (EKF). Besides, considering the complexity of aging influences, sufficient correlation analysis, and variable selection are performed. Second, the relationship between external operating conditions and internal health characteristics is constructed by a multiobjective Gaussian process regression (MOGPR) algorithm. Finally, the aging path and remaining useful life (RUL) of the PEMFC are predicted under three operating conditions. The root mean square error is less than 0.024 V and 12.43 h. The results indicate that the proposed method can provide accurate PEMFC predictions.