Remaining-Useful-Lifetime Prediction of Proton Exchange Membrane Fuel Cell Considering Model Uncertainty Quantification on the Full-Time Scale

A prognostics and health management (PHM) system with prediction at its core optimizes the durability of the proton exchange membrane fuel cell (PEMFC). However, the aging behavior model has some uncertainty due to limited knowledge, affecting the predictive performance in remaining useful life (RUL) prediction. To address this issue, an RUL prediction method based on the Bayesian framework considering uncertainty quantification on the full-time scale is proposed. First, the state of health (SOH) of the PEMFC is estimated, and the behavior of uncertainty is quantified. Afterward, a long short-term memory (LSTM) neural network is employed to make a prediction for its behavior. Finally, the RUL of PEMFC is predicted based on historical SOH and the predicted behavior of uncertainty. Validation indicates that the proposed method can make a long-term prediction and provide RUL prediction with high accuracy. Under the dynamic operating condition, in terms of long-term prediction, compared to unscented Kalman filter (KF), adaptive unscented KF, double-input-echo-state-network, and bidirectional LSTM, the proposed method decreases the error by 88.12%, 41.99%, 13.82%, and 3.21%, respectively. Also, under dynamic operating conditions, the proposed method shows good stability. Moreover, the robustness of this method has also been verified.