Hybrid Health Indicator Based Fusion Ahead Degradation Prediction for Proton Exchange Membrane Fuel Cell

Vollständiger Abstract

Worum geht es in dieser Arbeit?

The decay state is the key indicator to reflect the health of a proton exchange membrane fuel cell (PEMFC). The existing static health indicators (HIs) fail to precisely represent the dynamic degradation of the PEMFC, resulting in low long-term forecasting precision of the data-driven methods. A hybrid HI and fusion ahead prediction method is proposed in this paper to achieve precise long-term decay prediction of PEMFC with limited data. Firstly, the RC-RLQ equivalent circuit equation is proposed to extract the polarization resistance from the electrochemical impedance spectroscopy data and form a hybrid HI with the voltage to accurately reflect the degradation behavior of the PEMFC under full operating conditions. Secondly, Empirical Mode Decomposition decomposes the hybrid HI into multiple Intrinsic Mode Functions to increase the inputs for Bidirectional Long Short-Term Memory (BiLSTM) in data-limited situations. Third, the sparrow search algorithm is employed to automatically optimize the optimal parameters of BiLSTM, which reduces the complexity of the prediction model and improves the model generalizability. Finally, a fusion ahead prediction method with hybrid HI is used to realize 10-step ahead decay prediction with limited data. The forecasting performance of the fusion ahead prediction algorithm is verified with real PEMFC data. With 185 h of training data, the 10-step RUL prediction error of the fusion ahead prediction method is only 0.63%. The results of the degradation prediction prove that the fusion ahead prediction method can precisely represent long-term degradation behavior of PEMFC, which is significant for the routine maintenance and control of PEMFC.