Prognosis of fuel cell degradation under different applications using wavelet analysis and nonlinear autoregressive exogenous neural network

Abstract

Abstract This paper presents the degradation prognosis of Proton Exchange Membrane Fuel Cell (PEMFC) operated under several conditions based on the combination of two types of data: data from postal fuel cell hybrid electric vehicles equipped with PEMFC and carrying out their postal delivery missions and PEMFC degradation data from laboratory. The prognosis is based on wavelet analysis and Nonlinear Autoregressive Exogenous Neural Network (NARX). The influences of historical state, operating conditions (load current, relative humidity, temperature, and hydrogen pressure), global degradation trend, and recovery phenomena on the degradation prognosis of PEMFC are considered. Firstly, the raw voltage degraded waveform of PEMFC is decomposed into multiple sub-waveforms by the wavelet analysis. Then, the degradation prognosis of each sub-waveform is made by NARX. Finally, the overall degradation prognosis of PEMFC is gotten by combing the degradation prognosis of each sub-waveform. Experimental results have shown that the novel prognosis method which exploits the two types of data results in a reliable model that covers PEMFC degradation over a wide range of operating conditions. The proposed prognosis method not only can make an accurate degradation prognosis of PEMFC with less learning data but also can use directly the raw experimental data with large fluctuation.