Fault Diagnosis of PEMFC Systems Based on Wavelet Packet Energy Decomposition and Long Short-Term Memory Neural Network

Abstract

The performance of proton exchange membrane fuel cell (PEMFC) fault diagnosis system plays a decisive role in the robustness of PEMFC system. Therefore, this study proposes a new fault diagnosis algorithm based on wavelet packet energy decomposition (WPD) and long short-term memory (LSTM) neural network, called WPD-LSTM. The input of WPD-LSTM algorithm is the sensor signal when PEMFC is running, and the output is the fault type of PEMFC (fault sequence number in normal state or multi-level and multi fault states). WPD plays the role of signal up sampling in the algorithm, which is used to mine the frequency domain characteristics of time-series sensor signals. LSTM is used to classify and diagnose the fault of the up sampled signal set. In this study, the effects of wavelet packet energy decomposition based on different wavelet bases are compared and analyzed, and the wavelet base with the best effect is applied to WPD-LSTM algorithm. Through the PEMFC bench fault diagnose experiment, the classification accuracy and recall of WPD-LSTM algorithm are 0.985 and 0.987 respectively, which are better than support vector machine (SVM), back propagation neural network (BPNN), recurrent neural network (RNN) and LSTM.