Short-term performance degradation prediction of a commercial vehicle fuel cell system based on CNN and LSTM hybrid neural network

Abstract

Short-term performance degradation prediction is significant for fuel cell system control and health management. This paper presents a hybrid deep learning method by combining the convolutional neural network (CNN) and long short-term memory (LSTM) network to predict the short-term degradation of a 110 kW fuel cell system used for the commercial vehicle. First, the complete ensemble empirical mode decomposition (CEEMD) is applied to decompose the nonlinear and non-stationary voltage sequence extracted by the sliding window into modality sequences with different characteristic time scales. Then, these modality sequences are input into the corresponding CNN-LSTM for voltage prediction. Experimental results show that the proposed CNN-LSTM can reduce the root mean square error (RMSE) by 13.55% and 34.40%, respectively, compared to the single CNN and LSTM because it combines the spatial feature extraction ability of CNN and the powerful prediction ability of LSTM. Furthermore, the CEEMD–CNN–LSTM can reduce RMSE by 36.92% compared to CNN-LSTM since the impact of exogenous factors on the recoverable decay and intrinsic decay of the fuel cell can be separated easily for better model learning. The CEEMD–CNN–LSTM is also compared with other recently published deep learning models based on the same data set, and the results show that the prediction framework in this paper has higher accuracy.