A Prediction Method for Fuel Cell Degradation Based on CNN-LSTM Hybrid Model

As one of the most potential development directions for new energy application, the fuel cell has attracted much attention recently. Facing the bottleneck problems of durability, estimating the remaining useful life of fuel cell accurately is especially vital for its rapid and large-scale application. The paper proposed a degradation prediction method for fuel cell on the basis of Long Short-Term Memory (LSTM) neural network. To overcome traditional LSTM defects in feature extraction of multidimensional data, a Convolutional Neural Network (CNN) is also employed. Firstly, method extracts the feature and reduces the dimension of the original degradation data of fuel cell by CNN. Then it use Bi-LSTM to predict the degradation trend. 1154-hour experimental analysis of fuel cell degradation indicates that for the method the mean absolute error is 0.00223 and root mean square error is 0.00179. Compared with the method using LSTM with Kernel Principal Component Analysis (KPCA), it is verified the proposed method has great performance on predictive accuracy improvement of fuel cell degradation which could support follow-up health management of the system.