Remaining useful lifetime prediction methods of proton exchange membrane fuel cell based on convolutional neural network-long short-term memory and convolutional neural network-bidirectional long short-term memory

As a promising energy conversion device, the proton exchange membrane fuel cell (PEMFC) has been widely used in many fields. However, its commercialization is limited by its useful lifetime, so it's very important to predict the remaining useful lifetime (RUL). In this paper, an RUL prediction method of PEMFC based on convolutional neural network (CNN) and long short-term memory (LSTM) is proposed. First, for data processing, we use Savitzky-Golay to smooth the datasets, a box plot to remove the outliers, and Z-score to normalize the datasets. Then, we perform experiments on different lengths of time series data to find the best parameters and test the generalization ability of the model to long-term and short-term forecasts. Eventually, the results indicated that CNN-LSTM and CNN-bidirectional LSTM (CNN-BiLSTM) can get very accurate predictions with the relative error values of CNN-LSTM being 0.07% and CNN-BiLSTM only 0.03%. Furthermore, we discover that the training and prediction speed of the models are improved due to the addition of CNN. Therefore, we can quickly and accurately predict the RUL of PEMFC in the long term and short term.