关键词: 燃料电池, 深度学习, 故障预测, 长短时记忆网络, 卷积神经网络

Abstract:

Proton exchange membrane fuel cells (PEMFC) are highly valued in fields such as vehicles, ships, and aerospace due to their advantages of being pollution-free, highly efficient, and low-noise. However, reliability issues have hindered the large-scale commercial promotion of fuel cells. To further enhance the reliability of fuel cells, this paper proposes a fault prediction method based on deep learning. First, data preprocessing methods such as standardization are used to eliminate the influence of different dimensions among monitoring parameters. Feature parameters are selected based on mechanistic knowledge to reduce the dimensionality of the original data and improve the computational efficiency of the fault prediction model. Then, a time series prediction model based on Long Short-Term Memory (LSTM) networks is constructed to predict the future operating state of the fuel cell. Finally, the predicted state data is input into a fault identification model based on Convolutional Neural Networks (CNN) to achieve fuel cell fault prediction. The proposed method is validated using experimental fault data from fuel cells. The experimental results show that the proposed fault prediction model can predict faults 10 time steps in advance with an accuracy of 96.8%.

Key words: fuel cells, deep learning, fault prediction, long short-term memory network, convolutional neural network