Fuel cell fault classification based on long and short-term memory full convolutional neural networks

The development of modern automotive industry has accelerated the technological development and commercial application of fuel cells due to the challenges of resources and environment. The establishment of a sound failure prediction and management (PHM) system for hydrogen energy vehicles can achieve the goal of improving product quality and saving energy. Proton exchange membrane fuel cell (PEMFC) fault classification is the key to achieve the PHM system. The dataset used in this paper is realtime data collected on a live fuel cell vehicle. Considering the impact of unbalanced fault samples on fault classification accuracy, hybrid sampling is used in the data preprocessing stage to balance the number of samples, and a long and short-term memory full convolutional neural network is proposed to enhance the deep learning-based time series classification method by using global temporal attention and temporal pseudo-Gaussian enhanced self-attention. The experimental results demonstrate that the method in this paper has higher classification accuracy and precision compared with the traditional methods.