| description abstract | Aiming at the evolutionary and simultaneous faults encountered in fuel cell applications, this paper proposes a fault diagnosis method for proton exchange membrane fuel cells that integrates 1DAlexNet, the self-attention mechanism, and the random forest algorithm. This approach incorporates deep learning and integrated learning techniques and utilizes a small number of features as diagnostic metrics, aiming to improve the accuracy of fault diagnosis, as well as reduce cost and simplify system design. The method initially extracts and identifies sparse feature information via deep convolutional networks and the self-attention mechanism, enabling the model to capture the nuances of faults more accurately. The distilled information is subsequently input into a random forest model, which capitalizes on the properties of ensemble learning, through a voting mechanism among multiple decision trees, to ascertain the definitive fault classification. The results demonstrate that this technique is highly effective in diagnosing evolutionary and simultaneous faults, offering a significant advantage in multifault scenarios. Meanwhile, it also exhibits high interference resistance and accuracy under different levels of noise interference. Furthermore, its generalization and superiority have been confirmed in fault diagnosis tests for both 100 kW evaporative cooling fuel cells and 80 W fuel cells. | |
