A Fast Fault Diagnosis Method of the PEMFC System Based on Extreme Learning Machine and Dempster-Shafer Evidence Theory

For purpose of solving the data-driven failure diagnosis problems of the proton exchange membrane fuel cell (PEMFC) system, improve the test accuracy and shorten the training time, a novel failure diagnosis method of the PEMFC systems based on data fusion is proposed, which combines extreme learning machine (ELM) and Dempster-Shafer (D-S) evidence theory. The characteristic vector extraction is carried out on the electrical quantities and the nonelectrical quantities of the PEMFC system under four different faults. The kernel ELM algorithm and online sequential ELM algorithm are, respectively, used to establish the failure diagnosis model of the PEMFC system based on electrical quantities and nonelectrical quantities. It is used for preliminary failure diagnosis of a PEMFC system. The diagnosis results of the above-mentioned two strategies are converted into the function values of the basic probability assignment by the squeeze function. The D-S evidence theory algorithm is used to fuse the diagnostic output at the decision level. The classification results of 154 samples of PEMFC system show that the novel model can diagnose four different degrees of high air stoichiometry failures. The average recognition rate is 98.79% and the operation time is only 0.2011 s. At the same time, the comparisons with the hack-propagation neural network and one-against-one support vector machine show that the data fusion algorithm can significantly improve the running speed while ensuring the correct recognition rate. It can be used for online failure diagnosis of the PEMFC systems.