Research on PEMFC Water Management Fault Diagnosis Method Based on Probabilistic Neural Network and Linear Discriminant Analysis

In order to accurately and quickly identify the water management subsystem fault problem of proton exchange membrane fuel cell (PEMFC), a PEMFC water management fault diagnosis method based on probabilistic neural network (PNN) and linear discriminant analysis (LDA) was proposed. In this method, the dimensional effect between the original data parameters was eliminated by normalization, and the normalized variables were extracted by linear discriminant analysis. Not only can the original experimental data be mapped to the same interval, but also can effectively reduce the data dimension. The probabilistic neural network was used to implement water management fault diagnosis for 5-dimensional fault eigenvectors. The diagnostic results of 17086 sets of PEMFC water management failure samples show that the proposed method can effectively identify three health states of water management system: normal state, flooded fault, and membrane dry fault. The diagnostic accuracy of the training set and test set is 99.80% and 93.48% respectively, and the computing time is only 14.04 seconds. Compared with BPNN and LDA-BPNN, the prediction accuracy of the test set of the novel method is 17.47% and 2.75% higher than that of BPNN and LDA-BPNN, respectively, and the calculation time is saved by 39.83 seconds and 28.37 seconds, respectively. Therefore, the novel method can quickly and accurately diagnose PEMFC water management failure. © 2019 Chin. Soc. for Elec. Eng.