基于概率神经网络和线性判别分析的PEMFC水管理故障诊断方法研究

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

为了准确迅速判断质子交换膜燃料电池(proton exchange membrane fuel cell,PEMFC)水管理子系统故障,提出基于概率神经网络(probabilistic neural network,PNN)和线性判别分析(linear discriminant analysis,LDA)的PEMFC水管理故障诊断方法。该方法采用归一化消除原始数据参量之间的量纲影响,使用线性判别分析对归一化后的变量进行特征提取。不但可以将原始实验数据映射到同一区间内,而且能有效降低数据维度。利用概率神经网络对5维故障特征向量实现水管理故障诊断。17086组PEMFC水管理故障样本的诊断结果表明:所提方法能有效识别正常状态、水淹故障和膜干故障共3种水管理健康状态,训练集和测试集的诊断正确率分别为99.80%和93.48%,运算时间仅为14.04s。与BPNN和LDA-BPNN相比:新方法测试集的预测精度分别高出17.47%和2.75%,计算时间分别节省39.83s和28.37s。因此,新方法能快速准确地诊断PEMFC水管理故障。

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.