基于PSO-RBF和PCA-PSO-PNN的六脉冲整流器故障诊断研究

Research on fault diagnosis of six-pulse rectifier based on PSO-RBF and PCA-PSO-PNN

针对六脉冲整流器器件故障诊断准确率较低的问题,提出了由两种优化的神经网络相结合的故障诊断方法。对六脉冲整流器进行仿真并模拟其故障情况,采集其输出电压经傅里叶变换的直流分量幅值和谐波幅值。选用粒子群算法(PSO)优化径向基函数神经网络(RBF),运用PSO-RBF对整流器器件的故障大类进行诊断。然后采集在各故障大类下整流器的输出电压值,由主成分分析算法(PCA)和PSO优化概率神经网络(PNN),运用PCA-PSO-PNN对故障大类中的故障小类进行进一步诊断。再将此方法与PNN、RBF以及结合RBF和PNN的诊断方法进行实验对比。结果表明,这些方法的测试集准确率分别为91.5955%、67.6768%、16.1616%、87.8500%。故PSO-RBF和PCA-PSO-PNN相结合的诊断方法可以更精确地对六脉冲整流器器件的故障类型进行诊断。

Aiming at the low fault diagnosis accuracy of six pulse rectifier devices,a fault diagnosis method combining two optimized neural networks is proposed.The six pulse rectifier was simulated and its fault condition was simulated.The amplitude of dc component and harmonic wave of its output voltage was collected by Fourier transform.The radial basis function neural network(RBF)was optimized by particle swarm optimization(PSO),and the fault classification of rectifier device was diagnosed by PSO-RBF.Then,the output voltage values of the rectifier in each fault class were collected,and PCA-PSO-PNN was used to further diagnose the fault subclasses in the fault class.This method was compared with PNN,RBF and the diagnostic method combined with RBF and PNN.The results show that the accuracy of these methods are 91.5955%,67.6768%,16.1616%and 87.8500%respectively.Therefore,the combined diagnosis method of PSO-RBF and PCA-PSO-PNN can diagnose the fault types of six pulse rectifier devices more accurately.