基于网格搜索的改进SVM模拟电路故障诊断方法

针对模拟电路故障识别与诊断问题,提出了一种基于K最近邻的一对一SVM分类器(KNN-OSVM)的故障诊断方法。将K最近邻算法与用网格搜索法优化后的一对一SVM模型相结合,建立KNN-OSVM模型,有效解决了SVM因存在不可分域造成的误分问题,提高了故障诊断率。采用小波分析法提取输出端电压信号作为故障特征值,采用网格搜索对核函数、惩罚参数寻优。采用两个模拟电路进行仿真实验,并将改进的SVM与传统SVM进行对比。结果证明了该故障诊断方法的可行性。

An Improved SVM Analog Circuit Fault Diagnosis Method Based on Grid Search

Aiming at the fault identification and diagnosis problems of analog circuit, a fault diagnosis method based on K nearest neighbor and one against one support vector machine (KNN-OSVM) classifier was proposed. The KNN-OSVM model was established by combining the K nearest neighbor algorithm with the one against one SVM model optimized by the grid search method, which could effectively solve the problem of misclassification caused by the non-separable support vector machine and improve the fault diagnosis rate. The wavelet transform method was used to extract the fault features from the output voltage signals, and the grid search method was adopted to optimize the kernel functions and penalty parameters. The simulation experiments were carried out by two analog circuits. The improved SVM was compared with the traditional SVM. The simulation results showed the feasibility of the algorithm.