用于高压断路器机械故障诊断的AM-ReliefF特征选择下集成SVM方法

Integrated SVM Method with AM-ReliefF Feature Selection for Mechanical Fault Diagnosis of High Voltage Circuit Breakers

针对高压断路器机械故障诊断过程中存在的原始特征维度过高,导致过拟合现象,而传统Relief-F算法在筛选特征过程中对诊断模型没有针对性,以及支持向量机(support vector machine,SVM)算法受参数选择限制导致诊断精度不佳的问题,提出一种AM-ReliefF特征选择下集成SVM的诊断算法。该算法对原始特征空间进行有效筛选,生成适应模型的最优特征子集,并将SVM作为基学习器与Ada Boost算法有效集成,提高诊断性能。首先对LW30-252型SF6高压断路器典型6种工况的合闸振动信号提取特征,构成原始特征空间,然后利用AM-ReliefF特征选择算法构造与集成SVM模型匹配的最优特征子集,最后用集成SVM模型进行故障诊断。与原始特征下单一SVM算法对比,所提方法使故障诊断精度由83.0%提升到98.9%,为高压断路器机械故障诊断研究提供了新思路。

For the high-voltage circuit breakers(HVCBs)mechanical fault diagnosis process,the original feature dimension is too high,which leads to over-fitting.However,the traditional Relief-F algorithm is not specific to the diagnosis model in the process of screening features,and support vector machine(SVM)diagnosis algorithm is limited by parameter selection,resulting in poor diagnosis accuracy.This paper proposes an integrated SVM diagnosis algorithm with AM-ReliefF feature selection.The algorithm sorts and effectively filters the original feature space,generates an optimal feature subset of the adaptive model,and effectively integrates SVM as a base learner with the AdaBoost algorithm to improve the accuracy of mechanical fault diagnosis of HVCBs.Firstly,features were extracted from the closing vibration signals of LW30-252 SF6 HVCB under six typical working conditions to form the original feature space,and then used the AM-ReliefF feature selection algorithm to generate the optimal feature subset matching the integrated SVM model,and finally used Integrated SVM model for diagnosis.The comparison with the diagnosis accuracy of the single SVM algorithm under the original features shows that the proposed method improves the fault diagnosis accuracy from 83.0%to 98.9%,which provides a new idea for the mechanical fault diagnosis of HVCBs.