基于LMD边际谱能量熵与FWA-SVM的变压器绕组松动诊断

Diagnosis for transformer winding looseness based on LMD marginal spectrum energy entropy and FWA-SVM

变压器空载合闸的振动信号包含了丰富的机械特征信息,为了实现对变压器绕组松动故障诊断,提出了一种局部均值分解(LMD)边际谱能量熵与烟花算法优化支持向量机(FWA-SVM)的方法。通过LMD提取若干反映变压器合闸过程绕组状态信息的乘积函数(Product Function, PF)分量;依据各PF分量相关系数与能量分布,将前6阶PF分量进行希尔伯特变换,并求取对变压器绕组状态变化敏感的边际谱能量熵作为特征向量;将特征向量输入到烟花算法(FWA)优化的支持向量机(SVM)分类器,实现变压器绕组轻微松动故障早期预警。实验结果表明:基于LMD边际谱能量熵能准确反映故障特征,FWA-SVM诊断方法在少量样本情况下仍有较高的故障辨识度。

The vibration signal of transformer no-load closing contains abundant mechanical characteristic information. In order to effectively realize the fault diagnosis of transformer winding looseness, a fault feature extraction method that combines local mean decomposition(LMD) marginal spectrum energy entropy and fire-works algorithm-support vector machine(FWA-SVM) is proposed in this paper. Several product function(PF) components that reflect the mechanical state information of operating process are extracted by LMD. The top 6 th order PF components are transformed by Hilbert transform based on the energy distribution and correlation coefficients of PF components and their marginal spectrum energy entropy are calculated as the feature vectors. The fire-works algorithm(FWA) is used to optimize the support vector machine(SVM) classifier to realize the early warning of transformer winding looseness. The experimental results show that the fault features based on LMD-Hilbert marginal spectrum energy entropy extraction can accurately reflect the fault characteristics, and the FWA-SVM diagnosis method has a good identification effect on the case of a small number of samples.