基于支持向量机GIS局放小波包能量谱故障诊断

Fault Diagnosis of Wavelet Packet Energy Spectrum of Partial Discharge Pulse in GIS Based on Support Vector Machine

为了研究气体绝缘全封闭组合电器(GIS)在绝缘故障中如何区分故障类型进而判断故障源具体位置的问题,在实验室中设计了GIS高压导电杆上金属尖刺缺陷、盆式绝缘子表面尖刺缺陷2种物理模型,选取了脉冲电流法对局部放电信号进行量测。针对测量信号非平稳、频率高的特点,选择小波包变换将信号从时域转化为频域,细化频带以减少谐波及噪声对故障模式分量分离的干扰,计算主要频带能量并提取特征。在此基础上,融合了智能分类算法——支持向量机,将3个特殊频段能量占比代入支持向量机算法中进行训练并分类。实验结果表明,不同缺陷下能量谱图具有明显不同的特征。若将此故障在线诊断方法应用在GIS实际运行中,可以大大加强供电可靠性、运行安全性,减少人员工作量。且此方法具有时效性和一定的预见性,在GIS故障在线诊断中有较好的应用前景。

In an attempt to study how gas insulated switchgea(r GIS)can distinguish types of fault and specifically locate the source of fault in the insulation fault,two physical models of metal spike defects on the GIS high-voltage guide rod and disc insulator surface spike defects were designed in the laboratory.The pulse current method was selected to measure the partial discharge signal.Targeting the characteristics of non-stationary and high-frequency of the measurement signals,wavelet packet transform was adopted to convert the signals from time domain to frequency domain;the frequency band was refined to reduce the interference of harmonics and noise on the separation of fault mode components;energy of the main frequency band was calculated and features were extracted.Meanwhile,an intelligent classification method-the support vector machine(SVM)was incorporated,three special frequency band energy proportions were put into support vector machine algorithm for training and classification.The experimental results show that the energy spectrums of different defects have obviously different characteristics.If this online diagnosing method for faults is applied into the actual operation of GIS,the reliability of power supply and operation safety will be enhanced substantially while personnel workload will be lessened.With timeliness and some predictability,this method shows great prospect in online diagnosis of GIS faults.