GK模糊分类算法在GIS局部放电模式识别中的应用

摘要

为了分析不同绝缘缺陷所激发的局部放电类型,在GIS内模拟了四种典型缺陷模型,根据局放信号与相位之间的关系,提取脉冲序列、幅值和相位信息,得到Hqmax - Phi、Hqmean - Phi及Hn - Phi等二维相位分布,然后利用统计参数偏斜度Sk、陡峭度Ku、峰值数量Pe及互相关因数CC等获取二维分布正负半周期的特征指纹.介绍一种新型Gustafson-Kessel (GK)模糊分类方法,根据特征指纹对四种缺陷进行分类,最后根据聚类有效性分析,验证了GK分类算法与模糊C-均值(FCM)分类方法都可达到较好的分类效果.%In order to analyze the partial discharge (PD) pattern of different defects in gas insulated switchgear (GIS), four common defects in GIS which are the floating electrode, protrusion, particle and void are proposed. According to the relation between PD and phase, the pulse sequence, amplitude and phase are extracted, and three two-dimensional phase distributions of Hqmax~Phi. The Hqmean~phi and Hn~phi are acquired. And based on above, the statistical parameters of skewness (Sk), kurtosis (Ku), number of amplitude (Pe) and cross coefficient (CC) are used to achieve the characteristic fingerprints of the positive and negative half phase distributions. At last, a new Gustafson-Kessel (GK ) fuzzy classification method is introduced to classify the four kinds of defects according to the fingerprints, and the cluster validity analysis proves that both the GK and fuzzy classification method (FCM) can achieve good classification results.