模糊聚类算法参数优选方法及其在局部放电模式识别中的应用

Optimization Method of Parameter for Fuzzy Clustering Algorithm and Application in the PD Pattern Recognition for GIS

为了研究GIS中不同缺陷所激发的局部放电类型,设置了悬浮电极、针尖电晕、自由微粒以及气隙等4种常见缺陷模型。对采集的局放数据,建立了最大放电量、平均放电量和放电次数等二维相位分布函数,在此基础上提取24组指纹特征参数。在采用模糊C均值(FCM)和Gustafson-Kessel(GK)等聚类算法对局放数据进行聚类分析时,针对聚类有效性,即样本集的类数c和模糊加权指数m的优选问题,介绍了一种新型聚类有效性评估指标U(c),发现U(c)值越大,得到的聚类数越接近于真实值。最后与其它有效性指标对比,验证了U(c)指标的准确性和有效性。

In order to research the pattern of partial discharge（PD）excited from different defect models in gas insulated switchgear（GIS）,we prepared four kinds of defects,such as the floating electrode,protrusion corona,freedom particle and void.By adopting the PD signals,three two-dimensional distributions,such as the maximum discharge quantity,the mean discharge quantity and the amount of discharge,were achieved,thereby,24kinds of statistical fingerprint parameters were extracted.According to problems of the cluster validity,in the choice of the cluster number c and the fuzzy weighting exponent mfor the given PD samples when using the fuzzy clustering algorithms such as the fuzzy C-means（FCM）and the Gustafson-Kessel（GK）methods,we introduced a new evaluation indicator for cluster validity,namely,U（c）.The conclusion can be drawn that,the higher the value of U（c）is,the closer to the true value the cluster number c will be.The comparison with other validity indices verifies the efficiency and accuracy of the U（c）.