油纸绝缘电热老化的聚类判别算法  

Clustering Algorithm for Electrical-thermal Aged Oil-paper Insulation

作　　者：张锋[1,2] 李剑[1] 杨丽君[1] 廖瑞金[1] 王小维[1] 

机构地区：[1]重庆大学输配电装备及系统安全与新技术国家重点实验室,重庆400044 [2]杭州萧山供电局,杭州311200

出　　处：《高电压技术》2009年第5期1020-1025,共6页High Voltage Engineering

基　　金：国家自然科学基金(50425722)~~

摘　　要：油纸绝缘老化诊断是保障充油电力设备安全运行的重要手段之一。为研究采用局部放电参数进行油纸绝缘老化诊断的新方法,设计了新的油纸绝缘缺陷模型模拟变压器匝间绝缘缺陷,通过电热联合老化试验,采集老化过程中缺陷模型的局部放电信号及绝缘纸样本聚合度。采用因子分析方法,从传统的27个局部放电统计算子中提取一组新的9个局部放电主成分因子向量,组成统计矩阵进行聚类及老化状态的判别。采用c-均值聚类算法和模糊c-均值聚类算法分别对两种向量进行聚类分析,聚类结果表明,基于局部放电统计算子和其主成分因子,采用模糊c-均值聚类算法对油纸绝缘老化状态识别,都可获得稳定的识别结果,正确识别率均>70%,优于c-均值聚类算法。Aging diagnosis to oil paper insulation is an effective approach to ensure security of oil-filled electrical power equipment in operation. We established a new oil-paper cavity model to simulate the turn-to-turn insulation defects in transformers, in order to study a novel approach of aging diagnosis to oil-paper insulation based on partial discharge （PD） parameters. Aging experiment of this model under electrical and thermal stresses were taken and partial discharges and degree of polymerization of specimens were measured during the experiment. 9 principal parameters were extracted from the 27 statistical operators by factor analysis. The PD parameter vectors composed by either the statistical operators or the principal parameters were clustered by two types of clustering analysis methods, the c-mean clustering ,and the fuzzy c-mean clustering. The clustering results show that the robust clustering results can be obtained and the correctness ratio exceeds 70 percent by using the fuzzy c-mean clustering with both the PD statistical operators and principal parameters. FCM shows its advantages to the c-mean clustering method for oil-paper aging diagnosis.

关 键 词：油纸绝缘 局部放电 电热老化 缺陷模型 聚类算法 因子分析 

分 类 号：TM855[电气工程—高电压与绝缘技术]