基于改进模糊聚类算法的变压器油色谱分析  

Analysis of Transformer Oil Chromatography Based on Improved Fuzzy Clustering Algorithm

作　　者：李恩文[1] 王力农[1] 宋斌[1] 方雅琪 Li Enwen;Wang Linong;Song Bin;Fang Yaqi(School of Electrical Engineering Wuhan University Wuhan 430072 China)

机构地区：[1]武汉大学电气工程学院,武汉430072

出　　处：《电工技术学报》2018年第19期4594-4602,共9页Transactions of China Electrotechnical Society

摘　　要：变压器油色谱分析对变压器的运行和维护具有重要意义,聚类算法是油色谱分析的一种重要智能算法。但是传统模糊聚类算法(FCM)无法实现变压器油中溶解气体分析(DGA)数据的有效故障分类。该文针对传统FCM隶属度函数存在较多局部极值点的缺陷,重构了FCM的隶属度计算方法。通过构建指数形式的相似性函数,得到随距离单调变化的隶属函数,消除了隶属度函数的局部极值点;将相似性计算分为两个步骤,先根据样本每个属性计算子相似性,再融合得到样本的综合相似性,进而得到隶属度。实例分析表明,该方法提高了FCM进行DGA故障模式识别的能力,改善了算法的分类性能,具有重要的现实应用价值。Transformer oil chromatography analysis is of great significance for the operation and maintenance of the power transformers,and fuzzy clustering algorithm is an important intelligent algorithm for oil chromatographic analysis.However,the traditional fuzzy c-means algorithm(FCM)cannot achieve the fault classification of dissolved gas analysis(DGA)data effectively.The traditional FCM membership function has many local extreme points,and this is obstructive to DGA data classification.This paper reconstructed the membership degree calculation method of fuzzy clustering algorithm.The similarity function of exponential form was constructed,and the membership function with monotonicity of distance was obtained.This method eliminated the local extreme point of membership function.The similarity calculation was divided into two steps.First calculated the subsimilarity according to each attribute of the sample,and then merged them into the final membership.The example shows that the scheme improves the ability of fuzzy clustering algorithm to identify DGA fault pattern,and improves the classification performance of the algorithm.

关 键 词：模糊聚类分析 变压器 故障诊断 油中溶解气体分析 

分 类 号：TM72[电气工程—电力系统及自动化]