基于油中溶解气体分析数据挖掘的变压器绝缘故障诊断  

INSULATION FAULT DIAGNOSIS FOR POWER TRANSFORMERS BASED ON DISSOLVED GAS ANALYSIS DATA MININIG

作　　者：董立新[1] 肖登明[1] 李喆[1] 刘奕路[2] 

机构地区：[1]上海交通大学电子信息与电气工程学院,上海市200240 [2]美国弗吉尼亚理工大学电气工程系

出　　处：《电力系统自动化》2004年第15期85-89,共5页Automation of Electric Power Systems

基　　金：国家自然科学基金资助项目(50128706)

摘　　要：充分利用粗糙集理论对知识的约简能力与模糊径向基函数(RBF)神经网络优良的分类诊断能力,基于粗糙集与RBF网络实现数据挖掘的电力变压器绝缘故障诊断。该方法一方面将粗糙集作为RBF神经网络的前置,对经离散化的样本集进行约简,形成精简的规则集,将高于一定可信度的挖掘规则用于电力变压器故障诊断;另一方面,将粗糙集挖掘的低于可信度要求的规则所对应的挖掘样本,作为模糊RBF神经网络的训练样本集,同时将粗糙集对这些样本的聚类结果作为模糊RBF神经网络的聚类因子,在此基础上构建改进的4层RBF神经网络,用来诊断不能用粗糙集挖掘的规则诊断的事例。经检验,系统具有较好的分类诊断能力。In this paper, rough set and fuzzy RBF neural network (RBFNN) data mining based insulation fault diagnosis for power transformer is proposed. It makes full use of the knowledge reduction ability of rough set and the excellent classified diagnosis ability of RBFNN. On one side, the rough set is used as the front of RBFNN to simplify the samples set and to form the simplified rule sets. The mined rules, whose confidence is higher than the required, are used to offer fault diagnosis for power transformers. On the other side, the mining samples corresponding to the mined rules based on rough set, whose confidence is lower than the required, are used as training samples set of RBFNN and these samples' clustering result is used as the clustering gene of RBFNN. The improved four-layer RBFNN is proposed to diagnose the cases which can not be diagnosed by the mined simplified rules. Test results show the system has good classified diagnosis ability. This work is supported by National Natural Science Foundation of China (No. 50128706).

关 键 词：故障诊断 变压器 粗糙集 径向基函数神经网络 数据挖掘 

分 类 号：TM41[电气工程—电器] TM855