基于多域特征融合及概率神经网络的GIS绝缘故障诊断  

GIS Insulation Fault Diagnosis Based on Multi-domain Feature Fusion and Probabilistic Neural Network

作　　者：彭曼 史钰潮 PENG Man;SHI Yuchao(Guangzhou Shike High-tech Co.,Ltd.,Guangzhou 510540,China;Guangzhou Zhixin Power Technology Co.,Ltd.,Guangzhou 510400,China)

机构地区：[1]广州市世科高新技术有限公司,广东广州510540 [2]广州致新电力科技有限公司,广东广州510400

出　　处：《电工技术》2024年第1期55-59,62,共6页Electric Engineering

摘　　要：目前GIS绝缘监测存在较高的误报率。针对该现象,对GIS内部绝缘机理进行研究,搭建了实验模拟平台,提出了等相位角的UHF数据格式技术,采用了短时能量法进行噪声过滤,有效提升了局放数据信息量。研究了基于统计量、时频等方法的局放数据特征提取技术,采用了概率神经网络分类器的机器学习方法建立诊断模型,实际诊断数据结果显示该方法具有极高的诊断精度。成果已应用于多个工程,应用效果良好。At present,GIS insulation monitoring has a high false positive rate.In view of this,the present work studied the internal insulation mechanism of GIS,built the experimental simulation platform,proposed the UHF data format technology with equal phase angle,and realized obvious increase in partial discharge information quantity by adopting short-time energy method for noise filtering.The feature extraction technology of partial discharge data based on statistics,time-frequency and other methods was studied,and the machine learning method of probabilistic neural network classifier was used to establish the diagnosis model.The actual diagnosis data results show that the method has high diagnostic accuracy,and the results have been applied to many projects with good application effect.

关 键 词：GIS绝缘故障 等相位角采样 机器学习 UHF 

分 类 号：TM855[电气工程—高电压与绝缘技术]