油纸绝缘典型缺陷局放特性及缺陷类型识别  

Partial discharge characteristics and defect type identification of typical defects in oil-pressboard insulation

作　　者：池明赫[1,2,3] 夏若淳 罗青林 张潮海 曹津铭[1] 关毅 陈庆国[1] CHI Ming-he;XIA Ruo-chun;LUO Qing-lin;ZHANG Chao-hai;CAO Jin-ming;GUAN Yi;CHEN Qing-guo(MOE Key Laboratory of Engineering Dielectrics and Its Application,Harbin University of Science and Technology,Harbin 150080,China;TBEA Co.,Ltd,Changji 750306,China;School of Electrical Engineering and Automation,Harbin Institute of Technology,Harbin 150006,China)

机构地区：[1]哈尔滨理工大学工程电介质及其应用教育部重点试验室,黑龙江哈尔滨150080 [2]特变电工股份有限公司,新疆昌吉750306 [3]哈尔滨工业大学电气工程及自动化学院,黑龙江哈尔滨150006

出　　处：《电机与控制学报》2022年第2期121-130,共10页Electric Machines and Control

基　　金：国家电网公司总部科技项目(GYB17201800078)。

摘　　要：针对油浸式变压器纸质绝缘件内部缺陷导致局部放电和绝缘劣化的问题,结合三种典型缺陷纸板的局部放电图谱信号和深层神经网络的手段提出一种缺陷识别方法。根据不同类型油浸式变压器纸质绝缘件局部放电图谱信号的特点确定图谱的统计参数作为特征量。分析对比深层神经网络不同参数对识别效果的影响,寻找最优的深层神经网络结构。通过各类局部放电信号的特征量和深层神经网络来进行局部放电模式识别。研究结果表明:统计参数能够表征局部放电图谱信号的分布特征,优化深层神经网络可以提高模型的收敛速度和准确度。局部放电图谱信号的统计参数和深层神经网络相结合能够识别不同类型下油浸式变压器纸质绝缘件内部缺陷的局部放电信号,识别结果高于K-邻近法、支持向量机与反向传播神经网络。In view of the problem of partial discharge and insulation deterioration caused by internal defects in the oil-paper insulation of oil-immersed transformers,an identification method is proposed based on partial discharge signals of three typical defective oil-pressboards and deep neural networks.According to the characteristics of the partial discharge spectrum signals from different types of the oil-paper insulation in oil-immersed transformers,the statistical parameters of the spectrum were determined as the characteristic quantities.The influence of different parameters of the deep neural network on the recognition effect was analyzed and compared and the optimal deep neural network structure was found.The partial discharge pattern recognition was carried out through the characteristic quantities of various partial discharge signals and deep neural networks.The research results show that statistical parameters can characterize the distribution characteristics of the partial discharge pattern signals.Optimizing the deep neural network can improve the convergence speed and accuracy of the model.The combination of the statistical parameters from the partial discharge map signal and the deep neural network can identify the partial discharge signals of the different internal defects of the oil-paper insulation in oil-immersed transformers.And the recognition result is higher than the K-neighbor method,support vector machine and back propagation neural network.

关 键 词：油纸绝缘 局部放电 深层神经网络 模式识别 缺陷 

分 类 号：TM85[电气工程—高电压与绝缘技术]