基于局部放电因子向量和BP神经网络的油纸绝缘老化状况诊断  

Diagnosis of Aging Condition in Oil-Paper Insulation Based on Factor Vectors of Partial Discharge and BP Neural Network

作　　者：周天春[1] 杨丽君[1] 廖瑞金[1] 汪可[1] 郑升讯[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

出　　处：《电工技术学报》2010年第10期18-23,共6页Transactions of China Electrotechnical Society

基　　金：国家重点基础研究发展计划资助项目(973项目)(2009CB724505-1)

摘　　要：模拟真实变压器的内绝缘运行环境,设计了一种单因子加速热老化试验。提取了局部放电脉冲相位分布(PRPD)模式的四个图谱及其对应的27个特征量,利用因子分析方法从27个特征量中提取了10个主成分因子。在局部放电10个主成分因子的基础上,选择BP神经网络对油纸绝缘的老化状况进行诊断。输入训练样本数据分别采用BP标准算法和四种改进算法对网络进行了训练,并对测试样本进行了老化诊断。基于10个主成分因子的BP网络在一定程度上能够诊断油纸绝缘的老化状况。各种算法的诊断结果表明L-M算法是比较合理的油纸绝缘老化诊断的BP网络算法。In this paper,thermally accelerated aging is studied experimentally for simulating the environment of the transformer insulation in service.Four statistical spectra and 27 characteristic values are extracted based on the phase-resolved partial discharge（PRPD） pattern of aging sample data.Thereafter,10 principal component factors are extracted from 27 characteristic values by factor analysis.Furthermore,BP neural network was chosen to diagnose the aging condition of oil-paper insulation based on 10 principal component factors.To train the network,the standard and four improved algorithms of BP network are compared with the training sample data,and then the trained network is used to diagnose the aging condition of the test sample data.It is concluded that BP network based on 10 principal component factors could be used for the aging condition diagnosis of oil-paper insulation to a certain extent.Moreover,it is suggested the L-M algorithm is the most reasonable algorithm for the aging condition diagnosis of oil-paper insulation according to diagnosis results of different algorithms.

关 键 词：油纸绝缘 局部放电 变压器 老化诊断 

分 类 号：TM835[电气工程—高电压与绝缘技术]