基于概率神经网络的油纸绝缘老化诊断技术研究  

Research on Ageing Diagnosis Technology for Oil-paper Insulation Based on Probabilistic Neural Network

作　　者：孙长海[1] 李天明 陈百通 郭佳彬 鞠爽 SUN Changhai;LI Tianming;CHEN Baitong;GUO Jiabin;JU Shuang(School of Electrical Engineering,Dalian University of Technology,Dalian 116024,China;School of Control Science and Engineering,Dalian University of Technology,Dalian 116024,China)

机构地区：[1]大连理工大学电气工程学院,辽宁大连116024 [2]大连理工大学控制科学与工程学院,辽宁大连116024

出　　处：《绝缘材料》2021年第6期107-115,共9页Insulating Materials

摘　　要：对油纸试样进行加速热老化处理,根据聚合度变化将其老化过程划分为5个老化阶段;基于气隙放电模型进行局部放电试验,采集不同老化阶段油纸试样的PRPD图谱;利用统计算子提取特征量,采用因子分析法(FAM)对原始特征数据降维,比较降维前后特征数据的聚类特性;建立概率神经网络(PNN)模型识别油纸绝缘的老化阶段,作为对照,搭建反向传播(BP)神经网络模型以及支持向量机(SVM)模型,使用相同的数据对其进行训练,比较三者的识别结果。结果表明:老化会导致纸板内部产生孔隙,从而促进局部放电的发生;与其他模型相比,FAM-PNN模型在识别准确率和运算效率上具备明显优势,使用FAM-PNN模型可以准确高效地对变压器油纸绝缘的老化状态进行评估。An oil-paper sample was conducted accelerate thermal ageing treatment,and its ageing process was divided into five ageing stages according to the variation of polymerization degree.Partial discharge tests were conducted on the air gap discharge model,and the PRPD patterns of the oil-paper sample were collected at different ageing stages.The feature quantities were extracted by using statistical operator,the dimension of the original feature data was reduced by factor analysis method,and the clustering characteristics of the feature data before and after dimension reduction were compared.A probabilistic neural network model(PNN)was established to identify the ageing stages of oil-paper insulation,and a back propagation(BP)neural network model and a support vector machine(SVM)model were built as comparison.The three models were trained by the same data,and their recognition results were compared.The results show that ageing will cause pores in the pressboard,which promotes the occurrence of partial discharge.Compared with other models,the FAM-PNN model has obvious advantages in recognition accuracy and operation efficiency.The ageing state of transformer oil-paper insulation can be evaluated accurately and efficiently using the FAM-PNN model.

关 键 词：油纸绝缘 气隙放电 老化阶段识别 因子分析法 概率神经网络 

分 类 号：TM215[一般工业技术—材料科学与工程]