2017年第45卷第6期 2017，Vol.45No.6
陕西电力
SHAANXI ELECTRIC POWER
特别推荐 Featured Focus
基于振动噪声SC与SVM的变压器局部过热故障诊断
许洪华1，李勇1，施恂山2.吴秋池2,马宏忠2(1.国网江苏省电力公司南京供电公司，江苏南京210019；
2.河海大学能源与电气学院，江苏南京211100）
摘要；为了更加及时准确地诊断变压器内部局部过热故障，提出了基于报动噪声SC与SVM的变压器过热故障诊断方法。首先，以稀疏编码(SC)算法为工具，对不同过热故障模型的频率信号进行预处理，提取出识别特征量，并作为后续支持向量机(SVM)学习输入量。然后，因SVM泛化能力受参数C和8影响较大，以均分训练样本所得多个模型的平均准确率作为评价函数，为SVM参数优选提供评价标准。通过10kV配电变压器过热模拟实验，生成训练样本和测试样本，并对SVM进行训练和测试。试验结果表明，采用[0,1]作为归一化方式，径向基函数作为核函数以及交叉验证(CV)寻优的CV-SVM模型，经学习后能够对局部过热不同故障程度与故障模型进行准确区
分。该研究为基于报动噪声在线监测变压器状态提供了参考。关键调：稀疏编码；支持向量机；局部过热；变压器：故障诊断
中图分类号：TM411
文献标志码：A
文章编号：16737598(2017)06001107
TransformerLocalOverheatingFaultDiagnosisBasedonSC and
SVMWithVibrationAnd Noise
XUHonghua',LI Yong',SHIXunshan?,WU Oiuchi?,MAHongzhong
(1. State Grid Jiangsu Nanjing Power Supply Company ,Nanjing 210019,China;2. College of Energy and
Electrical Engineering, Hohai University,Nanjing 211100,China)
Abstract: In order to diagnose intemal local overheating fault of transformers more timely and accurately, a fault diagnosis method of transformers overheating fault based on SC and SVM with vibration and noise is presented. Firstly, with the help of sparse coding(SC), frequency signals of different overheating fault models are preprocessed to extract feature quantities for diagnosis. These feature quantities can be used as leaming inputs of the following support vector machine (SVM). Then, parameter C and 8 have greater impact on the performance of SVM, so average accuracy rate of multiple models resulting from average training samples can be used as evaluation function to provide evaluation criteria for the parameter optimization of SVM.Overheating simulation tests of 10kV distribution transfomer are used to generate the training samples and testing samples, which are also used to train and test SVM. Finally experimental results show that the trained CVSVM model, which adopts [0, 1] as the normalized manner, radial basis funetion as the kernel fanction and cross validation (CV) as the optimization, can properly identify diferent fault degrees and fault types of local overheating, This study provides a reference for transformer onlinemonitoring conditions based on vibration and noise. Key words: SC; SVM; partial overheating; power transformer; fault diagnosis
与稳定。据统计，局部过热是造成变压器发生故障的
0
引言
电力变压器作为电力系统的核心设备之一，其
运行可靠性直接关系到整个区域电能传输的安全
基金项目：国家自然科学基金项目资助(51577050) 万方数据
一个重要原因1-2。当变压器内部导电回路或者磁回路出现局部过热，若不及时发现和处理，过热点温度可能达到上千度，会加速过热点表面的氧化腐蚀和形变，最终形成变压器过热事故。因此，有必要对其状态进行实时监测与诊断。
目前，油中溶解气体分析法（DissolvedGas AnalysisDGA)3-和电气试验法&-是对变压器内部