基于M-SVR算法的变压器内绝缘老化状态研究  

Study on Aging State of Transformer Internal Insulation Based on M-SVR Algorithm

作　　者：韩志 HAN Zhi(Chengde Company,State Grid Jibei Electric Power Co.,Ltd.,Chengde 067000,China)

机构地区：[1]国网冀北电力有限公司承德供电公司,河北承德067000

出　　处：《自动化仪表》2023年第2期59-64,共6页Process Automation Instrumentation

摘　　要：为进一步提升电力变压器内绝缘状态的评估水平,采用对制备的绝缘纸样品在不同老化程度和不同水分含量情况下进行试验的方法,就样品介损因数与阻抗相位频域谱展开研究,研究样品聚合度(DP)值、含水量对频域介电谱(FDS)测试法参数的作用情况。构造多输出支持向量回归(M-SVR)算法模型,结合自组织映射(SOM)神经网络聚类分析情况,对径向基函数(RBF)神经网络就绝缘纸样品老化情况进行评估对比。结论如下:M-SVR算法可以实现高精准预测纸样中的水分含量,精度高于RBF神经网络;DP值对SOM聚类结果的作用伴随含水率升高而变小,且在含水率大于4.7%的时候所受影响几乎可以忽略;M-SVR算法对纸样老化情况判断较为准确,误差最低为8.54%。对M-SVR算法的针对性研究,对现场变压器内部绝缘水平判断给出了新方向。In order to further improve the assessment of the internal insulation condition of power transformers, experiments are conducted on prepared insulation paper samples with different aging levels and different moisture contents, and the frequency domain spectra of sample dielectric loss factor and impedance phase are studied to investigate the effects of sample degree of polymerization(DP) value and moisture content on frequency domain dielectric spectroscopy(FDS) test method parameters. The multi-output support vector regression(M-SVR) algorithm is constructed and combined with self-organrzing map(SOM) neural network clustering analysis. The radrical basis function(RBF) neural network is evaluated and compared on the aging of insulation paper samples. The conclusions are as follows: the M-SVR algorithm can predict the moisture content of paper samples with higher accuracy than the RBF neural network;the effect of DP value on the SOM clustering results decreases with the increase of moisture content and is almost negligible when the moisture content is greater than 4.7%;the M-SVR algorithm is more accurate in determining the aging condition of paper samples, with a minimum error of 8.54%. A targeted study of the M-SVR algorithm gives a new direction for judging the internal insulation level of transformers in the field.

关 键 词：变压器 多输出支持向量回归算法 绝缘纸 状态评估 频域介电谱 自组织映射 

分 类 号：TH183.3[机械工程—机械制造及自动化]