基于半监督分类方法的变压器故障诊断

Transformer Fault Diagnosis Based on Semi-supervised Classifying Method

变压器油中溶解气体分析(DGA)是电力变压器故障诊断的重要方法。针对现有方法建立分类器时需用到大量已知类别数据而没有利用待分类数据的问题,将半监督分类(SSC)方法引入变压器故障诊断问题中,建立了一个新的变压器故障诊断模型。SSC方法在学习过程中能同时利用已知类别数据和未知类别数据,获得更多的信息,因而有更好的学习效果。采用模糊近邻标签传递的半监督分类(FNNLP-SSC)方法进行变压器故障诊断,所提方法依据样本与其K个近邻的模糊相似性连接,使类别标签从标签数据向未标签数据传递,最终实现未标签数据的分类。对故障DGA样本的诊断实例结果表明,所提FNNLP-SSC方法比模糊C均值(FCM)方法和IEC 3比值法有更高的诊断正确率,验证了所提方法在变压器故障诊断中的有效性和可行性。

Dissolved gas analysis（DGA） of oil in transformer is one of the most important methods to diagnose fault of power transformer.Most of existing diagnosis models will need large amount of labeled data to construct classifiers,while normally ignoring without unlabeled data,thus the semi-supervised classifying（SSC） method is introduced to build a new fault diagnosis model for power transformers.In its learning process,the SSC consider both labeled data and unlabeled data to get more knowledge,i.e.it learns better.A SSC method adopting fuzzy nearest neighborhood label propagation（FNNLP-SSC） is adopted for actual fault diagnosis of power transformers.In this method,based on the similarity connections between a sample and its K nearest data,the model classifies the unlabeled data by making the labels propagate from the labeled data to unlabeled data.Results of diagnosing a DGA sample of fault show that the proposed FNNLP-SSC method performs better than the fuzzy C-mean（FCM） method and the three ratios method of IEC,so it is concluded that the proposed method is feasible and valid for diagnoses of transformer fault.