基于谱聚类集成的变压器在线故障诊断

为了提高基于油中溶解气体分析技术(DGA)的变压器故障诊断准确率,本文提出了一种基于谱聚类集成的变压器在线故障诊断(TOFD-SCE)方法.以加权二次抽样算法抽取样本、构建基础谱聚类的样本集,以基础谱聚类学习问题的局部知识;平衡多样性与正确性选择集成成员;集成多个成员谱聚类的结果来提高变压器故障诊断的准确率.传统变压器故障诊断方法基于历史数据建立模型,缺乏在线学习的能力;TOFD-SCE利用历史与在线新增两种DGA数据来训练、修正模型,提高了其故障诊断的准确率.对SSP300000/500型变压器的故障诊断实验结果表明:TOFD-SCE的准确率优于IEC三比值法、BP神经网络法及支持向量机法,验证了其有效性.

The Transformer On-line Fault Diagnosis Based on Spectral Clustering Ensemble

To improve the accuracy of the transformer fault diagnosis based on dissolved gas analysis in oil(DGA),a transformer on-line fault diagnosis based on spectral clustering ensemble (TOFD-SCE) was proposed in this paper.The weighted double sampling algorithm create the samples set of the basic spectral clustering,which learned the local knowledge of the problems.The accuracy was improved by integrating the results of ensemble members,which were picked up form the basic spectral clustering in terms of the accuracy and variety.The conventional models are only trained by the historical data,and can't learn on-line.TOFD-SCE is trained and modified by both historical and new online data,and the accuracy is improved.The TOFD-SCE was validated by diagnosing the fault of SSP300000/500 transformers.Comparing with IEC three ratio,BP-neural networks and support vector machine,TOFD-SCE is more outstanding.