摘要
变压器是电力系统中重要的电气设备,其运行状态对系统安全运行起着重要作用,将代价敏感学习机制引入相关向量机,提出了代价敏感相关向量机(Cost-Sensitive Relevance Vector Machine,CS-RVM)。该算法以误诊损失代价最小为目标,按贝叶斯风险理论预测新样本的故障类别。用典型算例验证了CS-RVM具有较高的诊断正确率,同时可在一定程度上避免故障漏诊、高危故障误诊为低危故障。在此基础上,文章尝试将其应用于变压器绝缘状态评估,提出了基于CS-RVM的油浸式电力变压器故障诊断方法,以克服现有变压器故障诊断方法未考虑误诊代价差异的问题,并采用基于DGA数据的变压器故障诊断实例对该诊断方法的有效性进行了验证。
The transformer is an important electric equipment in the power system, whose running state plays animportant role in the safe operation of the system.This paper introduced thecost-sensitive learning mechanism to theRelevance Vector Machine, and put forward the term： the Cost-Sensitive Relevance Vector Machine （CS-RVM）.With the minimum misdiagnosis loss cost as the goal, the algorithm predicted the fault categories of newsamplesaccording to the Bayesian risk theory. With typical examples the paper verified the CS-RVM had highdiagnostic accuracy, and could to some extent avoid the misdiagnosis of faultor misdiagnosing high-risk failure aslow risk failure. On this basis, the paper tried to apply the CS-RVM to the insulation condition assessment of thetransformerproposed the fault diagnosis method of the oil-immersed power transformer based on the CS-RVM so asto overcome the problem that the existing transformer fault diagnosis method did not take into account the issue ofmisdiagnosis cost difference, and finally verified the effectiveness of the proposed diagnostic method by means of thetransformer fault diagnosis example based on the DGA data.
出处
《电测与仪表》
北大核心
2014年第12期29-33,119,共6页
Electrical Measurement & Instrumentation
作者简介
王斯妤（1994-），女，本科生，从事电气设备住线监测与故障诊断疗面的研究。Email：1446318897@qq.com.
