代价敏感相关向量机的研究及其在变压器故障诊断中的应用  

CS-RVM and its application in fault diagnosis of power transformers

作　　者：尹金良[1] 刘玲玲[2] 

机构地区：[1]天津理工大学自动化学院,天津300384 [2]天津理工大学工程训练中心,天津300384

出　　处：《电力自动化设备》2014年第5期111-115,共5页Electric Power Automation Equipment

基　　金：河北省自然科学基金资助项目(E2009001392)~~

摘　　要：实际中不同变压器故障类型的误分引发的危害程度往往不同,仅追求正确率并不一定会带来符合实际意义的分类结果。针对此,提出了代价敏感相关向量机(CS-RVM)。CS-RVM以误分代价最小为目标,按贝叶斯风险理论预测新样本类别。在用典型算例验证了CS-RVM具有代价敏感性的基础上,尝试将其应用于变压器故障诊断。基于溶解气体分析(DGA)数据的变压器故障诊断实例分析表明,CS-RVM全局诊断正确率略高于BP神经网络和支持向量机,略低于多分类相关向量机(M-RVM),但CS-RVM趋于提高误诊代价高的故障类型的诊断正确率,具有代价敏感性;CS-RVM的诊断速度足以满足变压器故障诊断的工程要求。Since different severities of damage may be induced by misclassification of transformer faults,the classification correctness alone may not be practically meaningful,for which,the CS-RVM（Cost-Sensitive Relevance Vector Machine） is proposed. It takes the minimum misclassification cost as its objective and applies Bayesian risk theory to predict the class of new sample. A typical case is studied to verify its cost sensitivity,based on which,it is adopted to the transformer fault diagnosis. The transformer fault diagnosis based on DGA（Dissolved Gas Analysis） shows that,the global diagnosis correctness of CS-RVM is slightly higher than that of BPNN（BP Neural Network） or SVM（Support Vector Machine） and slightly lower than that of M-RVM,while the diagnosis correctness of CS-RVM for the fault class with higher misdiagnosis cost is higher,showing its cost sensitivity. The diagnosis speed of CS-RVM meets the requirement of projects for transformer fault diagnosis.

关 键 词：变压器 代价敏感学习 相关向量机 误分类代价 故障诊断 

分 类 号：TM41[电气工程—电器]