基于最大相关最小冗余准则的变压器故障诊断特征选择

Feature selection for transformer fault diagnosis based on maximal relevance and minimal redundancy criterion

电力变压器油中溶解气体分析(DGA)技术广泛应用于变压器内典型故障诊断,其中基于DGA数据的人工智能诊断方法在变压器故障诊断领域具有较高的识别率,但该类方法在选择故障特征量时尚无统一的标准。鉴于此,本文尝试引入最大相关最小冗余算法(mRMR),以互信息理论为基础挖掘变压器故障特征量之间以及特征量与故障类型之间的关联关系,通过分析大量的DGA在线监测数据挖掘出最优的变压器故障特征量集,并采用支持向量机(SVM)分类器对比优选特征量集和传统的特征量集合在变压器故障诊断的效率。最后,通过与SVM智能分类、IEC推荐的三比值分类方法的对比测试表明该方案的故障诊断准确率优于传统的故障诊断方案,故障识别效率高于新型的人工智能诊断方案,更适合于现场的工程应用及推广。

The power transformer oil dissolved gas analysis(DGA)is widely applied in the transformer fault diagnosis,and the artificial intelligent diagnosis method based on DGA has high data recognition rate in the field of transformer fault diagnosis.However this intelligent method has no unified standard in transformer fault selection.This paper tries to introduce the maximal relevance and minimal redundancy(mRMR)algorithm on the basis of the principle of mutual information.Based on the DGA on-line monitoring data,the mRMR get the optimal features set through analyzing the relationship between the features and the relationship between the features and the fault types.The SVM classifier is employed to compare the fault diagnostic effect with original feature set and the optimized feature set.Finally,compared with the SVM classification and the IEC recommended three ratio classification methods,the proposed fault diagnosis accuracy is superior to the traditional fault diagnosis method and recognition speed is faster than the intelligent diagnosis method,and this method is more suitable for engineering application on site.