A Multi-layer KMC-RS-SVM Classifier and DGA for Fault Diagnosis of Power Transformer

Sheng-wei      Fei; Yong      He

Abstract

Support vector machine (SVM) which can solve the problem of ‘over-fitting’, local extremum has great application perspective in fault diagnosis of power transformer. However, classification ability of SVM is influenced by the training sample including excessive attribute. In order to solve this difficulty, a new fault diagnosis method for power transformer based on K-mean clustering algorithm (KMC), rough sets(RS) and support vector machine(SVM) is presented in this paper. K-mean clustering algorithm is used to gain discrete data in diagnostic decision table, and decision rule is gained by rough sets. Sample attribute is simplified to construct optimal SVM model through information reduction approach of RS, then the optimal SVM model is used for fault diagnosis of power transformer efficiently and exactly. Finally, the effectiveness and correctness of this method are validated by the result of fault diagnosis examples.

Keywords: Dissolved gas analysis, fault diagnosis, K-mean clustering algorithm, knowledge reduction, power transformer, rough sets, SVM, Fault Diagnosis, Power Transformer Diagnostic Mode, Knowledge Acquisition