SYNTHESIZED DIAGNOSIS ON TRANSFORMER FAULTS BASED ON BAYESIAN CLASSIFIER AND ROUGH SET

As available testing data for transformer fauldiagnosis are incomplete and biased, and a Bayesian networkhas strong capability of processing uncertain information, NB(naive Bayesian) classifier model, TAN (tree augmented naiveBayesian) classifier model and BAN (Bayesian networkaugmented na?ve Bayesian) classifier model for transformersfault diagnosis are presented. To ensure the diagnosingcorrectness when there is shortage of several transformer testingdata, a new diagnosing approach, which integrates the Bayesiannetwork classifiers with rough set (RS), is proposed initially. Theapproach uses the results of dissolved gas-in-oil analysis (DGAand conventional electrical tests as the necessary attributes toclassify power transformer’s fault types. The relating hybridclassifiers are NB-RS, TAN-RS and BAN-RS, have strongability to deal with the lack of data, and have the error-tolerancecapability. So they have overcome the weakness of theover-rigidity of rough set based diagnosing approach. Thecomputing tests of diagnosing actual samples of transformefaults show that the diagnosing performance of the proposedhybrid approach prevails that of separated Bayesian networkbased classifiers and the rough set based approach.