Belief network classifier for evaluation of DGA data of transformers

A method to improve the assessment capability of power transformers by using belief network classifier is proposed. Two different belief network classifiers, the Naive Bayes classifier and the tree augmented Naive Bayes classifier, are compared using utilities' DGA data analysis. Their respective advantages and shortcomings are also shown by the detailed comparison. More than hundreds of historical DGA data has been used to demonstrate the capability of the method. Classification results show that the two classifiers are suitable for interpretation of DGA data and for diagnosis of incipient faults in transformers.