Improvement of power transformer insulation diagnosis using oil characteristics data preprocessed by SMOTEBoost technique

Abstract

This paper proposes a novel method for power transformer insulation assessment using oil characteristics. A hybrid algorithm, named as SMOTEBoost is implemented in the paper to improve the diagnosis accuracy and consistency. The SMOTEBoost can significantly enhance the generalization capability of artificial intelligence (AI) algorithms for transformer insulation diagnosis. This will provide important benefits for applying AI techniques in utility companies, i.e., an AI algorithm with its model built upon on a local dataset can be utilized globally to make transformer insulation diagnosis. The SMOTEBoost adopts Synthetic Minority Over-sampling Technique (SMOTE) to handle the class imbalance problem, in which data points belonging to different fault types or insulation conditions are unevenly distributed in the training dataset. By using this boosting approach for reweighting and grouping data points in the training dataset, the SMOTEBoost facilitates AI algorithms consistently attaining desirable diagnosis accuracy. To verify the advantages of SMOTEBoost algorithm, it is integrated with a number of representative AI algorithms including support vector machine (SVM), C4.5 decision tree, radial basis function (RBF) network and k-nearest neighbor (KNN) to make transformer insulation diagnosis using various oil characteristic datasets collected from different utility companies. A statistical performance comparison amongst these algorithms is presented in the paper.