Knowledge Acquisition for Transformer Condition Assessment Using Synthetic Minority Over-sampling Technique and Decision Tree Algorithm

Abstract

This paper proposes a knowledge acquisition method for transformer condition assessment using Synthetic Minority Over-sampling Technique (SMOTE) and decision tree algorithm. SMOTE oversampling algorithm is utilized to supplement the minority samples. The transformer condition assessment is a classification process because its results are divided into four classes: normal condition, attentive condition, abnormal condition, and serious condition. Moreover, a decision tree is constructed using C4.5 algorithm to handle this classification problem. Key indicators and rules of transformer condition assessment are derived for the decision tree. The case study is performed on 110kV oil-immersed transformers using inspection and maintenance records. These records are collected from a power distribution company. The results indicate that the proposed method can automatically acquire the knowledge of condition assessment, which is capable for decision support.