Feature selection in power transformer fault diagnosis based on dissolved gas analysis

Dissolved gas analysis is an important test to diagnose the condition of a power transformer. Based on key gases, various features are recommended for the purpose of fault classification. A feature selection method can be used to reduce the number of features by excluding less useful or irrelevant features. Selecting useful features not only reduces the computational complexity, but also enhances the classification performance. The novelty of this paper is to use various techniques of feature selection, including Student´s t-test, Kolmogorov-Smirnov test and Kullback Leibler Divergence test, to rank features´ order based on discriminative power of different features. The ordered features are tested with the K-Nearest Neighbour classification algorithm to evaluate their importance based on fault classification accuracy.