A new optimal feature selection scheme for classification of power quality disturbances based on ant colony framework

Abstract

The optimal feature selection problem in classification of power quality disturbances is concerned with the elimination of redundant features. Its primary objective is to improve the classification accuracy and computational time. Limited literature on this topic and possible inefficiency in the existing works, makes it imperative to explore the new feature selection techniques. This paper, thus presents an ant colony optimization model for the same. A new multiobjective feature selection scheme is proposed which minimizes the product of feature set size and classification error. A feature set is declared feasible, if and only if it offers lesser error and features than other sets. The proposed scheme is compared against two other schemes from literature in same optimization framework. S-transform and time–time transform are employed for detection and feature extraction. This combination has a proven ability of offering precise time–frequency details and a wide range of features for optimal selection problem. Three classifiers, namely decision tree, k-nearest neighbour, and support vector machine are used to check the generality of proposed methodology. Disturbances simulated as per IEEE-1159 are used for initial testing, which confirm that the previous schemes may render accuracies even worse than those obtained without optimal feature selection. Conversely, the proposed scheme ensures a remarkable accuracy even when validated with real disturbances from an experimental database. A performance comparison with some recent works is also shown.