Comparison of Classifiers for Power Quality Disturbances with Effective Features

Abstract

In this paper, the authors aim to investigate Power Quality (PQ) disturbance recognition accuracy using
effective and discriminative features. It computes a few discriminative statistical parameters from the
Stockwell Transform (ST) that can represent the PQ disturbances more accurately. A total of eight
synthetically generated PQ disturbances and the pure tone signal have been chosen for this purpose.
Several widely popular conventional classifiers such as the Naïve Bayes’ (NB), K-nearest neighbor
(KNN), Decision Tree (DT), Discriminant Analysis (DA), Support Vector Machine (SVM), and Random
Forest (RF) have been simulated using the set of extracted ST statistical parameters to compare the
recognition accuracy. The average recognition accuracy has shown to be highest for the RF classifier
whereas the KNN performance has been the poorest. However, the classification time for the RF and the
SVM remains highest as compared to other classification algorithms. Among the chosen classifiers, the
DT remains the fastest as revealed from our results.