LAUSR

Abstract This paper presents an automated recognition approach for the classification of power quality (PQ) disturbances based on adaptive filtering and a multiclass support vector machine (SVM). Empirical wavelet transform-based adaptive filtering technique is suitable for nonstationary signals and therefore has been adopted to extract features of PQ disturbances. It primarily estimates the actual frequencies present in the signal by means of the fast Fourier transform following a divide to conquer principle. Second, a set of adaptive filters is designed in the frequency domain to extract the mono-frequency components of a distorted signal. Then six efficient features reflecting the characteristics of disturbances are extracted from these components as well as the signal. Lastly, these features are fed as inputs to a multiclass SVM for classification of the most frequent PQ disturbances. The PQ disturbances considered in this work include eight single disturbances and seven two-combination disturbances. The simulation results elucidate the efficiency and robustness of the proposed approach against noise and different degrees of disorder. The performance of the one-against-one and one-against-all approach based SVM classifiers is compared to determine the best in terms of recognition accuracy and computation time. Further, the classifier is also verified on a few measured disturbance signals.