Power Quality Signal Classification using Convolutional Neural Network

Researchers are exploring challenging techniques for the analysis of power quality signals for the detection of power quality (PQ) disturbances. PQ disturbances have become a serious problem for the end users and electric utilities. Numerous algorithms have been developed for the classification of unstructured data. In this paper, the classification of power quality signals are performed based on autoencoders and convolutional neural networks (CNN). The work focuses on classifying power quality signals into known categories of waveforms such as swell, sag, harmonics and their combinations. Signals are fed to the autoencoder for the extraction of features, which are then classified using Support Vector Machine. This classification result is compared with another technique, where the signals are given to convolutional neural network and are subjected to classification using softmax regression. The second method gave an accuracy of 97%, which is superior to that obtained by an autoencoder.