An image-based deep transfer learning approach to classify power quality disturbances

Abstract

Power quality disturbances (PQDs) consist in deviation of voltage and current waveforms from the ideal sinusoid at fundamental frequency, and need to be monitored to ensure a reliabile electrical supply. While, traditionally, power quality monitoring has been performed using signal processing techniques, coupled with shallow Machine Learning classifiers or wave change detection methods, more recently, new approaches, based on Deep Learning, have been proposed. These methods have the potential to achieve high classification accuracy and to remove the need of extensive data pre-processing, hence being more suitable for real-time deployments. However, high classification performance has been only demonstrated using synthetically generated data. In order to address limitations related to processing time and accuracy, this paper proposes a novel end-to-end framework for automated detection of PQDs based on Deep Transfer Learning. The proposed approach uses a small set of images of voltage waveforms to train the model and classify different types of PQDs. This method leverages on the high performance of existing pre-trained models for image classification and shows consistent high accuracy for data with varying resolution. The proposed methodology provides a pathway towards effective deployment of Deep Learning in power quality monitoring systems and real-time applications.

Topics & Concepts