A Deep Learning Approach for Discrimination of Single- and Multi-Source Corona Discharges
Moein Borghei, Mona Ghassemi
Abstract
Insulation system health is crucial for reliable, lifelong operation of almost any electrical apparatus. While many studies have focused on the testing, modeling, and analysis of insulation aging mechanisms, research is needed to overcome new challenges in electric power systems. Fortunately, the progress in data analytics methods has opened up new opportunities to extract information from datasets. This study aims to make use of deep learning algorithms to lay the foundation for an online condition monitoring system that is capable of discriminating single- and multi-source corona discharges. In this article, we report the results of experimental testing and conversion of the data into phase-resolved partial discharge images, which we fed into deep neural networks. We begin by reviewing some of the most successful image recognition models including AlexNet, Inception-V3, residual network (ResNet), and DenseNet. Thereafter, we develop and optimize a ResNet model to achieve the highest accuracy model with the lowest computational cost.