An anomaly-based technique for fault detection in power system networks

Abstract

In recent years, fault detection in electrical power systems has attracted substantial attention from both research communities and industry. Although many fault detection methods and their modifications have been developed during the past decade, it remained very challenging in real applications. Moreover, one of the most important parts of designing a fault detection system is reliable data for training and testing which is rare. Accordingly, this paper proposes an anomaly-based technique for fault detection in electrical power systems. Furthermore, a One-Class Support Vector Machine (SVM) model and a Principal Component Analysis (PCA)- based model are utilized to accomplish the desired task. The used models are trained and tested on VSB (Technical University of Ostrava) Power Line Fault Detection dataset which is a large amount of real-time waveform data recorded by their meter on Kaggle. Finally, performance and Receiver Operating Characteristic (ROC) curves analyses of our results are exploited to verify the effectiveness of the proposed technique in the fault detection problem.