Ensemble learning based fault detection using PMU data in imbalanced data condition

Abstract

Significant advancements in the electrical grid include enhanced regulation, communication, metering, and customer interaction, driven by information communication technologies (ICTs) and cyber-physical systems (CPS). The adaptation of synchro phasor devices like phasor measurement units (PMUs) enables real-time monitoring and control, aiding in power system security assessment. PMUs record voltage and current phasors with GPS time stamps, transmitting data to phasor data concentrators (PDCs) for decision-making. However, ensuring the stability and security of this method against cybersecurity threats is crucial due to its reliance on Internet Protocol (IP) networks. Dynamic security assessment utilizes PMU data, reported up to 30–60 times per second, to evaluate power system safety. To address security issues, a Python-based fault detection system employing a stack ensemble learning algorithm is developed. This approach consistently outperforms traditional methods, producing satisfactory results with superior AUC-ROC curves, validated through correctness checks and graphical analysis. The dataset includes both natural and man-made security threats, facilitating comprehensive assessment and mitigation strategies. The ensemble learning algorithm performed better than the individual algorithms by obtaining 95% in the AUC-ROC curve.