Real-Time Power System Event Detection: A Novel Instance Selection Approach

Abstract

This study presents a novel adaptation of the Hoeffding Adaptive Tree (HAT) classifier with an instance selection algorithm that detects and identifies cyber and non-cyber contingencies in real time to enhance the situational awareness of cyber-physical power systems (CPPS). Wide-area monitoring, protection, and control (WAMPAC) systems allow system operators to operate CPPS more efficiently and reliably. WAMPAC systems use intelligent devices such as phasor measurement units (PMUs) to monitor the CPPS state. However, such devices produce continuous and unbounded data streams, posing challenges for data handling and storage. Moreover, WAMPAC devices and the communication links connecting them are vulnerable to cybersecurity risks. In this study, we consider several cyber and non-cyber contingencies affecting the physics and monitoring infrastructure of CPPS. Our proposed classifier distinguishes disturbances from cyberattacks using a novel instance selection algorithm with three algorithmic stages to ease data management. A cost and complexity analysis of the algorithm is discussed. With reduced computational effort, the classifier can handle high-velocity, high-volume, and evolving data streams from the PMUs. Six case studies with extensive simulation results corroborate the merits of the proposed classifier, which outperforms state-of-the-art classifiers. Moreover, the classifier demonstrated a high performance using a dataset outside the contingency detection domain. Finally, the real-time applicability of the proposed methodology is assessed, and its limitations are discussed.

Keywords

• Classification
• cyber-attacks
• disturbances
• instance selection
• streaming data