Speaker: Pengyuan Wang, ECpE Graduate Student
Advisor: Manimaran Govindarasu
Title: Decentralized Power System Integrity Protection Schemes based on MAS: Anomaly Detection and Self-Adaptive Load Shedding
Abstract: With the wide application of the communication and information technologies in modern power system, malicious cyber attacks have become great concerns, especially for the wide area protection and control schemes. Traditional cyber security technologies can be applied to alleviate the situation, but the problem will not get solved by merely securing the periphery of schemes. To make the critical functions per se situation-aware and attack-resilient is a promising path. This paper presents a peer-to-peer multi-agent architecture for power system integrity
protection and control schemes (SIPS), which provides an alternative implementation for the non-resilient legacy SIPS and the centralized ones. Every agent within the architecture is embedded into one substation/power plant, and it collects local data as well as information from its community. All the agents possess the similar level of intelligence and interact in a peer-to-peer manner. To verify the validity of the proposed SIPS architecture, coordinated anomaly detection and situation adaptive optimal load shedding are further investigated as two sub-problems. Anomaly detection takes place within each intelligent agent based on the local data but the final detection conclusion will be achieved by the consensus of interconnected agents. A data-driven algorithm named as Support Vector Machine embedded Layered Decision Tree (SVMLDT) is proposed for the anomaly detection. Global data propagation lays the foundation for the situation adaptive load shedding such that a protocol is proposed to facilitate the data propagation. Dynamic programming is utilized to get the remedial actions for each agent. A load rejection SIPS adopted in reality is mapped onto IEEE 39 bus system as a study case. Via evaluation, it shows that the proposed decentralized SIPS attains better attack resiliency than the centralized solution under Denial of Service (DoS) attack and when part of the decentralized SIPS gets compromised, the rest of it can still deliver the remedial actions desired.