Date of Award

Spring 5-23-2021

Degree Type

Dissertation

Degree Name

Doctor of Philosophy (PhD)

Department

Electrical Engineering and Computer Science

Advisor(s)

Eftekharnejad, Sara

Keywords

controlled islanding, cyber security, false data attack, machine learning, optimization, remedial action scheme

Subject Categories

Electrical and Computer Engineering | Engineering

Abstract

The global energy market has seen a massive increase in investment and capital flow in the last few decades. This has completely transformed the way power grids operate - legacy systems are now being replaced by advanced smart grid infrastructures that attest to better connectivity and increased reliability. One popular example is the extensive deployment of phasor measurement units, which is referred to PMUs, that constantly provide time-synchronized phasor measurements at a high resolution compared to conventional meters. This enables system operators to monitor in real-time the vast electrical network spanning thousands of miles. However, a targeted cyber attack on PMUs can prompt operators to take wrong actions that can eventually jeopardize the power system reliability. Such threats originating from the cyber-space continue to increase as power grids become more dependent on PMU communication networks. Additionally, these threats are becoming increasingly efficient in remaining undetected for longer periods while gaining deep access into the power networks. An attack on the energy sector immediately impacts national defense, emergency services, and all aspects of human life. Cyber attacks against the electric grid may soon become a tactic of high-intensity warfare between nations in near future and lead to social disorder. Within this context, this dissertation investigates the cyber security of PMUs that affects critical decision-making for a reliable operation of the power grid. In particular, this dissertation focuses on false data attacks, a key vulnerability in the PMU architecture, that inject, alter, block, or delete data in devices or in communication network channels.

This dissertation addresses three important cyber security aspects - (1) impact assessment, (2) detection, and (3) mitigation of false data attacks. A comprehensive background of false data attack models targeting various steady-state control blocks is first presented. By investigating inter-dependencies between the cyber and the physical layers, this dissertation then identifies possible points of ingress and categorizes risk at different levels of threats. In particular, the likelihood of cyber attacks against the steady-state power system control block causing the worst-case impacts such as cascading failures is investigated. The case study results indicate that false data attacks do not often lead to widespread blackouts, but do result in subsequent line overloads and load shedding. The impacts are magnified when attacks are coordinated with physical failures of generators, transformers, or heavily loaded lines. Further, this dissertation develops a data-driven false data attack detection method that is independent of existing in-built security mechanisms in the state estimator. It is observed that a convolutional neural network classifier can quickly detect and isolate false measurements compared to other deep learning and traditional classifiers. Finally, this dissertation develops a recovery plan that minimizes the consequence of threats when sophisticated attacks remain undetected and have already caused multiple failures. Two new controlled islanding methods are developed that minimize the impact of attacks under the lack of, or partial information on the threats. The results indicate that the system operators can successfully contain the negative impacts of cyber attacks while creating stable and observable islands. Overall, this dissertation presents a comprehensive plan for fast and effective detection and mitigation of false data attacks, improving cyber security preparedness, and enabling continuity of operations.

Access

Open Access

Share

COinS