Graduation Year

2025

Document Type

Dissertation

Degree

Ph.D.

Degree Name

Doctor of Philosophy (Ph.D.)

Degree Granting Department

Electrical Engineering

Major Professor

Mia Naeini, Ph.D.

Committee Member

Nasir Ghani, Ph.D.

Committee Member

Ismail Uysal, Ph.D.

Committee Member

Kaiqi Xiong, Ph.D.

Committee Member

Ning Wang, Ph.D.

Keywords

AI in Cyber-security, Artificial Intelligence, Cyber-physical Systems, Graph Neural Network, Graph Theory, Power Systems

Abstract

Enhancing the reliability and security of smart grids is critical for ensuring their seamless operation and resilience against disruptions. The increasing integration of distributed energy resources, advanced measurement devices, and cyber-physical elements introduces both opportunities and challenges for grid management. While these advancements provide enhanced visibility and operational control, they also expose the grid to vulnerabilities from cyber-physical stresses, such as cyber-attacks, equipment failures, and fluctuating power demands. Traditional methods for reliability assessment and threat detection often rely on model-based approaches that struggle to adapt to the complexity and dynamic nature of modern smart grids. These limitations necessitate novel data-driven solutions to improve situational awareness, predictive maintenance, and system reliability.

This dissertation uses a Graph Neural Network (GNN) framework tailored to enhancing the reliability and security of smart grids. GNNs, with their ability to model complex topological relationships and non-linear interactions among system components, offer a powerful tool for analyzing the unique spatial and temporal characteristics of smart grids. The proposed framework integrates GNNs with advanced data analytics to address key challenges such as state estimation, fault localization, and anomaly detection. Specifically, spatio-temporal GNN architectures are adopted to capture both the spatial features of grid and the dynamic variations in system states. These models are evaluated under diverse operating conditions, including normal operations, cyber-physical attack scenarios, and topological disturbances.

Moreover, this dissertation investigates the potential of a resilient GNN-based framework to topological inaccuracies caused by known or unknown cyber-physical stresses, noise, and attacks. By leveraging the physical and operational characteristics of the grid, the framework ensures consistent and enhanced state estimation accuracy over time. This resilience is attributed to its novel architecture, which effectively mitigates the impact of topological inaccuracies, thereby maintaining a secure and reliable network. Experimental evaluations using simulated data highlight the framework's ability in preserving and improving the reliability and security of smart grids compared to conventional approaches.

Through the development of these novel GNN-based approaches, this dissertation contributes to advancing the field of situational awareness in power systems.

Download

Share

COinS