CAREER: Learning Power System Graph Signals for Cascade Resiliency

University Of South Florida, Tampa FL

Investigators

Abstract

This NSF CAREER project aims to improve the resiliency of smart transmission grids to cascading failures. Cascading failures in power grids are successive interdependent failures of components, which can lead to large blackouts with significant societal and economical impacts. The project will bring transformative change in supporting operation and mitigation functions during and before cascading failures through various graph-empowered predictive, descriptive and prescriptive analyses. These will be achieved by bridging the gap between graph-based and data-driven modeling and analyses of cascading failures in power grids. The intellectual merits of the project include developing a graph signal learning framework for analyzing cascading failures in power systems using graph signal processing (GSP) and graph-empowered machine learning techniques. The broader impacts of the project include an integrated education and workforce training component to foster interdisciplinary training in the area of energy data analytics through new course developments and mentoring and advising efforts with the help of industry and academic partners, as well as efforts to increase participation of underrepresented students in STEM disciplines. This project will develop new methodologies to enhance the reliability of power transmission grids to cascading failures by integrating the system’s structural and components interaction data along with temporal data, capturing dynamics of the states, in the form of graph signals. The proposed research will make new discoveries in the dynamics and properties of power systems’ graph signals during cascading failures through vertex domain, graph-frequency domain, and the joint vertex-frequency domain analyses and through modeling power system dynamics in a GSP framework using tools including graph filters. The analyses through the proposed framework will enable developing new techniques for detecting the proximity to cascade transition and identifying areas for protective control. Moreover, graph-empowered machine learning techniques, including graph neural networks, will be developed to learn the signatures and patterns of cascade stresses in various graph signal domains and to develop tools to support optimizing corrective and preventive decisions for improving cascade resiliency. This award reflects NSF's statutory mission and has been deemed worthy of support through evaluation using the Foundation's intellectual merit and broader impacts review criteria.