NSF Award
2500682
Rapid decarbonization and deployment of flexible, distributed resources in the electricity energy sector are quickly transforming the real-time operation paradigms of the interconnected power grid infrastructure. These changes have led to growing concerns over power system dynamics and stability, due to the reduced capability of grid inertia and increasing levels of external disturbances and variability. Meanwhile, the electricity infrastructure has benefited significantly from the ongoing deployment of sensing and cyber resources, which give rise to a huge amount of high-rate, high-quality data and information collected during real-time operations. Thanks to the enriched data availability, machine learning advances are envisioned to play an increasingly important role to address the challenges in power system dynamics and stability. This project aims to bridge domain-specific machine learning tools to transform the current grid dynamic modeling, inference, and stability-enforcing solutions. At a societal level, the anticipated outcomes can improve energy efficiency and security, and facilitate higher and smoother penetration of renewables and carbon-free resources. This project will further benefit industry practices with advanced algorithmic solutions, as well as education efforts by providing student training opportunities and reaching out to pre-college students via interactive demos. <br/><br/>This project will develop data-enabled and physics-informed modeling, monitoring, and optimization algorithmic solutions targeting power system dynamics. The proposed activities put forth and explore three creative, original, and potentially transformative ideas: i) Correlating synchrophasor data collected at two arbitrary grid locations can efficiently unveil the impulse response of the associated linear time-invariant (LTI) system under certain assumptions, which can be waived leveraging physics-informed analysis; ii) Gaussian processes (GPs) constitute a powerful tool for inferring signals occurring in LTI systems, and thanks to the underlying physics, GPs can be uniquely adapted to learn grid dynamic signals and their derivatives from heterogeneous, noisy, spatially and temporally incomplete, and/or multirate synchrophasor readings; iii) Well-established grid stability metrics can be expressed as convex functions of the steady-state operating point, and stability-aware OPFs can be handled via a semidefinite program relaxation. The outcome will be a comprehensive suite of computational tools dealing with grid dynamics from learning to power system operations, evaluated by both real-event synchrophasor datasets, and synthetic datasets generated from realistic power systems such as a Texas 2000-bus case in collaboration with ERCOT. The research results will also be integrated into engineering educational activities at the secondary and higher education levels. In addition to standard dissemination venues, close collaboration with grid operators will assist in showcasing the effectiveness of the project findings on real-world systems and lead to quick adoption.<br/><br/>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.
