AMPS: Graph-based Quickest Prediction and Assessment of Dynamic Anomalies in Power Systems

This research program significantly improves the resiliency of energy scheduling and management in power grids, which will have significant economic and environmental benefits. One of the major challenges still facing the modern power systems are their vulnerability to large-scale disruptions, failures, or even outages. Large-scale disruptions in power systems follow a sequence of less severe anomalous events that gradually stress the system over time. System operators are often unaware of these gradual changes since such changes often either appear insignificant or are not even detectable when they fall within a normal range of noisy variations. As a result, the system operators remain oblivious to the emerging anomalies that lead to a progressively less secure system state and a sequence of disruptions that gradually become more severe. The objective of this project is to develop theoretically-principled and efficient algorithms that can predict and rectify large-scale disruptions based the gradual anomalous changes . This research program addresses the following research questions. Question 1: How can we quickly and reliably determine whether there are ambient-level anomalies in the system and distinguish them from random noise? Question 2: If the system is deemed to contain anomalies, how can we quickly localize them? Question 3: When the existing anomalies are localized, how can we determine their subsequent risks to the system? Do the localized anomalies have locally-contained risks, or can they lead to anomalies of larger scales and risks? Our approach to developing provably correct algorithms that recognize all the complex constraints of power system dynamics involves advancing the existing theories in change-point detection, active sensing, probabilistic graphical models, and graphical neural networks. 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.