Data-Driven Constrained Control for Power Systems

Abstract

Power systems are evolving rapidly along several fronts. Renewable energy installations at the distribution and transmission levels are introducing new levels of variability and volatility, which are only exacerbated by increasing occurrences of extreme whether. Meanwhile, inverter-based controls and battery energy storage systems are providing new levels of flexibility while phasor measurement units and smart meters are improving the observability of power systems. Taken together, these changes both necessitate and enable new strategies for power system operation and control. These new strategies must utilize the fast responsiveness of inverter-based devices to compensate for increased volatility of power systems. They must do so in a way that respects the engineering constraints of power systems while handling uncertainty effectively. This dissertation addresses the design of policy functions for frequency and voltage regulation in modern power systems considering the joint challenges of computational complexity, uncertainty, and safety. First, we consider the problem of safe exploration for frequency regulation from the perspective of centralized control. We then move to a decentralized setting motivated by building energy management and develop new algorithms that yield probabilistic safety guarantees at execution time. Theoretical results are backed up by simulations demonstrating the advantages of the proposed~methods.