With increasing renewable integration and phase-out of conventional generators, the power system dynamics faces reduced inertia and becomes more dynamic and uncertain because of frequent power variations and uncertainties of renewables and system contingencies. These changes lead to an ever-increasing number of control nodes across the wide-area transmission systems. They also present operational challenges on how effective real-time control can be realized using under-utilized yet fast-responding power electronics-based resources and advanced sensing technologies under normal conditions and extreme events. The grid control paradigm and technologies currently adopted in the power industry are incapable of handling a significantly increased number of control points and utilizing the fast control capability of inverter-based resources (IBRs) and high-resolution measurement data. Thus, new architecture and advanced control technologies need to be developed to enable system operators to provide faster, robust, and uncertainty-tolerant real-time control on a vast number of control points so that reliable and fast grid services can be delivered to ensure efficiency under normal operations and resilience under extreme events.
In this project, we aim to propose a new stochastic control method that shapes the distribution of grid frequency towards a desired Gaussian distribution. Our key contribution is a novel data-driven, distribution-shaping model predictive control method for solving an infinite dimensional, continuous action space Markovian decision process, such as the distribution of the power grid system. We tested this algorithm on a two-area grid system with varying wind and solar power generation. The results show that our method can effectively shape the probability distribution function of the power grid frequency in five minutes to be more Gaussian.
This work is supported by DE-AC05-00OR22725 with the US Department of Energy (DOE), Office of Electricity and Office of Energy Efficiency & Renewable Energy, subcontract through Oak Ridge National Lab (ORNL).