The operation of future electric grids will likely rely on solving large-scale, dynamic optimization and control problems involving hundreds of thousands of devices jointly optimizing millions of variables. These systems will include various types of active dynamic devices, such as distributed generators based on solar and wind, batteries, deferrable loads, curtailable loads, and electric vehicles, whose control and scheduling amount to a very complex nonlinear power management problem.
The main objective of this talk is to highlight the challenges in the operation of future energy systems and address some of them. I propose high-performance computational methods for the control and operation planning of power grids, by deploying various concepts in optimization theory, graph theory, power systems, control theory, distributed computation, low-rank matrix completion, and numerical algorithm. I demonstrate our proposed methods in two examples. The first one is about solving a power optimization problem for a real-world power grid with over 16,000,000 parameters, for which we obtain a feasible solution with a global optimality degree of at least 99.9%. The second one is to perform real-time optimal distributed control for the New England Power Grid to maximize the penetration of renewable energy. I conclude the talk by discussing the implementation of our algorithms in three toolboxes (named OPF solver, ODC solver, and ADMM-SDP solver), which are able to efficiently solve problems with millions of variables.