Title: Learning Congestion Patterns in Optimal Power Flow Problems
Date: February 12, 2020
Bio: Line A. Roald is an Assistant Professor and Grainger Institute Fellow in the Department of
Electrical and Computer Engineering at the University of Wisconsin-Madison. Prior to joining UW
Madison, she received her Ph.D. in Electrical Engineering and her M.Sc. and B.Sc. in Mechanical
Engineering at ETH Zurich in Switzerland and worked as a postdoctoral researcher at Los Alamos
National Laboratory. Her research interests are in energy systems and optimization, with a focus on
integration of renewable energy and the transition to a more sustainable and resilient electric grid.
Abstract: In power system operations, optimization problems such as the optimal power flow (OPF)
are solved over and over and over again. In this tutorial, we discuss how to use statistical learning and
bound tightening techniques to make this repeated solution process more efficient. A key observation is
that traditionally, the solutions to the OPF problem have been characterized by a small number of
binding constraints (and so-called active sets), giving rise to a limited number of operational patterns.
The first part of the tutorial discusses how the small number of operational patterns makes it possible to
learn the optimal power flow solution from previous solutions, both for statistical learning algorithms
and for human operators. In the second part of the tutorial, we discuss how the introduction of renewable
energy may change this picture of stable, recognizable operating conditions.