Daniel Bienstock, PhD
Liu Family Professor of Operations Research & Applied Mathematics
Columbia University
Friday, March 1, 2024
2:15-3:05 pm Tickle 410
Abstract: The ACOPF (or, Alternating Current Optimal Power Flow) problem concerns the operation of power grids so as to deliver energy from generators to loads at minimum cost, while obeying nonlinear and nonconvex laws of physics. Large-scale problem cases, in single-time period mode, can be handled by nonlinear solvers such as Ipopt or Knitro so as to obtain very good solutions; multi-time period formulations are beyond the scope of such solvers. Additionally, the solvers do not provide any guarantee as to actual solution quality. Moreover, the solvers do not provide pricing –i.e., incremental cost– information, which is one of the desired pieces of information. Nonlinear but convex relaxations which are also tight are known, but do prove quite challenging for the best solvers.
In this talk we describe linearly constrained informations that quickly and robustly handle even the largest cases for ACOPF. Most important, the formulations are warm-startable, an important feature given that in actual operation of power systems, an optimization is never run from scratch and, indeed, must be evolved out of a pre-existing solution for recent (and slightly different) data. This is joint work with my student Matías Villagra.
Bio: Daniel Bienstock is the Liu Family Professor of Operations Research and Applied Mathematics, with courtesy appointment in Electrical Engineering, at Columbia University. His research interests concern methodological, computational and experimental aspects of optimization, with parallel interest in power systems operations. He received the PhD in Operations Research from MIT. He has been an Informs Fellow since 2013 and received the 2022 Khachiyan Prize in Optimization.