IEOR professors Javad Lavaei and Alper Atamturk have been awarded $360,000 in funding by the National Science Foundation to investigate computational methods for mixed-integer programs in power systems. The research will take place from 2018-2021.
Title: Computational Methods for Mixed-Integer Programs in Power Systems
Abstract: The goal of this project is to design provably efficient computational methods for the optimal operation of power systems and to facilitate their transformation into sustainable systems. Since power systems are large-scale interconnected networks with tens of thousands of devices connected to one another via a physical infrastructure, power operators periodically solve a series of highly complex optimization problems to be able to run these systems. One major power optimization problem is unit commitment (UC), which optimizes the production schedules of the participating generators and is the backbone of the US electricity market with the value of exceeding $300B annually. In addition, the emerging problem of optimal transmission switching (OTS) enables a further improvement of the operation of power systems by co-optimizing the interactions among the resources in the infrastructure. Since these problems are highly nonlinear, well-established optimization algorithms cannot efficiently solve them consistently and suffer from major drawbacks. This project aims to address the pressing need to develop effective techniques that are able to solve much larger power optimization problems on a much shorter time scale with a higher accuracy, compared to the current capabilities. This project leverages the underlying structures of real-world systems to develop customized computational techniques for power optimization problems with strong theoretical and practical guarantees. The focus is on the UC problem (with binary variables at the nodes of the system) and the OTS problem (with binary variables on the links of the system), since other mixed-integer power problems mathematically resemble a combination of UC and OTS. The proposed approach relies on advanced topics in graph theory, conic optimization, valid inequalities, rounding techniques, penalization methods, branch-and-bound techniques, robust optimization, and algebraic geometry.