The IEEE Greece Young Professionals in collaboration with IEEE Greece PES Chapter, IEEE NTUA SB, PES Chapter of IEEE NTUA SB and Smart grids research unit of ECE (SmartRUE) invite you to a lecture titled ‘’Decision Trees for the global optimum of the AC OPF and inferring guaranteed costs with renewable energy” by Dr. Panayiotis Moutis (Assisant Prof., CCNY of CUNY) on 10th of January 2024 at 3pm at the multimedia amphitheater of NTUA Library & Information Centre. More information can be found in the abstract and the bio!
Abstract
The AC Optimal Power Flow (OPF) seeks to minimize the cost of using certain resources or the levels of some performance metrics in electrical grid operation. Due to the non-linear grid physics, it is a non-convex problem that cannot be solved efficiently in most of its practical instances. Relaxations cannot guarantee an AC OPF globally optimum solution for all grids or cases, while they also require lifting the problem to higher dimensions, requiring increased processing times. Among the non-linear, artificial intelligence and machine learning solvers, invariable challenges arise ranging from failure to converge to the global optimum to difficulties in modeling the problem in each solver. This talk proposes an AC OPF feasible space search. The space is sampled with random dispatches using power flows. The AC OPF infeasible samples are marked as False. The feasible AC OPF samples with cost greater than the median of all feasible samples are also marked False; the remaining are marked True. A heuristically inducted binary decision tree is trained with these samples and will tighten the constraints of the AC OPF variables towards the space of the feasible samples of cost less than the mentioned median. The recursive sampling and constraint tightening leads to convergence to the global optimum of the AC OPF instance based on Bayesian guarantees. Several benchmark instances from the PGLib and NESTA libraries are solved favorably with the proposed method compared to IPOPT and other relaxations and solvers. The time requirements arising from sampling the AC OPF space are discussed in terms of “hot starting” the method with statistically advantageous linear approximations. Lastly, because renewables are, most typically, zero cost assets and the method converges to the global optimum by the successive improvement of the median cost of AC OPF feasible samples, it is shown that the method can yield optimal dispatches for any level of renewable energy available.
Bio
Panayiotis (Panos) Moutis, PhD, is Assistant Professor at the Dept. of Electrical Engineering at the City College (CCNY) of the University of New York (CUNY). He has previously been special faculty (2018-23) and postdoctoral research associate (2016-18) at Carnegie Mellon University. Panos studies pragmatic-data-driven optimization, control and planning of electrical grids with high shares of renewables. He has recently been working with the grid operator of Portugal, REN, the moonshot factory of Google, X, and the grid operator of NY, NYISO. In 2018-20 he served as a Marie Curie Research Fellow with DEPsys, Switzerland, on distribution grid awareness. In 2014 he was awarded a fellowship by Arup, UK (through the University of Greenwich), to study microgrids for residential communities. During 2007-15, as part of Prof. Nikos Hatziargyriou’s research group he contributed to over a dozen R&D projects funded by the European Commission. Panos received both his diploma (2007) and his PhD (2015) degrees in Electrical and Computer Engineering at the National Technical University of Athens, Greece, and has published more than 30 papers and contributed to 4 book chapters. He has over 10 years of industry experience on Renewable Energy and Energy Efficiency, and serves as advisor and executive in energy start-ups. He is Chair of the IEEE-USA Energy Policy Committee, senior editor of IEEE & IET scientific journals, member of IEEE standard working groups, a senior member of the IEEE, and leads the Distribution Task Team at the North American Synchro-Phasor Initiative and the Power & Energy Community at the Climate Change AI initiative. Personal Website for more information: https://panay1ot1s.com/
