AI-Assisted Algorithms for Automatic AC Power Flow Model Creation based on DC Dispatch

University Of Tennessee Knoxville, Knoxville TN

Investigators

Abstract

This NSF project aims to develop user-friendly algorithms that automatically create realistic snapshots of real U.S. electric grids for many futuristic operating scenarios with variable proportions of conventional and renewable energy generators. This project will bring transformative change to the process of risk identification and mitigation of electric grids by significantly reducing model development time and increasing the operating conditions that can be considered. The intellectual merits of the project include: 1) deployment of a multi-stage approach that combines physics-based principles and artificial intelligence-based methods to convert dispatch scenarios of grid generators to full power system cases, 2) development of a data-driven problem discovery algorithm to assist human intervention. The broader impacts of the project include: a) facilitation of high-renewable energy power grids towards the achievement of national clean energy goals, b) involvement of undergraduate and graduate students from underrepresented groups, c) provision of opportunities, including lab tours and presentations, for K-12 students to learn about the potential use of artificial intelligence in real power grids. Higher penetration of renewable energy resources has led to increased variations in daily generation mix, thus, the AC power flow (ACPF) solution of a DC power flow (DCPF) dispatch case is no longer a good initialization to obtain the ACPF solution of the next operating condition. Currently, there are neither reliable algorithms nor vendor products to automatically convert DCPF dispatch cases to converged ACPF cases, thus, conversion requires manual analysis and tuning. This project aims to solve this problem using several innovations: 1) a physics-guided machine learning initializer (PMLI) to replace flat start initialization in Newton Raphson solution method for ACPF, 2) a transfer learning process to reuse developed PMLI on multiple power systems, and 3) a hot-start incremental algorithm with automatic reactive power compensation selection. This work will provide an important step to make simulation of the power grid functions in real time a reality. This award reflects NSF's statutory mission and has been deemed worthy of support through evaluation using the Foundation's intellectual merit and broader impacts review criteria.