Machine Learning Aided Inertia Constrained Day Ahead Energy Scheduling

A C2SR Colloquia Series | Distinguished Webinar Series.
The Distinguished Speaker Webinar Series is aimed at advancing the state-of-the-art concepts and methods in artificial intelligence and cyber security areas. The series is jointly hosted by the Centers for Cyber Security and AI Research and the School of Electrical Engineering and Computer Science (SEECS) at the University of North Dakota College of Engineering & Mines.
Speaker Biography:
Dr Xingpeng Li is currently an Assistant Professor in the Department of Electrical and Computer Engineering at the University of Houston (UH). He previously worked for ISO New England and PJM Interconnection. Prior to joining UH in 2018, he was a senior engineer with ABB’s Power Grid division, which is now Hitachi Energy. He received his Ph D in electrical engineering from Arizona State University in 2017. He received the NAS Gulf Research Program’s Early Career Research Fellowship in 2023 and the NSF CAREER award in 2024. His research interests include power system operation, control and planning, microgrid optimal sizing and energy management, and applied machine learning and optimization.
About the Webinar:
To ensure power system operate in a reliable and efficient manner, day-ahead scheduling is performed daily by solving a security-constrained unit commitment ( problem Maintaining system frequency within acceptable limits is critical for power system stability However, with increasing penetration of variable renewable energy in the power system, the number of conventional synchronous generators committed on will be much less than before This results in a significant reduction in system synchronous inertia, which would negatively affect system frequency stability This makes it necessary to ensure inertia requirements in SCUC for future renewables dominated low inertia power grids To address this issue, this talk will describe a general procedure for enforcing grid dynamic performance in look ahead optimal energy scheduling models Particularly, this talk will demonstrate this procedure by explaining how to create a machine learning ( based frequency stability predictor that is then integrated into SCUC, creating a novel inertia constrained unit commitment model This talk will also present multiple computational enhancement approaches such as sparse neural networks to ensure quality solutions can be obtained in a timely manner.