報告主題：Physics-Informed Deep Reinforcement Learning for Power System Optimization and Control
報 告 人：Prof. Junbo Zhao
Junbo Zhao is an Associate Director of Eversource Energy Center for Grid Modernization and Strategic Partnerships and assistant professor of the Department of Electrical and Computer Engineering at the University of Connecticut. He received the Ph.D. degree from Bradley Department of Electrical and Computer Engineering Virginia Tech, in 2018.He is currently the Chair of the IEEE Task Force on Power System Dynamic State and Parameter Estimation, the IEEE Task Force on Cyber-Physical Interdependency for Power System Operation and Control, and the Co-chair of the IEEE Working Group on Power System Static and Dynamic State Estimation.
He has published three book chapters and more than 140 peer-reviewed journal and conference papers. His research interests are cyber-physical power system modeling，estimation，security, dynamics and stability, uncertainty quantification, renewable energy integration and control, robust statistical signal processing and machine learning. He serves as the associate editor of more than 6 international journals, including the IEEE Transactions on Power Systems, IEEE Transactions on Smart Grid and IEEE Power and Engineering Letters. He received the 2020 IEEE PES Outstanding Engineer Award, the 2021 IEEE PES Outstanding Volunteer Award, and the 2021 IEEE Transactions on Power Systems Best Paper Award. He has been listed as the 2020 and 2021 World's Top 2%Scientists released by Stanford University in both Single-Year and Career tracks.
For the past decades，the integration of intermittent renewable sources，responsive loads and other new technologies significantly complicates today's power systems. It becomes more and more challenging to build accurate models for power grid planning，operation and control. This leads to the fruitful development of machine learning-based power system applications. However，the existing machine learning methods are data-driven， and the underlying physical models are typically not considered. As a result, a large number of high-quality training data and complex neural network structures are required. There are also serious concerns of the physical interpretability of the machine learning results. Motivated by the advancements of constrained machine learning methods that consider some critical physical constraints, we develop a physics informed deep reinforcement learning framework for power system optimization and control, such as optimal power flow, preventive stability control, and Volt-VAR optimization.