The U.S. power and energy system is being reshaped into an active multi-physics grid, where data centers form massive load clusters while inverter-based renewables and storage dominate new generation. This new regime departs from original assumptions and strains existing monitoring and control. Data-driven solutions are increasingly applied, and physics-informed mechanisms are growing, but exact physics priors are difficult to collect in practice. We present machine learning methods that remain physically consistent while being flexible to data. We begin with expanding power system
models with limited system information and sparse observability, where conventional modeling and monitoring approaches become inadequate, motivating implicit physics regularizations. We first discuss invertible neural networks (INNs) that enforce forward–inverse consistency in power flow modeling and state estimation. Building on these monitoring insights, we demonstrate how feasible operational patterns can be extracted while prioritizing high-information transitions through offline reinforcement learning (RL) for reliable decision-making. The talk will conclude with future directions on extending implicit physics regularizations toward more adaptive, multi-scale digital twins that integrate monitoring with real-time control.

Bio:

Jingyi Yuan graduated with a Ph.D. in Electrical Engineering from Arizona State University. Her research focuses on physical-consistent machine learning methodologies for power system monitoring and control, including power flow
modeling, renewable hosting capacity analysis, state estimation, and voltage control. Her interdisciplinary publications span leading power systems journals (IEEE Transactions on Power Systems and IEEE Transactions on Smart Grid) and
machine learning venues (ACM KDD, ICDM, and ECML-PKDD). She has received the ASU University Graduate Fellowship (2019), the Best Paper Award at NAPS (2019), and was a finalist in the IEEE PES Grid Edge Dissertation Prize Challenge (2023).