PNNL

Abstract

This report presents multiple new results related to the development of multi-scale analytics and control designs for grid oscillations and stability in presence of large-scale power electronics based resource integration. This work investigates two main oscillation types. In the initial part of the report, we consider wide-area inter-area oscillations under high grid-forming inverter (GFM) penetration. Our findings show that significant integration of GFMs replacing conventional generation alter the conventional coherency structure of the grid, necessitating modifications to wide-area control. We propose a distributed, risk-constrained reinforcement learning–based wide-area control design that accounts for this perturbed coherency. Additionally, we study how GFM droop, inverter capacity, and active power headroom affect natural low-frequency modes in heavily GFM-integrated grids, highlighting the importance of droop tuning. A zeroth-order gradient-based algorithm is proposed to optimize such settings and ensure adequate stability margins. Subsequently, we focus on sub- and super-synchronous control interactions (SSCIs). We extend the Dissipative Energy Flow (DEF) method to detect oscillation paths at faster frequencies, validated through PSCAD EMT cases with superimposed sub and super synchronous oscillations. Using an extended MiniWECC EMT case, we show oscillation propagation through MTDC links between IBR and grid-side resources. To this end, we develop an EMT-in-the-loop learning framework for adaptive SSCI mitigation. Finally, we propose a physics-informed neural learning framework to emulate proprietary inverter dynamics, addressing challenges posed with proprietary inverter models. Our Physics-Informed Latent Neural ODE Model (PI-LNM) integrates physics with learning layers to capture black-box inverter dynamics, improving accuracy in grid simulations and control studies.