PNNL

Abstract

Virtual synchronous generator (VSG) is typically deployed for grid-forming inverters interfacing various generation resources to the grid. With the growing penetration of distributed generation and the rising demand from data centers, grid conditions can change significantly. These variations challenge the conventional P–Q decoupling assumption in fixed-gain VSGs, resulting in reduced robustness, oscillations, long settling times, and overshoot under stiff-grid conditions. This paper analyzes the effects of coupling and grid strength on the transient P–Q performance of conventional VSGs over a range of short-circuit ratio (SCR) and reactance-to-resistance (X/R) ratio values. To address these challenges, an adaptive control framework is proposed that integrates full-state feedback with a Physics-Informed Neural Network (PINN) based online grid impedance estimator. A standard state-space model of the VSG is developed to explicitly represent coupling effects and parameter dependencies, supporting systematic controller design. Then a full-state-feedback control is designed via pole placement and enhanced by the PINN estimator, enabling real-time tuning of VSG gains to achieve specified settling times and damping ratios under varying grid conditions. Offline and real-time simulations in MATLAB/Simulink and on an OPAL-RT platform demonstrate well-damped power responses with minimal overshoot and consistent dynamic performance across wide SCR and X/R ranges.