Spectral admittance modeling and learning-based stability assessment for 100% inverter-dominated power systems

This paper presents a unified framework for evaluating and predicting the small-signal stability of 100% inverter-dominated power systems. The method begins with a spectral admittance formulation, where the dynamics of grid-forming (GFM) and grid-following (GFL) inverters are represented through frequency-domain admittances combined with the network model via Kron reduction. From this reduced model, a spectral grid strength index (SGSI) is derived to measure how strongly the system resists oscillations at each operating point. To enable real-time assessment, a data-driven model based on the frequency-enhanced decomposed transformer (FEDformer) is trained to forecast SGSI trajectories from online operating measurements together with known inverter control parameters and offline identified equivalent descriptors. The framework is tested on a modified IEEE 39-bus New England system comprising both GFM and GFL inverters and validated through electromagnetic transient simulation. Results show that the proposed method reproduces eigenvalue-based stability trends and accurately forecasts SGSI, with the FEDformer achieving the best predictive performance.