Data-Driven Approach to Capturing Wide-Area Frequency Response Dynamics

The rise in Converter Interfaced Generation (CIG) is causing system frequency response to become increasingly localised. Traditionally, frequency studies rely on the Centre of Inertia (COI)-based approach, which assumes a global frequency response and cannot adequately capture locational frequency variations. This limitation increases the risk of unforeseen relay activation, potentially leading to large-scale blackouts. This paper demonstrates the potential for locational frequency variations due to high CIG penetration, which purely COI-based methods may fail to detect. To address this, we propose a machine learning (ML) approach to effectively capture and represent these nonlinear frequency dynamics. Simulation results on the modified IEEE 39-bus test network indicate that the proposed ML approach is computationally more efficient than model-based methods while maintaining accuracy comparable to computationally intensive time-domain dynamic simulations.