PAGN: A Prony-based Adaptive Graph Network for Wideband Oscillation Source Localization

The transition from traditional power grids to renewable-integrated systems has introduced higher-frequency oscillations as a major challenge in modern power systems, heightening the need for automatic localization of wideband oscillation sources. Existing research primarily focuses on localizing low-frequency oscillations, which poses limitations on a wider frequency range. To address this issue, a Prony-based Adaptive Graph Network (PAGN) is proposed to adaptively localize the wideband oscillation in a finer granularity, enabling precise localization of oscillation sources. Specifically, the optimal window size for measurement segmentation is determined by a Prony adapter to enhance the model's ability to effectively handle wideband oscillations, taking into account both spatial and temporal domain sources. Extensive experimental results demonstrate that PAGN achieves state-of-the-art performance, surpassing both baseline and advanced models by a significant margin, underscoring its effectiveness in wideband forced oscillation localization tasks.