Threshold-free full-process monitoring of power system oscillation based on data-structure feature fusion

Tools

Guo, Mou-Fa and Xu, Sheng-Tao and Zhang, Bin-Long and Hong, Qiteng (2026) Threshold-free full-process monitoring of power system oscillation based on data-structure feature fusion. IEEE Transactions on Power Systems. pp. 1-15. ISSN 0885-8950 (https://doi.org/10.1109/tpwrs.2026.3697595)

[thumbnail of Guo-etal-2026-Threshold-free-full-process-monitoring-of-power-system-oscillation]
Preview
 Text. Filename: Guo-etal-2026-Threshold-free-full-process-monitoring-of-power-system-oscillation.pdf
Accepted Author Manuscript
License: Creative Commons Attribution 4.0 logo
Download (2MB)| Preview

Abstract

The increasing penetration of renewable energy and extensive use of power electronic devices have made oscillations more prevalent in power systems. Accurate identification and tracking of oscillation evolution are essential for early warning and source tracing. However, precisely characterizing and recognizing these oscillations in real time is highly challenging due to their nonlinear, nonstationary, and multi-time-scale nature. This paper presents a full-process adaptive approach for oscillation monitoring based on data-structure feature fusion. Firstly, the intrinsic localized mode (ILM) theory is employed to qualitatively analyze the three-stage evolution of oscillation. Next, the ensemble empirical mode decomposition (EEMD) and variance contribution rate (VCR) criterion are applied to extract the dominant intrinsic mode function (IMF) components. From these components, Gramian angular summation field (GASF) phase maps and Hilbert-Huang spectrum (HHS) time-frequency maps are subsequently constructed to represent the structural features of the data. A Swin-Cross Attention Transformer (S-CAT) network is further designed to fuse heterogeneous image features and classify evolution patterns, and a decision strategy is introduced for full-process monitoring. Experimental tests show that the method achieves high recognition accuracy and offers early warning during the initial disturbance stage, demonstrating strong engineering applicability.

ORCID iDs

Guo, Mou-Fa, Xu, Sheng-Tao, Zhang, Bin-Long and Hong, Qiteng ORCID logoORCID: https://orcid.org/0000-0001-9122-1981;
CORE (COnnecting REpositories)