A measurement-driven graph learning framework for Wideband oscillation localization in power systems

Area of Science:

• Electrical Engineering
• Power Systems Analysis
• Signal Processing

Background:

• Converter-dominated power systems face challenges with wideband oscillations due to their complex, non-stationary nature.
• Conventional localization methods struggle with multi-band oscillations and weak spatial coherence, particularly for medium- and high-frequency components.

Purpose of the Study:

• To develop an advanced method for wideband oscillation source localization in power systems.
• To address the limitations of existing techniques in handling complex oscillation characteristics.

Main Methods:

• A novel spatiotemporal graph learning framework, LCGS-Net (Local-Contrast Global-Smooth Network), is proposed.
• A dual-branch representation disentangles global smooth patterns from local perturbations.
• Hierarchical channel interaction and adaptive fusion strategies enhance feature representation and information balancing.

Main Results:

• LCGS-Net demonstrates superior performance compared to baseline models in simulations on IEEE benchmark systems.
• The framework shows significant improvements in localizing high-frequency oscillation sources.
• Initial validation on the IEEE 39-bus system suggests scalability.

Conclusions:

• LCGS-Net effectively captures the local contrastive spatial characteristics of wideband oscillations.
• The proposed method offers a promising solution for wideband oscillation source localization in complex power systems.
• Further validation on large-scale, realistic grids is recommended.