An interpretable hybrid graph pooling scheme for system-scale adaptive small-signal stability assessment

Aimed at increasingly challenging operation conditions in modern power systems, online small-signal stability assessment (SSA) acts as a significant tool to detect latent oscillation risks and provide abundant information for preventive controls. Existing machine learning-based SSA methods fail under small system-scale changes and encounter efficiency loss when applied to large-scale systems. The model inference lacks enough interpretability for preventive controls. In this paper, we propose an I nterpretable hyb R id gr A ph P ooling-based SSA scheme (IRAP-SSA) with excellent robustness against system-scale changes. A sparse edge contraction-based attention pooling (ECAP) is stacked to dynamically simplify the network structure without loss of representation differences. A spectral graph pooling (SGP) module works to generate fixed-dimensional area representations. The advocated I nterpretable M odules with P ost- H oc I nterpretation (IM-PHI) unveil the rationality of the system-scale robustness and discriminate vulnerable areas and dominant generators for operators. The performance as well as interpretability and generalization of our scheme are validated on the IEEE 39 Bus system and the IEEE 118 Bus system under various operation topologies and system scales. • A hybrid pooling structure ensures satisfactory generalization across scenarios with different topologies or system scales. • Interpretable Modules with Post-Hoc Interpretation (IM-PHI) enhance the model interpretability concerning topology and area representations. • A small-signal stability assessment (SSA) scheme exhibits superior reliability in practical application with high robustness and interpretability. • Model interpretability is comprehensively demonstrated through novel visualization methods.