Edge-graph convolution and multi-hop attention jointly driven small-signal stability assessment against topology changes

Data-driven small-signal assessment (SSA) is promising for long-term power system operation security. Existing SSA schemes are sensitive to unseen topology changes, especially when aimed at inter-area modes. We propose a Multi-hop Edge-Graph based Deep Learning (MEGDL) model for the SSA based on steady-state measurements. The core MEGDL includes multiple blocks. Each block utilizes not only the edge-graph convolution to jointly merge electrical variables at buses (nodes) and transmission lines (edges), but also exploits the multi-hop graph attention mechanism to discover the impact of topology changes on the high-order neighborhoods. Then the frequencies and damping ratios of various modes can be predicted in parallel via a multi-task structure with sharing features and different downstream networks. Test results on the IEEE 39 Bus system and the IEEE 118 Bus system indicate the superiority of our scheme over existing models, as well as its robustness against significant topology changes. •A small-signal assessment scheme with superior adaptivity to significant topology changes.•An edge-graph convolution extracts both node and edge features.•A multi-hop graph attention enhances model generalization under various topology changes.•Effective graph learning strategies and the multi-task scheme promotes the performance in critical LFO modes.