•  
  •  
 

Keywords

node vulnerability assessment, graph convolutional network, multi-objective optimization

Abstract

The continued expansion and increasing structural complexity of new power systems have made the limitations of traditional node vulnerability analysis methods increasingly evident in terms of computational efficiency，nonlinear feature analysis capability，and assessment accuracy.To address this issue，a power grid node vulnerability assessment method based on a multi-objective optimization graph convolutional network （MOO-GCN ） was first proposed.Then，a multi-objective optimization mechanism was introduced into the traditional GCN framework to construct a three-component joint loss function integrating reconstruction loss，neighborhood embedding loss，and ranking consistency loss，thus achieving comprehensive assessment through dynamic weight adjustment.Finally，the IEEE 300-node，IEEE 500-node，and IEEE 1354 -node distribution networks were used as test cases，and the model performance was comprehensively evaluated using the TOP-K Jaccard similarity，Kendall ’s tau rank correlation coefficient，and the SI spreading model，thus verifying the feasibility and effectiveness of the proposed method.The results show that the MOO-GCN-based power grid node vulnerability assessment method outperforms both traditional methods and the GCN-based method in terms of infection speed and ranking consistency，and that the proposed algorithm improves the stability and accuracy of node ranking.

DOI

10.19781/j.issn.1673-9140.2026.04.002

First Page

12

Last Page

24

Download

Share

COinS