Learning Power System Vulnerabilities through Multi-View Topological Neural Networks

The interconnected, interdependent nature of wide-area modern power systems necessitates effective integration of increasingly larger-scale, heterogeneous, and spatially distributed physical assets with a multitude of ubiquitous cyber devices. Conventional admittance-matrix-based power grid topological models have been illustrated to lack analytical capabilities to effectively model ubiquitous flexibility, capacity, and prosumer profiles and enable granular and accurate controls of the ultrahigh-dimensional distributed energy resources. In this work, we propose a novel Multi-View Topological Neural Networks (MV-TopoNets), a class of simple yet effective topological deep learning built upon persistent homology, topological convolution, and graph convolution operations. The proposed MV-TopoNets model aims to integrate ubiquitous and high-dimensional information structures on transmission nodes by taking properties of power systems as special directions, and designs new perspectives at the intersection of algebraic topology, commutative algebra, and deep learning for power system contingency analysis. Real data experiments show that the proposed MV-TopoNets outperforms 7 state-of-the-art methods on two power grid datasets.