Learning Power Grid Outages with Higher-Order Topological Neural Networks

With the increase in cyber-physical threats and extreme weather events, resilience of power system has become a problem utmost societal importance. In this paper, we propose novel approach for improvement distribution networks, based on notions persistent homology simplicial neural networks (SNNs) which are new directions graph learning. particular, tools allow us to capture most essential topological descriptors network. turn, extending convolutional operation complexes network, using Hodge-Laplacian analytics, enables describe complex interactions among multi-node higher order substructures. Such substructures particular importance since change demand at load bus (or supplied from substation) will produce corresponding perturbation nodal variables (such as voltages) edge branch currents). We validate our Higher-Order Topological Neural Networks (HOT-Nets) model contingency classification three test IEEE 37-bus feeder, 123-bus 342-bus low voltage Our experiment results two case studies (i.e., (i) with sensors placed all buses alone (ii) partial observability networks) indicate that HOT-Nets substantially outperforms 9 state-of-the-art methods, yielding relative gains up 14.04% terms classification.