Topology-Aware Neural Networks for Fast Contingency Analysis of Power Systems

• DOI: 10.48550/arxiv.2310.04213
• Publication Date: 2023-01-01
• Authors: Agnes M. Nakiganda; Catherine Cheylan; Spyros Chatzivasi...
• Keywords: Control reconfiguration; Solver

Training Neural Networks able to capture the topology changes of power grid is one significant challenges towards adoption machine learning techniques for N-k security computations and a wide range other operations that involve reconfiguration. As number scenarios increases exponentially with increasing system size this renders such problems extremely time-consuming solve traditional solvers. In paper, we combine Physics-Informed both Guided-Dropout (GD) Network (which associates dedicated neurons specific line connections/disconnections) an edge-varrying Graph (GNN) architecture learn setpoints considers all probable single-line reconfigurations (all critical N-1 scenarios) subsequently apply trained models scenarios.We demonstrate how incorporating underlying physical equations network within training procedure GD GNN architectures, performs N-1, N-2, N-3 case studies. Using AC Power Flow as guiding application, test our methods on 14-bus, 30-bus, 57-bus, 118-bus systems. We find these topology-aware NNs not only achieve task contingency screening satisfactory accuracy but do at up 1000 times faster than Newton Raphson flow solver. Moreover, results provide comparison in terms computational speed recommendations their analysis