Predicting basin stability of power grids using graph neural networks

Abstract The prediction of dynamical stability power grids becomes more important and challenging with increasing shares renewable energy sources due to their decentralized structure, reduced inertia volatility. We investigate the feasibility applying graph neural networks (GNN) predict dynamic synchronisation in complex using single-node basin (SNBS) as a measure. To do so, we generate two synthetic datasets for 20 100 nodes respectively estimate SNBS Monte-Carlo sampling. Those are used train evaluate performance eight different GNN-models. All models use full without simplifications input nodal-regression-setup. show that can be predicted general significantly changes Furthermore, observe interesting transfer capabilities our approach: GNN-models trained on smaller directly applied larger need retraining.