Dynamic stability of power grids - new datasets for Graph Neural Networks

Anonymous

Keywords: Graph Neural Networks, Dynamic Stability, Power Grids, Complex Systems, Nonlinear dynamics, energy transition

TL;DR: New datasets are introdcued to the ML community regarding the dynamic stability of power grids that is very expensive and challenging to predict

Abstract: One of the key challenges for the success of the energy transition is the analysis of the dynamic stability of power grids. Graph Neural Networks (GNNs) are a promising method to reduce the computational effort of predicting dynamic stability of power grids, however datasets of appropriate complexity and size do not yet exist. In this paper we introduce new datasets of synthetic power grids and node-wise dynamic stability based on Monte-Carlo simulations. The datasets consist of a total of 20,000 grids instead of previously published work that has 2,000 grids. This enables the training of more complex models and can significantly increase the performance. The investigated grids have two sizes (20 and 100 nodes), which enables the application of transfer learning from a small to a large domain. Lastly, we provide several benchmark models to establish the feasibility of predicting dynamic stability from graph features. These models achieve surprisingly high performance, even in transfer learning, which opens the door for future application on real power grids. All Code and Data will be made available upon publication. We invite the community to improve on our benchmark models and thus aid the energy transition with better tools.