GAIM: Graph-based AI Monitoring Tools for Complex-Systems

The project aims to develop a heterogeneous Graph Neural Network (GNN) architecture capable of identifying and learning active topologies in distribution grids based on measurement data. This research seeks to address the challenges posed by varying topologies, ensuring a feasible solution that operates within a guaranteed convergence time. A novel GNN architecture will estimate the system state.

To validate the approach, real-world distribution grid data from Stedin, including scenarios with inaccurate topologies, will be collected and analyzed. The developed GNN model will be implemented and rigorously tested against this dataset, with a comparative analysis against Kalman filters and other commercial state estimators providing a benchmark for performance evaluation. The robustness of the approach under varying operating conditions and levels of uncertainty will also be assessed, ensuring the model’s reliability in real-world applications. In addition to performance evaluation, the study will examine the safety implications of integrating AI-based topology identification into power system operations, with recommendations provided to facilitate safe and effective decision-making by grid operators. The project’s expected outcomes include a proof-of-concept implementation of a GNN-based state estimator for distribution networks in collaboration with Stedin. Additionally, it will provide a performance comparison between the GNN-based approach and conventional state estimation methods, assessing the impact of topology variations and measurement uncertainties on model accuracy. Finally, this project will provide operational guidelines and recommendations to help operators leverage AI-based state estimators effectively and safely.

This research project is funded by NWO Topsector ICT Holland High Tech Systems and Materials as a public–private partnership project. The project duration is from 2025 to 2029 and is carried out in collaboration with Stedin.