GridFM
Small foundation models for the electric grid
GridSFM Is Open Source
The GridSFM topology and pipeline code are now available on our GitHub. The dataset of grid models, and the trained GridSFM-opem model are now available on HuggingFace
GridFM is a Microsoft Research initiative to build a foundation model (FM) for electric power grids, applying modern AI methods—similar to large language/weather models—to complex grid physics.
Traditional power‑flow solvers (like AC‑OPF) are accurate but extremely slow, taking minutes to hours on real-world grids with tens of thousands of components. As power systems grow more volatile due to datacenter expansion, renewable variability, electrification, and extreme weather, grid operators need fast, scalable, and generalizable models to evaluate thousands of scenarios in real time.
GridFM aims to deliver exactly that:
- Robust tools for planning, reliability analysis, and emergency management
- Rapid inference for operational decision‑making
- Physics‑informed modeling with high numerical fidelity
- Generalized representations that can be fine‑tuned to specific grid topologies