|
| 1 | +--- |
| 2 | +layout: post |
| 3 | +title: "JuMP powers foundational models for the electric grid" |
| 4 | +date: 2026-06-28 |
| 5 | +categories: [open-energy-modeling] |
| 6 | +author: "Oscar Dowson" |
| 7 | +--- |
| 8 | + |
| 9 | +Microsoft Research [recently published GridSFM](https://www.microsoft.com/en-us/research/blog/gridsfm-a-new-small-foundation-model-for-the-electric-grid/), |
| 10 | +a small foundation model for the electric grid. We were pleased to notice that |
| 11 | +the training pipeline is powered by JuMP and [PowerModels.jl](https://lanl-ansi.github.io/PowerModels.jl/stable/). |
| 12 | + |
| 13 | +To generate the training data for GridSFM, the Microsoft team solved hundreds of |
| 14 | +thousands of AC optimal power flow (AC-OPF) problems across 150+ real grid |
| 15 | +topologies. They did this using [PowerModels.jl](https://lanl-ansi.github.io/PowerModels.jl/stable/), |
| 16 | +a Julia package for power network optimization built on JuMP, with Ipopt as the |
| 17 | +underlying solver. Their code is open source on GitHub at [microsoft/GridSFM](https://github.com/microsoft/GridSFM/tree/1ca775fd436d7ce013a1c0ab946e61ac7ef59ad6/power_grid/US/topology_solver_pipeline). |
| 18 | + |
| 19 | +A related project is the [GridFM DataKit](https://github.com/gridfm/gridfm-datakit), |
| 20 | +which similarly uses JuMP and PowerModels.jl to generate training data for grid |
| 21 | +foundation models. The team behind it have [published an arXiv paper](https://arxiv.org/pdf/2512.14658) |
| 22 | +describing their work. An interesting aspect of GridFM DataKit is that the |
| 23 | +front-end is a Python package, but it calls Julia and PowerModels via |
| 24 | +[juliacall](https://juliapy.github.io/PythonCall.jl/stable/juliacall/). |
| 25 | + |
| 26 | +It's great to see JuMP and the broader Julia ecosystem being used at this scale. |
| 27 | + |
| 28 | +You can learn more about foundation models of the electric grid by watching the |
| 29 | +recordings of the [5th Workshop on Foundation Models of the Electric Grid](https://gridfm.org/harvard/). |
| 30 | + |
| 31 | +If your work uses JuMP, we'd love to hear about it. The best way to tell us is |
| 32 | +by [opening a GitHub issue](https://github.com/jump-dev/jump-dev.github.io/issues). |
0 commit comments