1 of 10

Grid Foundation Models(GridFM)� �

Community and Technology

2 of 10

Why AI for the electric Grid? To accelerate grid growth safely and securely…

*US Numbers

AI enabled?

*

3 of 10

Why AI for the electric Grid? �To cope with the increasing Complexity and Uncertainty of a rapidly expanding grid…

New generation

AI enabled?

Future

4 of 10

But why Foundation Models �vs. traditional AI?

  • Foundation Model Technology can be grid topology robust,
  • data efficient while accelerating simulations accurately

5 of 10

Foundation Model Technology is a perfect �platform to enable collaboration to advance AI for electric grid applications

Community

gridfm.org

Individual members

6 of 10

The GridFM Community was created…to develop GridFM

“IBM Yorktown”

workshop

Founding of

GridFM working group

(now 380+ members from 120+ orgs)

“Imperial College”

workshop

Joule GridFM

Perspective Paper

GridFM

Linux Foundation

Energy project

“Aachen” workshop

​

​

​

​

​

​

Dec 24

Mar 25

Jun 24

Feb 24

IBM Research © 2024 IBM Corporation

Subgroups launched

Technology, Governance, Collaboration

GridFM Press

Communication

“Argonne ” workshop

GridFM-v0

developed

Sept 25

Argonne National Laboratory

February 11-13, 2025, Lemont, Il, USA

​

​

E.ON Energy Research Center

Sept.8/9, 2025, Aachen, Germany

​

​

gridfm-datakit

gridfm-graphkit

Harvard workshop

?

​

​

​

​

​

​

Aug 25

Feb 26

7 of 10

GridFM Starter Kit – Try it Out!!

Accessible on Linux Foundation Energy:

  • Synthetic data generator: gridfm-datakit
  • Training & fine-tuning: gridfm-graphkit
  • License: Apache V2.0
  • PyPI Packages

8 of 10

Collaborative Training without Data Sharing

Collaboratively training GridFM across multiple institutions without sharing data

​

  • APPFL: Open-source privacy-preserving federated learning (PPFL) framework
    • Available in Python/PyTorch and Github.
    • Supports:
      • Privacy-preserving techniques
      • Scalable deployment (supercomputers, cloud, edge devices)
    • Extensively tested on: DOE supercomputers, AWS/Google cloud systems, edge devices
    • Funded by the U.S. Department of Energy Office of Science.

Privacy-Preserving Federated Learning

8

https://appfl.ai

Client 2

Client 1

Client 4

Server

Client 3

+

9 of 10

Join the GridFM Community

gridfm.org

Subscribe

10 of 10

GridFM Roadmap

#

Pre-training

Enabled Applications

1

Bus value

Reconstruction

​

​

​

​

  • Load flow
  • Optimal Power Flow
  • State estimation
  • (N-k) Contingency Analysis with k>1

​

2

+ Temporal

Reconstruction

​

​

​

​

  • Load forecasting
  • Renewable energy forecasting
  • Look-ahead power flow
  • Look-ahead state transition
  • Transient stability analysis

​

3

++ Edge

Reconstruction

​

​

​

​

  • Expansion planning
  • Cybersecurity
  • Control operations