From Open Data To Foundation Models: Building GridSFM for Power Grid An...
summary
GridSFM delivers an open-source pipeline and foundation model that converts OpenStreetMap, U.S. EIA, U.S. Census and Highfield boundaries into AC-OPF-solvable continental-scale models (48 states plus six multi-region cases, ~20k buses). The dataset and model train on ~200 grids and >500k scenarios, label feasibility, and predict operating points in milliseconds with 96% feasibility accuracy. Using predicted states to warm-start solvers yields an average 1.66× faster warm-start and speeds up 39 of 54 unseen test grids; code and checkpoints live on Hugging Face.
AI-written summary · about
chapters (5)
- 0:00Introduction to GridSFM
- 2:52Open data pipeline for grids
- 18:36GridSFM foundational model
- 25:24Open tools and future work
- 26:12Q&A session
related talks
31mGridFM - Towards Foundation Models for the Electric Grid
14mLF Energy’s Power Grid Model Project Tackles Complex Grid Calculations Efficiently
46mPanel: Expanding the OpenSynth Ecosystem: Open Source Data and AI for Power Systems Modelling
13mLightning Talk: Open-source Cutting Edge Algorithms for Mod...
10mLightning Talk: Integrating CGMES-Based Grid Models Using the Generic Branch in Power...
22mUnderstanding our cities with FOSS4G
25mThe Value of Open Source Short-term Energy Forecasting Demo...
13mOpen Source Can Help With How We Consume And Produce Electricity