GridForge: A Tractable Network-Flow Method for Constructing Open, Continental-Scale Synthetic Power Grids
The goal of this paper is to construct open, continental-scale synthetic transmission grids with an optimization method that stays tractable. Security restrictions keep models of the real grid out of open research and machine-learning datasets, and optimization-based construction is hard at scale: the number of candidate lines grows combinatorially, so electrically faithful methods prune the candidates in advance. This paper presents GridForge, which keeps every bus pair within physical length limits as a candidate, selects lines with Garver's length-weighted network-flow linear program instead of binary decisions, and resizes them with a sequential direct-current (DC) power flow. Two stages of 13 regional networks are evaluated. Their node-degree distributions are nearly as strongly log-linear as those of the real Eastern and Western Interconnections ($R^2$ of 0.92 to 0.98 against 0.971 and 0.981), although the areas under the fitted lines differ from the real ones by 39.9% on average. On the upper Midwest and Tennessee Valley footprints, resizing lowers the DC power-flow loss from 5.89% to 4.14% and from 14.55% to 7.05%, and the DC optimal power-flow loss from 4.03% to 2.02% and from 5.54% to 2.47%. Without relaxing any line limit, the DC optimal power flow is feasible in 78.81% and 0.82% of the scenarios. Most lines are built at 69 kV. The models span 250 to 78,484 buses, were built on single machines, and are freely redistributable.
Authors
- Thomas J. Overbye (ORCID: https://orcid.org/0000-0002-2382-2811)
- Christopher L. DeMarco (ORCID: https://orcid.org/0000-0002-8817-9804)
- Jonathan Snodgrass (ORCID: https://orcid.org/0000-0003-2305-8136)
- Bernard C. Lesieutre (ORCID: https://orcid.org/0000-0002-5274-324X)
Publication Details
- Journal
- arXiv (Cornell University)
- Published
- 2026-11-12
- DOI
- https://doi.org/10.48550/arxiv.2611.16222
- Primary Topic
- Electric Power System Optimization
- Type
- preprint