GridFM
Small foundation models for the electric grid
GridSFM is designed around four core tenets:
Topology Agnostic. A single model with shared weights processes grids of any size and shape. Buses are nodes, transmission lines are edges, and the same backbone handles a 500-bus benchmark or a 4,000-bus state-scale topology without per-grid retraining.
Feasibility Aware. Infeasibility is a first-class output, not a discarded label. GridSFM classifies every scenario as feasible or infeasible with a confidence score — useful for contingency screening, security assessment, and market-clearing validation. On the held-out test set, the classifier reaches 95.3% balanced accuracy (F1 = 0.945 on the feasible class).
Physics Grounded. Branch flows are not predicted directly; they’re derived analytically from predicted bus voltages and angles via the standard π-equivalent branch equations. Physics penalties (power balance, thermal, voltage) regularize training so outputs land on the AC-OPF manifold.
Data Efficient. Self-supervised physics constraints supplement supervised solver labels, reducing the per-topology label budget. On a brand-new grid, as few as ~10 fine-tuning scenarios already produce reasonable cost and dispatch estimates, and ~1,000 scenarios recover full in-sample performance.
“GridSFM predicts AC-OPF solutions in milliseconds: bus voltages, generator dispatch, branch power flows, and a feasibility classification without running a solver.”
What the model predicts
Given a grid topology, physical and operating constraints, generation characteristics, and a loading scenario, GridSFM produces a complete operating point — bus voltage magnitudes V and angles θ, generator active and reactive dispatch (Pg, Qg), branch active and reactive flows (Pij, Qij) — plus a feasibility verdict with a continuous margin.
Headline results (GridSFM-Open, 54-grid test corpus)
| Metric | Value |
|---|---|
| Cost MAPE | 3.35% (median 2.85%; 51/54 grids below 5%) |
| Voltage magnitude MAE | 0.0080 p.u. |
| Voltage angle MAE | 2.14° |
| Generator active power MAE | 0.092 p.u. |
| Feasibility classifier balanced accuracy | 95.3% |
| AC-OPF warm-start speedup over cold start | 1.66× geometric mean (wins on 41/54 grids) |
| Warm-start speedup vs. DC-OPF baseline | 1.59× faster than DC warm-start alone |
Used as a warm-start seed for the PowerModels.jl AC-OPF solver, GridSFM cuts solve time by 1.66× on average and captures ~61% of the theoretical headroom between a cold solve and the optimal-point ceiling. On the largest cases this reaches ~4× speedup (case1951_rte, case2868_rte) and 6–7× on a few (Texas2k summer peak, case2742_goc).
Out-of-distribution generalization and fine-tuning
On a grid 1.4× larger than anything seen in training (case6470_rte), zero-shot cost MAPE rises to ~14% — the model has learned generalizable angle and cost structure, but voltage magnitude and the feasibility classifier need calibration to the new grid. Fine-tuning on just 1,000 scenarios from the new grid restores accuracy (cost MAPE drops to 1.12%, feasibility F1 recovers to 0.99), and a held-out N-1 contingency split tracks the intact topology closely — fine-tuning on the base topology transfers cleanly to contingency variants.
