Physics-aware training improves power flow predictions at scale
Scaling Laws for Physics-Aware ACOPF Surrogate Learning
Machine Learning
Summary
Solving the AC optimal power flow problem is key to running electrical grids efficiently, but traditional methods can be slow. The authors study machine learning models that predict these flows faster, focusing on how well they obey physical constraints. They find that training with physics-aware objectives leads to more reliable predictions, especially as the power network grows in size. Their analysis shows this approach reduces constraint violations by a large margin, though it requires more training time.
What this means in practice
- •For power system operators: Speed up AC power flow predictions with improved physical feasibility by using physics-aware training objectives at scale.
- •For energy grid software developers: Incorporate augmented Lagrangian-based training for surrogate models to reduce constraint violations in large-scale power network simulations.
Authors
Yijiang Li, Emon Dey, Stefano Fenu, Massimiliano Lupo Pasini, Teja Kuruganti, Kibaek Kim
Abstract
Learning-based surrogates for AC optimal power flow (ACOPF) promise large speedups over classical solvers, but their operational value depends on physical feasibility as much as predictive accuracy. Physics-aware objectives such as the augmented Lagrangian (AL) improve constraint satisfaction at additional per-step cost, yet how this trade-off behaves with scale is uncharacterized. We sweep model and dataset sizes under both MSE and AL training, and characterize how constraint violation changes with network size across grids. Both objectives improve as power laws, but at different rates: MSE is governed primarily by model capacity, while AL is balanced across both. Violation grows roughly twice as fast with network size under MSE as under AL. On matched hardware, AL reduces violation by nearly $30\times$ for an order of magnitude more training time, with negligible added memory. The training objective determines not only where a surrogate lands but how its quality evolves with scale.