LUMINA: A Grid Foundation Model for Benchmarking AC Optimal Power Flow Surrogate Learning

2026-05-03

AI Summary

Existing learning-based surrogate models for AC optimal power flow (ACOPF) exhibit limited generalization to unseen grid topologies, hindering their practical deployment and “what-if” analyses. To address this, this work proposes LUMINA-Bench—the first comprehensive benchmark for multi-topology generalization in ACOPF surrogate learning—which systematically evaluates model prediction accuracy and physical constraint violations under pretraining, transfer, and adaptation scenarios. We release an open-source end-to-end framework encompassing data processing, training, and evaluation, and conduct comparisons across homogeneous and heterogeneous architectures as well as various constraint-aware loss functions, including mean squared error, augmented Lagrangian, and constraint-violation Lagrangian formulations. Our experiments comprehensively reveal the trade-offs between accuracy and robustness under different training objectives, establishing an empirical foundation for developing generalizable, physics-consistent surrogate models for power grids.

Abstract

AC optimal power flow (ACOPF) is foundational yet computationally expensive in power grid operations, driving learning-based surrogates for large-scale grid analysis. These surrogates, however, often fail to generalize across network topologies, a critical gap for deployment on grids not seen during training and for routine operational what-if studies. We introduce LUMINA-Bench, a comprehensive benchmark suite for ACOPF surrogate learning covering multi-topology pretraining, transfer, and adaptation. The benchmark evaluates homogeneous and heterogeneous architectures under single- and multi-topology learning settings using unified metrics that capture both predictive accuracy and physics-informed constraint violations. We additionally compare constraint-aware training objectives, including MSE, augmented Lagrangian, and violation-based Lagrangian losses, to characterize accuracy-robustness trade-offs across settings. Data processing, training, and evaluation frameworks are open-sourced as the LUMINA suite to support reproducibility and accelerate future research on feasibility-aware OPF surrogates.

Problem

• AC optimal power flow
• surrogate learning
• generalization
• grid topology
• feasibility-aware

Innovation

• AC optimal power flow
• surrogate learning
• multi-topology generalization
• physics-informed constraints
• benchmark suite