PowerModels.jl and Training Grid Foundation Models

Russel Bent – LANL

In this talk, we review the steady-state load shedding formulation developed within PowerModels.jl and PowerModelsRestoration.jl. These software frameworks are designed to support the computational evaluation of emerging power network formulations and algorithms within a unified platform. The architecture cleanly decouples problem specifications (e.g., Power Flow, Optimal Power Flow, Load Shedding) from power network formulations (e.g., AC, DC approximations, SOC relaxations), enabling systematic comparison of diverse modeling approaches under common problem definitions. We then describe how the load shedding formulation was integrated into GridPACK to enable massively parallel batch processing for large-scale solution data generation, supporting the training and fine-tuning of grid foundation models.