GridFM – Proof of Concept Empowering Diverse Downstream Tasks

Etienne Vos – IBM

Foundation models (FMs) have proliferated and transformed many industries over the past couple of years. GridFM represents an opportunity to apply the same FM paradigm to the electric grid by leveraging from abundant power system data available, and adapting (fine-tuning) it to various “downstream tasks”. In this presentation we will report on our progress towards developing GridFM, showing generalizability across network topologies and topology perturbations, to serve regression tasks such as power-flow and state-estimation. We will introduce our recent release of the GridFM “Starter-Kit”, consisting of the Graph-Kit and Data-Kit repositories, which serve as the first open-source code contribution to further the vision and democratize the capabilities of GridFM to the power system community. We will also report on the performance of GridFM’s capabilities in tasks like expansion planning, congestion management, and stability analysis, highlight GridFM’s computational advantages over traditional solvers, and discuss future plans. This endeavour is being driven by a collaborative effort under the LF Energy umbrella.