Argonne Team Takes Third in Grid Control Contest
A joint Argonne and SAGE lab team built an AI agent that speeds up power grid action selection 68-fold.
Updated on Nov. 4, 2026 in Energy
A team from Argonne National Laboratory's MCS division and Virginia Tech's SAGE lab took third place in an international grid-control competition. The team's agent ranks candidate grid actions and delivered a 68-fold speedup over conventional search methods.
Why it matters
Testing every candidate action against a physics-based simulator is too slow for real-time grid operation, so faster selection methods could make grid control more responsive. The approach keeps physics-based safety checks before final actions are applied.
The team reduced roughly 73,000 permissible grid reconfigurations to a curated library of about 1,000 and achieved a 68-fold speedup over conventional search methods. The source does not report results beyond the competition's own scoring.
The players
Argonne National Laboratory
Argonne National Laboratory is a U.S. Department of Energy science and engineering research facility whose MCS division contributed to the team's work.
Virginia Tech
Virginia Tech is a public research university whose SAGE lab and Ph.D. candidate were part of the team.
The details
Researchers ran the costly computation offline and deployed a fast agent for run time. The agent uses a four-layer graph attention neural network and a conditional value-at-risk criterion, and at run time it simulates the loss of every grid element before applying an action.
Timeline
Summer 2026: Ahmad Al-Tawaha served as a Givens Associate at Argonne.
Several weeks during the competition: The selection phase ran and allowed multiple submissions.
November 4, 2026: Argonne published the article announcing the third-place finish.
Big Picture
The result comes from a collaboration between Argonne's MCS division and Virginia Tech's SAGE lab, extending that institutional research effort into power grid control. It reflects the offline-computation, fast-runtime-agent pattern that the team used to meet real-time operating constraints.
If methods like this reach operational use, grid operators could evaluate control actions faster during disturbances. The source does not state any timeline for deployment or consumer-facing effects.
The takeaway
Offline computation paired with fast run-time agents is one route to making rigorous grid-safety checks practical in real time. The competition result shows the approach can rank many candidate actions quickly while keeping physics-based checks in place.
Further reading
Explore more on Energy developments in grid operations and power systems.
Source note: This article includes information reported by Argonne National Laboratory.
