Argonne says its grid AI ran calculations 4,000 times faster in demonstrations
GridAI examines changing power-system conditions to help utilities prepare for disasters and attacks. Its reported gains also include broader outage-risk analysis.
Loading page…
GridAI examines changing power-system conditions to help utilities prepare for disasters and attacks. Its reported gains also include broader outage-risk analysis.
Listen to this story
Argonne’s GridAI is being tested as a planning tool for utilities and federal grid-security teams, with demonstrations showing much broader risk analysis alongside faster calculations. Argonne security manager Duane Rudolph Verner reported 200 times the outage-risk scope and 99% overall risk coverage, based on tens of billions of contingency calculations; these are attributed demonstration results, not independently verified benchmarks or evidence of outage prevention. The effort also aims to speed infrastructure assessments: a process that once took a year can now produce information and reports faster.
Verner said GridAI’s demonstrations ran calculations 4,000 times faster than a conventional utility solver.
Argonne researchers are training the model on data from three national laboratory supercomputers and updating its grid understanding second by second.
The reported 99% risk coverage is not a claim that GridAI prevents 99% of outages.
Argonne National Laboratory’s grid-modeling AI completed demonstration calculations 4,000 times faster than conventional utility methods, its global energy security manager said November 18. Speaking at an AFCEA Bethesda summit in Washington, D.C., Duane Rudolph Verner also described a wider reach for outage-risk analysis, with potential uses in preparing for disasters and attacks.
The model, called GridAI, is being trained by Argonne researchers to understand the systems that connect and power the electric grid. Federal News Network reported that its training uses data from three national laboratory supercomputers. Verner said the model accounts for changes in power systems on a second-by-second basis.
Verner described the work as a collaboration between Argonne and other national laboratories to examine different power-grid configurations. His speed comparison was against a conventional solver—the calculation software used at a utility. He paired that result with two measures of the scope of outage risk GridAI could examine.
Verner said GridAI covers 200 times the “outage risk aperture,” or scope of outage risk examined.
He said GridAI reached this coverage through tens of billions of contingency calculations.
These are attributed demonstration results, not independently verified benchmarks or measured reductions in outages. In particular, 99% risk coverage does not mean 99% outage prevention. Verner described the practical benefit as helping utilities and grid operators improve resilience and preparedness during natural disasters or adversarial attacks.
At the summit, Energy Department officials described using AI to identify grid vulnerabilities and understand how critical systems depend on one another. Andrew McClure, who leads the department’s Office of Cybersecurity, Energy Security, and Emergency Response, pointed to one source of strain: new power plants far from cities using the most electricity. Connecting them requires costly transmission lines, he said.
Beyond GridAI’s calculations, Verner described a shorter path from infrastructure assessments to useful information. AI is helping laboratory researchers provide real-time grid information to McClure’s office and produce assessment reports faster. He contrasted that with earlier work in which sending a team to assess infrastructure and delivering its report could take a year.
Loading discussion...
Join the conversation
Explain whether planning tools deserve different scrutiny from systems that control equipment.
Be the first to share a perspective or an experience.
Reader comments
Newest comments first. Replies stay oldest first.