{"resourceId":"pnnl-genesis-awards-si-grid-planning-surrogates-20261029","versions":[{"version":"external-ff1d2dd20d7c7454f38646fe557c4a3ccee480359b254e9578579f47a0ed8cb3","resource":{"id":"pnnl-genesis-awards-si-grid-planning-surrogates-20261029","title":"PNNL wins two Genesis Mission awards to test AI surrogates for grid simulation and scenario screening, with six months to show results","organization":"Pacific Northwest National Laboratory, with Argonne, Sandia, Berkeley Lab, Arizona State, Penn State, UC San Diego, Washington State, Siemens, Xcel Energy Services and ERCOT","sector":"National laboratory and university research consortia (energy systems modeling)","geography":"United States","publishedAt":"29 October 2026 (PNNL feature)","publicationDate":"2026-10-29","eventDate":null,"sourceName":"Pacific Northwest National Laboratory","sourceLabel":"PNNL, \"Two Genesis Awards Support SI-Powered Grid Planning,\" 29 Oct 2026","sourceUrl":"https://www.pnnl.gov/news-media/two-genesis-awards-support-si-powered-grid-planning","evidenceClass":"vendor-claim","outcomeClass":"emerging","topics":["infrastructure","operating-model"],"finding":"DOE awarded Genesis Mission funding to two PNNL-involved grid-planning projects. The first, led by PNNL's Marcelo Elizondo with Argonne, Sandia, Siemens, Washington State University, Xcel Energy Services and ERCOT, will train machine-learning approximations (\"a model of a model\") of electromagnetic transient simulations, which run at microsecond resolution and are computationally heavy. The second, led by Arizona State University with PNNL, Berkeley Lab, Penn State and UC San Diego, will build a model to pick significant scenarios out of roughly ten thousand produced by capacity expansion models, and will test it against a traditional ensemble. Both teams have six months to demonstrate their goals. No results are reported yet.","sledRelevance":"Shows how federal AI-for-science money is reaching university and lab teams: short, multi-institution proof-of-concept awards on computationally heavy models, with a stated comparison against the conventional method. Research offices and research computing groups can expect similar six-month deliverables and the compute and data demands that come with them.","evidence":"Read the full feature. It describes plans and an evaluation approach, not outcomes. Award amounts, compute allocations and success criteria are not given.","architectureImplications":"Surrogate models trained on existing simulation output need large training runs and access to the original simulators for comparison.","governanceImplications":"The planned comparison with a traditional ensemble is the key validation step; the article does not say who judges success.","securityPrivacyImplications":"Grid models from utilities and ERCOT can be sensitive; data-sharing terms are not described.","caveats":"Institutional news about newly funded work; optimistic quotes and no results. Six-month timelines may not allow independent validation. Grid relevance overlaps the utility sector; included here for the research-funding model.","streamIds":["research"],"roles":{"sales":"Lead with readiness for short federal AI-for-science awards. PNNL said on 29 October that two Genesis Mission projects, one led by Arizona State with Penn State and UC San Diego, have six months to show that AI surrogates can speed grid simulation and scenario screening. University research offices joining such teams need compute, data agreements and validation plans quickly. Offer a bounded readiness review. Do not claim speedups; the projects have reported no results.","engineering":"Frame fit around surrogate models trained on existing simulation output. The projects will approximate microsecond-resolution transient simulations and screen about ten thousand capacity-expansion scenarios, then compare against a traditional ensemble. Prerequisites are access to the original simulators, enough GPU and CPU capacity for training and comparison runs, and shared data terms with utility partners. Validate by reproducing a known set of scenarios before trusting the surrogate on new ones.","delivery":"Treat delivery as a six-month proof of concept with a pre-agreed comparison. Work covers data agreements among labs, universities and utilities, training runs, and the side-by-side test against conventional models. Ownership sits with the principal investigators, supported by research computing. Governance checkpoints cover data sensitivity and who signs off on validation. Acceptance criteria are proposals, such as matching conventional results within an agreed tolerance, not results PNNL reported."},"roleDetails":{"sales":{"customerProblem":"Grid and energy models now produce more scenarios and heavier simulations than teams can run or review in time.","stakeholders":["Vice president for research","Principal investigators in energy systems","Research computing director","Sponsored programs office"],"discoveryQuestions":["Are your researchers on Genesis Mission or similar short-deadline AI awards?","Can your clusters absorb surrogate training runs at short notice?","Are data agreements with utility partners in place?"],"valueHypothesis":"Prepared compute and data arrangements may help teams meet six-month proof-of-concept deadlines.","potentialEngagement":"A readiness review of compute, data agreements and validation plans for one funded project.","unsupportedClaims":["Measured simulation speedups","Accuracy of AI surrogates","Award sizes or follow-on funding"]},"engineering":{"fit":"Fits teams with mature physics simulators and enough output to train approximations.","architectureIntegration":"Machine-learning surrogates trained on simulator output, run alongside the original models for comparison.","prerequisites":["Access to original simulation codes","Training-scale GPU capacity","Shared data storage with partners"],"constraints":["Six-month timeline","Utility data sensitivity","No published success criteria"],"security":"Apply partner data-handling terms to grid models and keep surrogate training data in controlled storage.","proofOfValue":["Reproduce known scenarios with the surrogate","Compare scenario selection against a full ensemble","Record compute used versus the conventional run"]},"delivery":{"work":"Set up data agreements, run training, and conduct the planned comparison with conventional models.","dependencies":["Utility and ISO data partners","Lab and university compute allocations","Sponsored programs office"],"ownership":"Principal investigators with research computing support.","skillsAdoption":"Power-systems researchers learn surrogate-model training and validation.","governanceCheckpoints":["Data classification before sharing","Agreed validation tolerance before training","Sign-off on comparison results"],"acceptanceCriteria":["Surrogate matches conventional results within an agreed tolerance (proposal)","Comparison completed within the award period","Data terms documented for all partners"],"risks":["Surrogates failing on unusual scenarios","Timeline too short for validation","Data-sharing delays"]}},"retrievedAt":"2026-10-30T22:01:39-05:00","enrichedAt":"2026-10-30T22:01:43-05:00","enrichmentBasis":"retrieved source","accessibilityWorkforceImplications":"Not addressed.","procurementImplications":"Industry partners (Siemens, Xcel, ERCOT) are named; intellectual property and data terms are not described.","operatingModelImplications":"Multi-institution teams with short deadlines need fast access to compute and shared data agreements.","peerReview":"unknown","sourceVerification":{"openedUrl":"https://www.pnnl.gov/news-media/two-genesis-awards-support-si-powered-grid-planning","referenceExcerpt":"The two teams have six months to demonstrate their research goals","promptVersion":"sled-research-v4.0","model":null,"basis":"agent-reported inspection"}}}]}