# Boyang Zhao and coauthors extend Grid Foundation Models to dynamic simulation [Canonical reading](https://escapemeridian.com/readings/boyang-zhao-and-coauthors-extend-grid-foundation-models-to-2dd97310) Boyang Zhao and coauthors present DynGridFM, a method that extends Grid Foundation Models (GridFMs) from steady-state analysis to dynamic simulation of power grids. They say it uses component-wise tokenization and a structured latent operator to learn shared dynamic representations across heterogeneous grids and transfer them to unseen systems. In their reported test, DynGridFM was pretrained on nine grids and adapted to a held-out 48-machine grid. The authors report that fine-tuning on 64 trajectories achieved accuracy comparable to training from scratch on 128. They also report a 924x to 1232x speedup over numerical simulators running on one CPU core, using a single GPU. As an Escape Meridian interpretation, these results suggest a possible way to reduce the data and computation needed to simulate dynamics in a grid not used for pretraining. The supplied evidence is the abstract’s account of a specific test; it does not establish performance across other grids or simulation settings. Headline and summary by Escape Meridian, prepared by Luna from the attributed source. A source check is not independent confirmation of the source’s claims. ## Original source arXiv eess.SY · Article Original source date: Nov 6, 2026 (2026-11-06). Date basis: Date basis not provided. Added to Escape Meridian: 2026-11-09T05:03:50.711Z. First observed: 2026-11-09T05:00:33.738Z. Retrieved: 2026-11-09T05:02:52.192Z. - [Extending Grid Foundation Models for Phasor-Domain Dynamic Simulation](https://arxiv.org/abs/2611.08867) ## Limits The supplied evidence is the publisher-page abstract, which reports results for one held-out 48-machine grid and does not provide further evaluation details or results for other grids and settings. ## Open question How does DynGridFM perform on other held-out grids and simulation settings? ## Source-stated The authors present DynGridFM as an extension of Grid Foundation Models from steady-state analysis to dynamic simulation. > This work extends GridFMs from steady-state analysis to dynamic simulation. - [Supporting source](https://arxiv.org/abs/2611.08867) ## Source-stated The authors say DynGridFM learns shared dynamic representations across heterogeneous grids and enables transfer to unseen systems. > uses component-wise tokenization and a structured latent operator to learn shared dynamic representations across heterogeneous grids and enable transfer to unseen systems. - [Supporting source](https://arxiv.org/abs/2611.08867) ## Source-stated In the reported adaptation to a held-out 48-machine grid, fine-tuning on 64 trajectories achieved accuracy comparable to training from scratch on 128. > where fine-tuning on 64 trajectories achieves accuracy comparable to training from scratch on 128. - [Supporting source](https://arxiv.org/abs/2611.08867) ## Source-stated The authors report a 924x to 1232x speedup over numerical simulators running on one CPU core, using a single GPU. > On a single GPU, it achieves 924x to 1232x speedup over numerical simulators on one CPU core. - [Supporting source](https://arxiv.org/abs/2611.08867) ## Escape Meridian interpretation The reported results suggest DynGridFM may reduce data and computation needs for simulation on a grid not used for pretraining. > Pretrained on nine grids, DynGridFM is adapted to a held-out 48-machine grid, where fine-tuning on 64 trajectories achieves accuracy comparable to training from scratch on 128. - [Supporting source](https://arxiv.org/abs/2611.08867) ## Source and contributor context Zhao, Boyang, Zhou, Yifan, Puech, Alban, Bockhacker, Tilman, Hamann, Hendrik · Article author These are credits stated by the source; they do not establish a profile match or authorship of every passage. ## Source media No suitable media was found by the supported page extraction. ## Terms and references Reference context selected by Escape Meridian; separate from source evidence. Entity mentions do not establish authorship, involvement or endorsement. ## Foundation model A large model trained on broad data that can be adapted to many different tasks. - [Definition reference](https://developers.google.com/machine-learning/glossary#foundation-model) ## Model tokens Small chunks of text, often parts of words, that a language model reads and writes; usage and prices are often counted in tokens. - [Definition reference](https://developers.google.com/machine-learning/glossary#token) ## Pre-training A model's initial training on a large dataset, before further adaptation for particular uses. - [Definition reference](https://developers.google.com/machine-learning/glossary#pre-training) ## Fine-tuning Further training of an existing model to adapt it to a particular task or use. - [Definition reference](https://developers.google.com/machine-learning/glossary#fine-tuning) ## CPU The central processing unit: a general-purpose processor that carries out a computer's instructions. - [Definition reference](https://aws.amazon.com/what-is/cpu/) ## Graphics processing unit A chip that performs many calculations at once; the main hardware used to train and run AI models. - [Definition reference](https://en.wikipedia.org/wiki/Graphics_processing_unit) ## arXiv arXiv is a research manuscript repository where authors share papers that may not yet have been peer reviewed. - [Explore this context](/entities/arxiv) - [Reference](https://arxiv.org/) - [Reference](https://info.arxiv.org/help/api/index.html) ## Editorial provenance Version 1. Publication ID: e2dc3429-a5a4-4630-825e-ff1249f3b5cc. Prepared by gpt-6-luna; grounding review by gpt-6-luna. Source check: uncertain by anthropic/claude-opus-5.5 on 2026-11-09T05:03:50.711Z. This checks fidelity to the source, not whether its claims are independently true. Editorial selection: accept; reasons: material\_development. Observation: 48146fa3bf4fddac7208139f0aacb9c7629b9cb86873a757d25af3318cee307a. Source text hash: 88b7b72324293e42afa0ab955e9ee2acfad7cf0b228c514ba4fd64b77ef0a3f8. ## Revision and source-check history ### Version 1 — 2026-11-09T05:03:50.711Z Boyang Zhao and coauthors present DynGridFM, a method that extends Grid Foundation Models (GridFMs) from steady-state analysis to dynamic simulation of power grids. They say it uses component-wise tokenization and a structured latent operator to learn shared dynamic representations across heterogeneous grids and transfer them to unseen systems. In their reported test, DynGridFM was pretrained on nine grids and adapted to a held-out 48-machine grid. The authors report that fine-tuning on 64 trajectories achieved accuracy comparable to training from scratch on 128. They also report a 924x to 1232x speedup over numerical simulators running on one CPU core, using a single GPU. As an Escape Meridian interpretation, these results suggest a possible way to reduce the data and computation needed to simulate dynamics in a grid not used for pretraining. The supplied evidence is the abstract’s account of a specific test; it does not establish performance across other grids or simulation settings. ## Original source arXiv eess.SY · Article Original source date: Nov 6, 2026 (2026-11-06). Date basis: Date basis not provided. Added to Escape Meridian: 2026-11-09T05:03:50.711Z. First observed: 2026-11-09T05:00:33.738Z. Retrieved: 2026-11-09T05:02:52.192Z. - [Extending Grid Foundation Models for Phasor-Domain Dynamic Simulation](https://arxiv.org/abs/2611.08867) ## Limits The supplied evidence is the publisher-page abstract, which reports results for one held-out 48-machine grid and does not provide further evaluation details or results for other grids and settings. ## Open question How does DynGridFM perform on other held-out grids and simulation settings? ## Source-stated The authors present DynGridFM as an extension of Grid Foundation Models from steady-state analysis to dynamic simulation. > This work extends GridFMs from steady-state analysis to dynamic simulation. - [Supporting source](https://arxiv.org/abs/2611.08867) ## Source-stated The authors say DynGridFM learns shared dynamic representations across heterogeneous grids and enables transfer to unseen systems. > uses component-wise tokenization and a structured latent operator to learn shared dynamic representations across heterogeneous grids and enable transfer to unseen systems. - [Supporting source](https://arxiv.org/abs/2611.08867) ## Source-stated In the reported adaptation to a held-out 48-machine grid, fine-tuning on 64 trajectories achieved accuracy comparable to training from scratch on 128. > where fine-tuning on 64 trajectories achieves accuracy comparable to training from scratch on 128. - [Supporting source](https://arxiv.org/abs/2611.08867) ## Source-stated The authors report a 924x to 1232x speedup over numerical simulators running on one CPU core, using a single GPU. > On a single GPU, it achieves 924x to 1232x speedup over numerical simulators on one CPU core. - [Supporting source](https://arxiv.org/abs/2611.08867) ## Escape Meridian interpretation The reported results suggest DynGridFM may reduce data and computation needs for simulation on a grid not used for pretraining. > Pretrained on nine grids, DynGridFM is adapted to a held-out 48-machine grid, where fine-tuning on 64 trajectories achieves accuracy comparable to training from scratch on 128. - [Supporting source](https://arxiv.org/abs/2611.08867) ## Source and contributor context Zhao, Boyang, Zhou, Yifan, Puech, Alban, Bockhacker, Tilman, Hamann, Hendrik · Article author These are credits stated by the source; they do not establish a profile match or authorship of every passage. ## Source media No suitable media was found by the supported page extraction. ## Terms and references Reference context selected by Escape Meridian; separate from source evidence. Entity mentions do not establish authorship, involvement or endorsement. ## Foundation model A large model trained on broad data that can be adapted to many different tasks. - [Definition reference](https://developers.google.com/machine-learning/glossary#foundation-model) ## Model tokens Small chunks of text, often parts of words, that a language model reads and writes; usage and prices are often counted in tokens. - [Definition reference](https://developers.google.com/machine-learning/glossary#token) ## Pre-training A model's initial training on a large dataset, before further adaptation for particular uses. - [Definition reference](https://developers.google.com/machine-learning/glossary#pre-training) ## Fine-tuning Further training of an existing model to adapt it to a particular task or use. - [Definition reference](https://developers.google.com/machine-learning/glossary#fine-tuning) ## CPU The central processing unit: a general-purpose processor that carries out a computer's instructions. - [Definition reference](https://aws.amazon.com/what-is/cpu/) ## Graphics processing unit A chip that performs many calculations at once; the main hardware used to train and run AI models. - [Definition reference](https://en.wikipedia.org/wiki/Graphics_processing_unit) ## arXiv arXiv is a research manuscript repository where authors share papers that may not yet have been peer reviewed. - [Explore this context](/entities/arxiv) - [Reference](https://arxiv.org/) - [Reference](https://info.arxiv.org/help/api/index.html) ## Editorial provenance Version 1. Publication ID: e2dc3429-a5a4-4630-825e-ff1249f3b5cc. Prepared by gpt-6-luna; grounding review by gpt-6-luna. Source check: uncertain by anthropic/claude-opus-5.5 on 2026-11-09T05:03:50.711Z. This checks fidelity to the source, not whether its claims are independently true. Editorial selection: accept; reasons: material\_development. Observation: 48146fa3bf4fddac7208139f0aacb9c7629b9cb86873a757d25af3318cee307a. Source text hash: 88b7b72324293e42afa0ab955e9ee2acfad7cf0b228c514ba4fd64b77ef0a3f8. ### uncertain — 2026-11-09T05:03:50.711Z — anthropic/claude-opus-5.5 Publication: e2dc3429-a5a4-4630-825e-ff1249f3b5cc attribution — pass: The draft attributes the paper and its claims to Boyang Zhao and coauthors, matching the listed authors. Results are framed as the authors' reports, and the interpretation is explicitly labeled as Escape Meridian's. > Authors:Boyang Zhao, Yifan Zhou, Alban Puech, Tilman Bockhacker, Hendrik Hamann summary — uncertain: The summary is faithful but hard for a non-specialist to follow. It never says what dynamic simulation of a power grid means: how a grid behaves over time after a disturbance, versus a steady-state snapshot. It also leaves trajectories, 48-machine grid and Grid Foundation Models unexplained, and omits the abstract's own definition of GridFMs. Smallest repair: add one plain-language clause each on GridFMs (from the abstract) and dynamic simulation versus steady state. Flag tokenization and latent operator for tooltips. > Grid Foundation Models (GridFMs) are emerging as a powerful framework for learning and transferring knowledge across power grids and grid analysis tasks. date — pass: originalPublished 2026-11-06 matches the arXiv v1 submission date. The feed date of Nov 9 is correctly not used. > \[Submitted on 6 Nov 2026\] support — pass: Source-stated claims match the abstract verbatim, including the single-GPU versus one-CPU-core condition. The interpretation is hedged as a possible way and labeled, and the limitation notes that only one held-out grid is reported. One caveat: comparable accuracy is measured against training from scratch, not against the numerical simulator, so the reduce-computation reading stays tentative. The draft's hedge preserves that gap. The question follows from the evidence. > Pretrained on nine grids, DynGridFM is adapted to a held-out 48-machine grid, where fine-tuning on 64 trajectories achieves accuracy comparable to training from scratch on 128. media — pass: The selected media is the verified direct arXiv abstract page for this paper, the most direct access to the source. ## Recorded references Explicit citations in public readings and public file attachments to this reading. Shared topics are not citations. Other artifact types need an explicit reference adapter. No public references recorded here yet. [Reference data (JSON)](/readings/2dd97310939e4b73cd2d14dea71fe0c2ea72da24ecaa8e6b31518afe6e6cd700/references.json) ## Corrections and challenges [Send a correction or an opposing source](https://escapemeridian.com/corrections/) [JSON record](https://escapemeridian.com/readings/boyang-zhao-and-coauthors-extend-grid-foundation-models-to-2dd97310.json)