![[2602.23188] Efficient Real-Time Adaptation of ROMs for Unsteady Flows Using Data Assimilation](https://arxiv.org/static/browse/0.3.4/images/arxiv-logo-fb.png){kind=logo}
[2602.23188] Efficient Real-Time Adaptation of ROMs for Unsteady Flows Using Data Assimilation
Summary
This article presents a novel retraining strategy for Reduced Order Models (ROMs) that enhances real-time adaptation for unsteady flows using data assimilation techniques, achieving high accuracy with reduced computational demands.
Why It Matters
The proposed method addresses the challenge of efficiently adapting models for complex fluid dynamics scenarios, which is crucial for real-time applications in engineering and scientific research. By utilizing sparse data and advanced machine learning techniques, this approach can significantly improve predictive capabilities while minimizing computational costs.
Key Takeaways
- The proposed ROM retraining strategy achieves accuracy comparable to full retraining with significantly less computational time.
- Utilizes a Variational Autoencoder and transformer networks to effectively model dynamics and capture temporal dependencies.
- Supports real-time adaptation with minimal data, enhancing predictive capabilities in fluid dynamics applications.
- Emphasizes the importance of latent manifold distortions as a primary source of forecasting errors.
- The lightweight adaptation procedure allows for efficient updates in dynamic systems.
Computer Science > Machine Learning arXiv:2602.23188 (cs) [Submitted on 26 Feb 2026] Title:Efficient Real-Time Adaptation of ROMs for Unsteady Flows Using Data Assimilation Authors:Ismaël Zighed, Andrea Nóvoa, Luca Magri, Taraneh Sayadi View a PDF of the paper titled Efficient Real-Time Adaptation of ROMs for Unsteady Flows Using Data Assimilation, by Isma\"el Zighed and 3 other authors View PDF HTML (experimental) Abstract:We propose an efficient retraining strategy for a parameterized Reduced Order Model (ROM) that attains accuracy comparable to full retraining while requiring only a fraction of the computational time and relying solely on sparse observations of the full system. The architecture employs an encode-process-decode structure: a Variational Autoencoder (VAE) to perform dimensionality reduction, and a transformer network to evolve the latent states and model the dynamics. The ROM is parameterized by an external control variable, the Reynolds number in the Navier-Stokes setting, with the transformer exploiting attention mechanisms to capture both temporal dependencies and parameter effects. The probabilistic VAE enables stochastic sampling of trajectory ensembles, providing predictive means and uncertainty quantification through the first two moments. After initial training on a limited set of dynamical regimes, the model is adapted to out-of-sample parameter regions using only sparse data. Its probabilistic formulation naturally supports ensemble generation, w...