Robust and expert-agnostic digital twin calibration via ensemble learning and Bayesian optimization
Name
3736425.3770105.pdf
Size
833.6 KB
Format
Adobe PDF
Checksum (MD5)
7c4f6123235a0fe2dd7dbed360bf7c0a
Author(s)
Zhan, Sicheng
Cui, Bosen
Date Issued
November 11, 2025
Publisher
ACM|The 12th ACM International Conference on Systems for Energy-Efficient Buildings, Cities, and Transportation
Citation
Sicheng Zhan and Bosen Cui. 2025. Robust and expert-agnostic digital twin calibration via ensemble learning and Bayesian optimization. In Proceedings of the 12th ACM International Conference on Systems for Energy-Efficient Buildings, Cities, and Transportation (BuildSys '25). Association for Computing Machinery, New York, NY, USA, 248–251.
Version
Final published version
Abstract
Digital twins have emerged as a critical tool in tackling climate change. Considering the data scarcity of complex systems, a promising approach to developing digital twins involves combining physics-based models with data assimilation. However, model calibration remains challenging due to uncertainties in both the physical models and observational data, and the reliance on domain knowledge. In this study, we develop an ensemble learning-based approach that aggregates sub-models with diversified calibration configurations. The proposed method streamlines calibration without expert-driven parameter screening and improves the digital twin's extrapolation capability, enabling more robust predictive applications. We demonstrate the effectiveness of our approach by calibrating the energy model of an office building, significantly reducing the extrapolation error and the associated risks. To the best of our knowledge, this is the first study to facilitate the calibration of physics-based models using ensemble learning, especially in the parameter space.
Description
BUILDSYS ’25, Golden, CO, USA
MIT Department
Massachusetts Institute of Technology. Department of Architecture
Terms of Use
Creative Commons Attribution
Persistent DSpace Link
DOI of Published Version
https://doi.org/10.1145/3736425.3770105
