View in EDS

Federated learning for enhancing extrapolation ability of HVAC models: case study on two real-life DOAS units.

Saved in:
Bibliographic Details
Title: Federated learning for enhancing extrapolation ability of HVAC models: case study on two real-life DOAS units.
Authors: Cho, Seongkwon1 (AUTHOR) gyflsnu@snu.ac.kr, Park, Cheol Soo2 (AUTHOR)
Source: Journal of Building Performance Simulation. Jun2026, Vol. 19 Issue 4, p622-637. 16p.
Subjects: Federated learning, Extrapolation, Independent variables, Artificial neural networks, Ventilation, Statistical models
Abstract: Conventional data-driven modeling often fails under unseen conditions. This study introduces bidirectional knowledge sharing via federated learning (FL) across multiple systems without data exchange. The method was tested on two dedicated outdoor air system (DOAS) units with imbalanced datasets. FL improved sensible heat prediction accuracy, reducing CVRMSE from 39.1 to 17.7% for DOAS 1 and 13.9 to 9.8% for DOAS 2 compared to a conventional ANN. Under unseen control conditions, FL further reduced CVRMSE from 33.1 to 15.1% for DOAS 1 and 21.3 to 14.6% for DOAS 2, demonstrating enhanced extrapolation. Intervention-based analyses also showed that FL promotes more physically consistent relationships between control variables and system responses. Although limited to a short-term case study of two similar systems, the results highlight FL's potential to improve the robustness of data-driven HVAC models. [ABSTRACT FROM AUTHOR]
Copyright of Journal of Building Performance Simulation is the property of Taylor & Francis Ltd and its content may not be copied or emailed to multiple sites without the copyright holder's express written permission. Additionally, content may not be used with any artificial intelligence tools or machine learning technologies. However, users may print, download, or email articles for individual use. This abstract may be abridged. No warranty is given about the accuracy of the copy. Users should refer to the original published version of the material for the full abstract. (Copyright applies to all Abstracts.)
Database: Engineering Source
Description
Abstract:Conventional data-driven modeling often fails under unseen conditions. This study introduces bidirectional knowledge sharing via federated learning (FL) across multiple systems without data exchange. The method was tested on two dedicated outdoor air system (DOAS) units with imbalanced datasets. FL improved sensible heat prediction accuracy, reducing CVRMSE from 39.1 to 17.7% for DOAS 1 and 13.9 to 9.8% for DOAS 2 compared to a conventional ANN. Under unseen control conditions, FL further reduced CVRMSE from 33.1 to 15.1% for DOAS 1 and 21.3 to 14.6% for DOAS 2, demonstrating enhanced extrapolation. Intervention-based analyses also showed that FL promotes more physically consistent relationships between control variables and system responses. Although limited to a short-term case study of two similar systems, the results highlight FL's potential to improve the robustness of data-driven HVAC models. [ABSTRACT FROM AUTHOR]
ISSN:19401493
DOI:10.1080/19401493.2026.2637735