Data-driven surrogate modeling for thermal-hydraulic codes via hybrid deep neural networks and quantile learning
Machine Learning: Science and Technology · 01.08.2026 · 7 (4) · 045017
DOI: 10.1088/2632-2153/ae7eef ↗Kurz gefasst
Fachartikel in der Zeitschrift Machine Learning: Science and Technology, online erschienen am 09.07.2026. Verfasst von Hyojun Yi und 2 weiteren Autor:innen. Die Arbeit nennt 38 Literaturangaben. Der Volltext ist frei zugänglich (Open Access, CC BY 4.0). Gefördert von: National Research Council of Science and Technology, Institute of Information & Communications Technology Planning & Evaluation.
Abstract (Originalsprache)
Nuclear energy is a clean, reliable power source, but realizing its potential requires strict safety measures in nuclear power plants. Thermal-hydraulic (TH) codes are used to simulate potential accident scenarios in probabilistic safety assessment (PSA). Their high computational cost becomes particularly problematic in dynamic PSA, which requires repeated simulations across large numbers of accident scenarios. Consequently, deep learning-based surrogate models (e.g. conditional autoencoders and recurrent neural networks) have been proposed to accelerate TH simulations. However, there remains substantial room for improvement in predictive accuracy, particularly in capturing abrupt fluctuations and complex temporal patterns. To address these limitations, we propose a novel surrogate model named QLCT that combines long short-term memory, one-dimensional convolutional neural networks, and transformer with multiple quantile regression, enhanced by temporal differential supervision. We compared QLCT against various deep learning models, and the results demonstrate that the proposed model achieves substantial improvements: 22.0%–40.0% reduction in mean absolute error, 12.4%–34.9% reduction in root mean squared error, and 27.9%–49.0% reduction in normalized mean absolute error. In addition, QLCT also shows superior performance in probabilistic forecasting compared to the state-of-the-art model, achieving improvements of 25.8%, 21.9%, and 15.2% in the winkler score, pinball loss, and coverage width-based criterion, respectively.
Förderung
- National Research Council of Science and Technology
- Institute of Information & Communications Technology Planning & Evaluation
Zitieren
APA
Yi, H., Kim, H., & Ryu, S. (2026). Data-driven surrogate modeling for thermal-hydraulic codes via hybrid deep neural networks and quantile learning. Machine Learning: Science and Technology, 7(4), 045017. https://doi.org/10.1088/2632-2153/ae7eef
Vancouver
Yi H, Kim H, Ryu S. Data-driven surrogate modeling for thermal-hydraulic codes via hybrid deep neural networks and quantile learning. Machine Learning: Science and Technology. 2026;7(4):045017. doi:10.1088/2632-2153/ae7eef
BibTeX
@article{Yi2026,
author = {Yi, Hyojun and Kim, Hyeonmin and Ryu, Seunghyoung},
title = {Data-driven surrogate modeling for thermal-hydraulic codes via hybrid deep neural networks and quantile learning},
journal = {Machine Learning: Science and Technology},
year = {2026},
volume = {7},
number = {4},
pages = {045017},
doi = {10.1088/2632-2153/ae7eef}
}
EUVIMED