Hybrid Modelling for Digital Twins of Complex Systems: A Greenhouse Microclimate Prediction Use Case
Digital Twins and Applications · 2026-01 · 3 (1)
DOI: 10.1049/dgt2.70044 ↗Kurz gefasst
Fachartikel in der Zeitschrift Digital Twins and Applications, online erschienen am 17.07.2026. Verfasst von Joshua Setyawan und 2 weiteren Autor:innen. Die Arbeit wurde bisher 1-mal zitiert (Crossref) und nennt 39 Literaturangaben. Der Volltext ist frei zugänglich (Open Access, CC BY-NC-ND 4.0).
Abstract (Originalsprache)
Agricultural production systems, including greenhouse horticulture, are often unpredictable and operationally complex because they depend on uncertain factors such as weather patterns and diseases. To address these issues, growers adopt various technologies ranging from sensors and real‐time monitoring systems to AI‐based control algorithms. This accelerates the adoption of Digital Twins (DTs), a technology that allows users to monitor, analyse or simulate actual objects virtually. For a successful implementation, DTs require a model that satisfies two characteristics— intelligence and complexity . Unfortunately, the most commonly used modelling methods, data‐driven (DD) or physics‐based (PB), alone cannot fulfil both characteristics simultaneously. This paper proposes two novel hybrid models, Parameterised (Param) and Physics‐Informed Neural Network (PINN), which combine the advantages of both modelling methods to address the trade‐off issue. This study focuses on greenhouse microclimate prediction, where these hybrid models combine a PB greenhouse model named GreenLight with a DD model, in the form of trainable weights and/or a neural network. Furthermore, this paper simplified the resulting hybrid models to assess the trade‐offs between the intelligence and complexity characteristics. Benchmark results showed that of the two proposed hybrid models, the Param model had more accurate and consistent predictions, and even outperformed the base DD and PB models. It achieved lower RRMSE values of 2.45%, 2.85% and 18.05% for predicting greenhouse air temperature, vapour pressure and concentration than the PB model, which had 5.95%, 2.85% and 18.08%. Moreover, the simplified hybrid models showed no statistical difference with their original counterparts.
Zitieren
APA
Setyawan, J., AlSkaif, T., & Babur, Ö. (2026). Hybrid Modelling for Digital Twins of Complex Systems: A Greenhouse Microclimate Prediction Use Case. Digital Twins and Applications, 3(1). https://doi.org/10.1049/dgt2.70044
Vancouver
Setyawan J, AlSkaif T, Babur Ö. Hybrid Modelling for Digital Twins of Complex Systems: A Greenhouse Microclimate Prediction Use Case. Digital Twins and Applications. 2026;3(1). doi:10.1049/dgt2.70044
BibTeX
@article{Setyawan2026,
author = {Setyawan, Joshua and AlSkaif, Tarek and Babur, Önder},
title = {Hybrid Modelling for Digital Twins of Complex Systems: A Greenhouse Microclimate Prediction Use Case},
journal = {Digital Twins and Applications},
year = {2026},
volume = {3},
number = {1},
doi = {10.1049/dgt2.70044}
}
EUVIMED