Academic Journal
A space mapping-based high-dimensional model representation for multi-fidelity surrogate modelling.
| Τίτλος: | A space mapping-based high-dimensional model representation for multi-fidelity surrogate modelling. |
|---|---|
| Συγγραφείς: | Zhong, Linjun1 (AUTHOR), Yang, Yang2 (AUTHOR) yangyang@mail.hzau.edu.cn, Zha, Zhijian3 (AUTHOR), Shu, Leshi1 (AUTHOR), Jiang, Ping1 (AUTHOR), Zhou, Qi3 (AUTHOR), Wang, Chu3 (AUTHOR) |
| Πηγή: | Engineering Optimization. Jun2026, Vol. 58 Issue 6, p1652-1678. 27p. |
| Θεματικοί όροι: | *High-dimensional model representation, *Metamaterials, *Reduced-order models, *Dimensional reduction algorithms, *Acoustic wave propagation |
| Περίληψη: | Surrogate modelling is extensively used for approximating expensive black-box problems. The effort grows dramatically with increasing design dimensionality, known as the 'curse of dimensionality'. Dimensionality reduction-based and auxiliary information-incorporated approaches have been studied. However, they are limited by sparse expensive samples available for sample collections and fusion patterns, and gaps remain in efficient high-dimensional model representation (HDMR) for multi-fidelity data. Therefore, a space mapping-based multi-fidelity modelling approach using the HDMR framework, termed SM-HDMR, is proposed. First, a Cut-HDMR model is constructed with low-fidelity data to capture the overall trend. Secondly, a multi-fidelity model is built by mapping low-fidelity predictions to the high-fidelity output space with the low-fidelity model as prior knowledge and high-fidelity samples distributed randomly. Six 10–50-dimensional numerical functions demonstrate the effectiveness and advantages of SM-HDMR. It is applied to a 50-dimensional acoustic field prediction of metamaterials involving expensive simulations, demonstrating its capabilities in supporting complex engineering designs. [ABSTRACT FROM AUTHOR] |
| Βάση Δεδομένων: | Academic Search Index |
καταχωρήστε σχόλιο πρώτοι!