Answers are grounded in this paper and cite their sources.
Unsupervised domain decomposition improved accuracy in modeling atmospheric boundary layer flows, highlighting clustering's role.
Reduced-order models emerge as a solution to efficiently predict atmospheric boundary layer flows. However, global reduced-order models, applying dimensionality reduction to the whole domain, struggle with accuracy in domains disturbed by buildings. To address this limitation, this study proposed an unsupervised domain-decomposition approach for reduced-order models using clustering to divide the domain into building-influenced and uninfluenced subdomains. Clustering enables the localized dimensionality reduction via principal component analysis and predictive mapping using Gaussian process regression in each subdomain. Two clustering methods, K-means and vector quantization principal component analysis, were used as the comparison to individually develop domain-decomposition reduced-order models. A steady atmospheric boundary layer flow over a building array across 64 operating conditions served as the test case. The development of the reduced-order models was based on computational fluid dynamics simulations using k-ω shear stress transport model. Compared to the global reduced-order model, the clustering-based domain decomposition reduced-order models reduced root mean square error by 9.5% for the stream-wise velocity field and 18.2% for the turbulent kinetic energy field and achieved a speed-up of approximately 105 times over full-order computational fluid dynamics simulations.
No takes yet. Share an insight, caveat, or question.
Li et al. (2025) studied this question.
Synapse has enriched 10 related papers. Consider them for comparative context: