Super-resolution reconstruction of simulated stochastic wind fields using ensemble conditional diffusion model

• Published in: Journal of wind engineering and industrial aerodynamics Vol. 267; p. 106249
• Main Authors: Xu, Zidong, Wang, Hao, Zhao, Kaiyong, Zhou, Rui, Lin, Yuxuan
• Format: Journal Article
• Language: English
• Published: Elsevier Ltd 01.12.2025
• Subjects: Deep learning; Denoising diffusion probabilistic model; Stochastic wind fields; Super-resolution reconstruction; Wind energy; Deep learning; Super-resolution reconstruction; Denoising diffusion probabilistic model; Wind energy; Stochastic wind fields
• ISSN: 0167-6105
• DOI: 10.1016/j.jweia.2025.106249

Conducting wind field super-resolution (SR) reconstruction using limited dataset is crucial for analyzing wind effects on wind energy equipment and optimizing wind energy utilization. Currently, most SR reconstruction methods are primarily applied to wind data (e.g., field measurement, CFD simulation) that contain complete turbulent physical structures, which facilitate the smooth execution of reconstruction. However, in engineering practice, multivariate stochastic processes are commonly simulated and regarded as the stochastic wind fields, which lack of fundamental fluid dynamic laws, making reconstruction more challenging. To this end, the ensemble conditional Denoising Diffusion Probabilistic Model (DDPM) is firstly proposed. Unlike classic DDPM, which directly use the low-resolution image as the conditional input, the ensemble model generates the input condition through the combination of the user-defined CNN and the transformer module. The effectiveness and accuracy of the ensemble model are validated through numerical experiment. The reconstruction results obtained by classic DDPM are also investigated for comparison purpose. Results show that compared to the classic DDPM, the reconstruction results based on the ensemble model demonstrate better alignment with target values in terms of wind speed time histories, turbulent spectral characteristics, similarity metrics, and wind power density. •Ensemble conditional diffusion model is utilized to conduct super-resolution reconstruction of stochastic wind fields.•A general framework for super-resolution reconstruction using the proposed DDPM is established and verified.•Transformer and custom-designed CNN modules are utilized to extract features from wind fields as a novel input condition.•The NUFFT-aided WSRM is utilized to generate 2-D turbulent wind field.•The ensemble DDPM outperforms the classic model in spatiotemporal wind field SR reconstruction.