Scholarlink

<p>Accurate prediction of three-dimensional wind fields over complex mountainous terrain is essential for renewable energy deployment and regional weather modeling. Existing computational fluid dynamics simulations are accurate but expensive and require laborious mesh generation, while many deep learning methods struggle to capture sharp local wind variations induced by complex terrain. Here, we present a transformer-based dual-attention neural-operator framework for rapid wind-field prediction over complex mountainous terrain. Trained on a large simulation dataset spanning diverse terrain geometries and inflow conditions, the framework enables rapid prediction of steady-state wind field while maintaining competitive accuracy. It transfers robustly to real-world mountainous sites, reducing relative error by about 10% compared with existing neural-operator baselines. With sparse observations covering only 1% of the domain, prediction error is further reduced by 16.89% relative to the model without observations and by 32.75% relative to advanced baselines on unseen terrains. This framework provides a practical tool for wind-resource assessment over complex mountainous terrain and atmosphere-surface interaction studies.</p>

错误：搜索内容不能为空，请输入英文关键词
错误：关键词超出字数限制，请精简
高级检索

Transformer-based neural operators for 3D wind field prediction over complex mountainous terrain

  • 期刊论文

  • 出版模式：

    开放获取

  • 发表日期：

    2026年7月20日

查看全文
  • Yujia Zhang,
  • Jiaxi Qi,
  • Ruiyan Chen,
  • Yong Liu,
  • Yuzhou Zhang,
  • Lyulin Kuang,
  • Rita Zhang,
  • Shengze Cai

摘要

Accurate prediction of three-dimensional wind fields over complex mountainous terrain is essential for renewable energy deployment and regional weather modeling. Existing computational fluid dynamics simulations are accurate but expensive and require laborious mesh generation, while many deep learning methods struggle to capture sharp local wind variations induced by complex terrain. Here, we present a transformer-based dual-attention neural-operator framework for rapid wind-field prediction over complex mountainous terrain. Trained on a large simulation dataset spanning diverse terrain geometries and inflow conditions, the framework enables rapid prediction of steady-state wind field while maintaining competitive accuracy. It transfers robustly to real-world mountainous sites, reducing relative error by about 10% compared with existing neural-operator baselines. With sparse observations covering only 1% of the domain, prediction error is further reduced by 16.89% relative to the model without observations and by 32.75% relative to advanced baselines on unseen terrains. This framework provides a practical tool for wind-resource assessment over complex mountainous terrain and atmosphere-surface interaction studies.