唤醒
自回归模型
物理
傅里叶变换
标量(数学)
接头(建筑物)
快速傅里叶变换
算法
地形
傅里叶分析
代表(政治)
操作员(生物学)
替代模型
傅里叶级数
平面的
有界函数
涡轮机
声学
流离失所(心理学)
人工神经网络
窗口函数
计算机科学
网格
领域(数学)
数学分析
笛卡尔坐标系
控制理论(社会学)
均方误差
克里金
相似性(几何)
高超音速
计算流体力学
人工智能
作者
E.P. Lee,Leonardo P. Chamorro,Sang Lee
出处
期刊:Physics of Fluids
[American Institute of Physics]
日期:2026-08-01
卷期号:38 (8): 088133-
摘要
A reconstructive channel-UNet-assisted Fourier neural operator (ReCUFNO) is proposed for joint three-dimensional (3D) reconstruction and temporal prediction of wind-turbine wakes from limited two-dimensional (2D) observations. The model reconstructs and forecasts the three-dimensional scalar field of streamwise velocity, u, on a 3D grid from two orthogonal planar u-velocity slices conditioned on yaw angle. The approach combines a reconstructive channel-assisted layer with Fourier neural-operator forecasting and UNet-based spatial refinement to improve multi-scale wake representation and autoregressive stability. To address the mismatch between 2D inputs and 3D outputs, a reconstruction-consistent rollout maps each predicted volume back into planar inputs through slice extraction and roll-and-update feedback. Evaluated on yawed large-eddy simulation (LES) datasets, ReCUFNO achieves a single-step wind turbine error of approximately 5% and a single-step structural similarity index of 0.66, while maintaining bounded error growth over 30-step rollouts. Additional wake-focused diagnostics, including wake deflection, wake angle, centerline correlation, and rotor-disk power-deficit estimates, show improved agreement with LES relative to a UNet baseline. These results indicate that ReCUFNO provides a stable surrogate for observation-limited reconstruction and short-horizon prediction of yawed wind-turbine wake velocity fields.
科研通智能强力驱动
Strongly Powered by AbleSci AI