Major research outcome
- M.S. Student Eunchan Lee and Professor Sang Lee introduce a new machine learning model for high-resolution far-wake prediction in wind turbines
- 관리자 |
- 2025-09-15 11:13:46|
- 377
- 2025-09-15 11:13:46|
For the effective operation of wind farms, it is essential to predict-both with high accuracy and at real-time speeds-the geometry and evolution of the far wake that drives turbine-turbine interactions. Yaw-angle control can increase total power output. However, high-fidelity numerical approaches such as large-eddy simulation (LES) demand massive computational resources and take considerable time to scale to farm-level analyses. In this context, there has been a need for a data-driven predictive framework that achieves both high accuracy and spatial high resolution while enabling low-cost real-time inference.
M.S. student Eunchan Lee and Professor Sang Lee proposed a new machine-learning model, the Channel-assisted Fourier Neural Operator (CFNO), and demonstrated that it can predict the far wake of a yawed wind turbine with high resolution over long horizons at low cost. CFNO retains the strengths of the Fourier Neural Operator (FNO), which performs global operations in the frequency domain, while explicitly incorporating yaw conditions into the model for the wind turbine wake prediction problem. The method introduces a new deep learning architecture that enhances training and inference stability so that the wake's asymmetric velocity deficit and vorticity structures do not collapse over long-horizon forecasts. As a result, compared with vanilla FNO, CFNO consistently reproduces wake deflection and cross-sectional distortion even in the far-wake region, and it dramatically reduces inference cost relative to high-fidelity simulations-substantially improving the practical feasibility of real-time operation and control scenarios.
This study has strong industrial applicability, including quantitative assessment of wake-steering strategies, mitigation of inter-turbine interference, wind-farm layout and operational optimization, and reduction of power-forecast uncertainty. It is also expected to contribute to advancing digital-twin-based wind farm operation technologies. The work was published in Physics of Fluids in September 2025 and was selected as an Editor's Pick. Physics of Fluids is the top-tier journal in fluid mechanics (within the top 6% in JCR).
https://doi.org/10.1063/5.0287914
M.S. student Eunchan Lee and Professor Sang Lee proposed a new machine-learning model, the Channel-assisted Fourier Neural Operator (CFNO), and demonstrated that it can predict the far wake of a yawed wind turbine with high resolution over long horizons at low cost. CFNO retains the strengths of the Fourier Neural Operator (FNO), which performs global operations in the frequency domain, while explicitly incorporating yaw conditions into the model for the wind turbine wake prediction problem. The method introduces a new deep learning architecture that enhances training and inference stability so that the wake's asymmetric velocity deficit and vorticity structures do not collapse over long-horizon forecasts. As a result, compared with vanilla FNO, CFNO consistently reproduces wake deflection and cross-sectional distortion even in the far-wake region, and it dramatically reduces inference cost relative to high-fidelity simulations-substantially improving the practical feasibility of real-time operation and control scenarios.
This study has strong industrial applicability, including quantitative assessment of wake-steering strategies, mitigation of inter-turbine interference, wind-farm layout and operational optimization, and reduction of power-forecast uncertainty. It is also expected to contribute to advancing digital-twin-based wind farm operation technologies. The work was published in Physics of Fluids in September 2025 and was selected as an Editor's Pick. Physics of Fluids is the top-tier journal in fluid mechanics (within the top 6% in JCR).
https://doi.org/10.1063/5.0287914
| Attach File |
|---|