基于U-Net的海上风场模式数据订正

Offshore Wind Field Model Data Revision Based on U-Net

我国沿海地区生产生活受海上风场的影响较大,目前海上高分辨网格化数据为模式数据,其主要通过物理数值模型推算得出,对于初值敏感,且无法准确描述近地面和海面的风速信息。需要考虑用浮标、船舶等实况参考数据订正,而传统订正方法只能订正参考数据附近的格点值。本文提出一种基于U-Net网络的风场数据订正算法。利用U-Net网络通过站点参考数据对整个目标区域进行订正,降低目标海洋区域订正后风速误差。实验结果表明海洋目标区域10 m风速平均订正误差RMSE值为1.96 m/s,比模式数据的平均误差降低了16.8%,U-Net网络模型可以有效订正模式数据,为后续数据融合等研究提供高质量数据支持。

The production and daily life in coastal areas of our country are greatly affected by offshore wind fields.Currently,high-resolution gridded marine data are modeled data,primarily derived from physical numerical models,while the modeled data are sensitive to initial conditions and cannot accurately describe wind speed information near the ground and sea surface.It is necessary to consider using real-time reference data from buoys,ships,and other sources for correction.However,traditional correction methods can only correct grid values near the reference data points.Therefore,in this study,a data correction algorithm based on U-Net was proposed for wind speed correction.It uses site reference data to correct the entire target marine area based on U-Net,reducing wind speed errors after correction.Experimental results show that the Root Mean Square Error(RMSE)of 10 m wind speed in the marine target area is 1.96 m/s.This is a 16.8%reduction compared to the average error of model data.The U-Net model can effectively correct model data,providing high-quality data support for subsequent data fusion studies.