摘要
[目的]在“双碳”目标背景下,海上风能是实现能源结构转型的重要支撑,为解决海上风速观测数据稀缺、分辨率有限的问题,[方法]提出一种融合合成孔径雷达(SAR)卫星观测与ERA5再分析数据的深度学习方法,构建基于卷积神经网络(CNN)、长短期记忆网络(LSTM)与Transformer网络架构的高分辨率风速重构模型。[结果]该模型整合了中国近海16个实测站点、12383景SAR影像及ERA5再分析数据,在海拔10 m与100 m高度风速预测中均取得优异表现,平均偏差分别为+0.19 m/s和+0.02 m/s,皮尔逊相关系数分别达到0.81和0.88,明显优于传统再分析数据。模型输出风速在威布尔分布形态上与实测结果高度一致,具备良好的统计一致性与泛化能力。在风向预测方面,模型同样展现出较高拟合度和稳定性。[结论]研究成果可为海上风能资源评估、风电场选址及规划提供高精度、高时空分辨率的数据支撑,具有广阔的工程应用前景和推广价值。
[Purpose]Under the"dual carbon"goals,offshore wind energy is recognized as a crucial support for achieving energy structure transformation.To address the scarcity and limited resolution of offshore wind speed observation data,[Method]a deep learning approach integrating SAR satellite observations and ERA5 reanalysis data is proposed.A high-resolution wind speed reconstruction model based on CNN,LSTM,and Transformer network architectures is developed.[Result]The model incorporates 16 offshore measurement stations in China,12383 SAR images,and ERA5 reanalysis data.Excellent performance is achieved in wind speed predictions at both 10 m and 100 m heights,with mean biases of+0.190 m/s and+0.02 m/s,and correlation coefficients of 0.81 and 0.88,respectively,significantly outperforming traditional reanalysis data.The reconstructed wind speed distribution highly aligns with measurements in Weibull distribution form,demonstrating strong statistical consistency and generalization capability.The model also exhibits high fitting accuracy and stability in wind direction prediction.[Conclusion]The research outcomes can provide high-precision,spatiotemporally continuous data support for offshore wind resource assessment,wind farm site selection,and planning,showing broad prospects for engineering application and promotion.
作者
龙海川
袁令
买小平
燕志婷
郝辰妍
LONG Haichuan;YUAN Ling;MAI Xiaoping;YAN Zhiting;HAO Chenyan(CSSC Windpower Development Co.,Ltd.,Beijing 100095,China)
出处
《船舶工程》
北大核心
2025年第S2期20-27,33,共9页
Ship Engineering
关键词
海上风能
SAR卫星
深度学习
风速重构
offshore wind energy
synthetic aperture radar(SAR)satellite
deep learning
wind speed reconstruction
