一种风矢量分解和RobustSTL-TimesNet-BiGRU的复杂地形风向预测  

WIND VECTOR DECOMPOSITION AND ROBUSTSTL-TIMESNET-BIGRU WIND DIRECTION FORECASTING IN COMPLEX TERRAIN

作　　者：刘洋[1] 王聪 张宏立 马萍 李新凯 Liu Yang;Wang Cong;Zhang Hongli;Ma Ping;Li Xinkai(School of Electrical Engineering,Xinjiang University,Urumqi 830047,China;School of Intelligence Science and Technology,Xinjiang University,Urumqi 830047,China)

机构地区：[1]新疆大学电气工程学院,乌鲁木齐830047 [2]智能科学与技术学院,新疆大学工程训练中心,乌鲁木齐830047

出　　处：《太阳能学报》2025年第3期576-588,共13页Acta Energiae Solaris Sinica

基　　金：国家重点研发计划(2021YFB1507000);新疆维吾尔自治区自然科学基金(2022D01E33,2022D01C367);国家自然科学基金(52267010);自治区重点研发计划(2022B03031)。

摘　　要：针对复杂地形下的风向预测场景,提出一种风矢量正交分解、鲁棒性局部加权回归下的周期趋势分解(RobustSTL)方法、TimesNet模型和融合双向门控循环单元网络(BiGRU)误差补偿的多步风向预测方法。首先,为了减少原始风向循环圆周特性带来的大幅度波动性,将风向与相关性强的风速利用风矢量正交分解方法转化为波动性较小的矢量风速,并利用RobustSTL将矢量风速分解为趋势项、季节项和剩余波动项。其次,将分解后的各项分别训练TimesNet网络并得到各项的初步预测结果,对各项进行求和并重构为初始预测风向。然后,为了进一步挖掘初步风向预测误差的深层特征,提高风向的预测精度,采用BiGRU对初步预测误差进行建模与训练。最后,将预测的误差与初步预测风向加和,得到最终的风向预测结果。采用实际复杂地形风电场数据进行验证分析,结果表明所提的多步风向预测混合模型具有较高的预测精度。Aiming at wind direction forecasting in complex terrain,this paper proposes a multi-step wind direction forecasting method based on wind vector orthogonal decomposition,Robust Seasonal-Trend decomposition using Loess(RobustSTL),the TimesNet model,and a bidirectional gated recurrent unit network(BiGRU)for error compensation.Firstly,to reduce the significant fluctuation caused by the circular characteristics of the original wind direction,the wind direction and the correlated wind speed are converted into a less fluctuating vector wind speed using the wind vector orthogonal decomposition method.This vector wind speed is then decomposed into trend,seasonal,and residual components using RobustSTL.Secondly,each component after decomposition is trained using the TimesNet model,and preliminary forecasting results are obtained.The components are then summed and reconstructed to form the initial wind direction forecast.To further extract deep features from the initial forecasting errors and enhance accuracy,the BiGRU network is used for modeling and training the forecasting errors.Finally,the predicted error is combined with the initial forecast to obtain the final wind direction forecast.The proposed multi-step wind direction forecasting hybrid model is validated using real data from a complex terrain wind farm,and the results demonstrate its high forecasting accuracy.

关 键 词：风电场 预测 误差补偿 TimesNet模型 风向 

分 类 号：TM614[电气工程—电力系统及自动化]