资源科学 ›› 2017, Vol. 39 ›› Issue (1): 116-124.doi: 10.18402/resci.2017.01.12

• 能源 • 上一篇    下一篇

风能数值预报的模式输出统计(MOS)研究

肖擎曜1, 2, 胡非1, 范绍佳3   

  1. 1. 中国科学院大气物理研究所,LAPC,北京 100029;
    2. 中国科学院大学,北京 100049;
    3. 中山大学大气科学院,广州 510275
  • 收稿日期:2016-09-10 修回日期:2016-12-05 出版日期:2017-01-20 发布日期:2017-01-20
  • 通讯作者: 胡非,E-mail:hufei@mail.iap.ac.cn
  • 作者简介:肖擎曜,男,广东广州人,博士生,主要从事大气湍流及风能应用研究。E-mail:xiaoqy@mail.iap.ac.cn
  • 基金资助:
    国家自然科学基金项目(11472272)

Model output statistics and wind power numerical prediction

XIAO Qingyao1, 2, HU Fei1, FAN Shaojia3   

  1. 1. LAPC,Institute of Atmospheric Physics,Chinese Academy of Sciences,Beijing 100029,China;
    2. University of Chinese Academy of Sciences,Beijing 100049,China;
    3. School of Atmospheric Sciences,Sun Yat-sen University,Guangzhou 510275,China
  • Received:2016-09-10 Revised:2016-12-05 Online:2017-01-20 Published:2017-01-20

摘要: 利用中尺度气象模式WRF输出的风速、温度、摩擦速度等变量作为待选因子,在其中选择合适的因子与风电场的实测风速建立多元线性模式输出统计(MOS)模型,对陕西省延长县风电场的风速进行预报。详细分析了因子数量、模型训练长度对预报结果的影响。研究表明:①所选因子不需要太多,推荐只选两个;②保证有600或以上时次的历史资料进行训练;③满足以上条件,MOS预报的均方根误差比单纯的模式预报误差平均可减少0.23 m/s。该方法实现简单,计算快捷,可实际应用于风电预报中。

关键词: WRF, 多元线性回归, 风电功率预报, 模式输出统计

Abstract: Instead of using the direct output from numerical weather prediction models,model post-processing using available observations are needed for meeting the rigorous requirements of wind speed prediction at wind farms. Here,a model output statistics (MOS)model is established using suitable factors within wind speed and temperature at different heights in the boundary layer,friction velocity,intensity of turbulence,Richardson number,air pressure and other meteorological elements from the mesoscale numerical weather prediction system WRF model output. Model fitting, selected factors and historical observation data in the same period with multi-linear regression formulas,and predicting wind speed at a wind farm in Shanxi,China was done. Effects on the prediction quality of a number of factors selected by the model and training length of the model are researched in detail. We found that generally there is no need for a large number of factors and the recommended number is two. Historical data with length of at least 600 are required for the quality of MOS prediction to be improved in evidence. Establishing the regressive formula,meeting the two conditions above,the root mean square error of MOS prediction would be about 0.23 m/s smaller compare to the WRF direct output. Considering the short calculation time and easy realization of this method it can be broadly applied to wind power prediction.

Key words: MOS, multi-regression, wind power prediction, WRF