孙全德1, 焦瑞莉1, 夏江江2, 严中伟2, 李昊辰3, 孙建华2, 王立志2, 梁钊明4
Adjusting wind speed prediction of the numerical weather forecast model based on machine learning methods
sunquande1, jiaoruili1, xiajiangjiang2, yanzhongwei2, lihaochen3, sunjianhua2, wanglizhi2, liangzhaoming4
(1.Beijing Information Science and Technology University;2.Chinese Academy of Sciences Institute of Atmospheric Physics;3.Peking University;4.State Key Laboratory of Severe Weather, Chinese Academy of Meteorological Sciences)
本文已被:浏览 390次   下载 256
投稿时间:2018-05-28    修订日期:2018-08-18
中文摘要: 对风速进行准确预测是精细化天气预报服务(如风能发电、冬奥会赛场条件保障等)的重要环节。本文基于三种机器学习算法(Lasso回归、随机森林和深度学习),对数值天气预报模式ECMWF预测的华北地区近地面10米风速进行订正。首先利用Lasso回归算法提取对10米风速有重要影响的气象要素特征集,作为三种机器学习算法的输入,建立相应模型对ECMWF预测的风速进行订正。用提取后的气象要素特征集建模有助于减少计算量和存储开销,并减小模型的复杂性,从而提高模型的泛化能力。将订正结果与传统订正方法模式输出统计(model output statistics,MOS)得到的订正结果进行对比。结果表明,三种机器学习算法的订正效果均好于MOS方法,显示了机器学习方法在改善局地精准气象预报方面的潜力。
Abstract:Accurate prediction of wind speed is crucial for local weather forecasting services (e.g., dealing with wind power industry and the Olympic winter game). Based on three machine learning algorithms (Lasso regression, random forest and deep learning), this paper demonstrates 3 models for adjusting the 10-meter wind speed in North China predicted by the numerical weather forecast model of ECMWF. Firstly, the Lasso regression algorithm is applied to identify the features which significantly affect the near-surface wind speed, among all the available meteorological elements. The extracted feature set is used as input for each machine learning algorithm to establish a model to adjust the ECMWF-predicted wind speed. Feature extraction helps to reduce the amount of computation, storage overhead and the complexity of the model, hence to facilitate the generalization of the model. The results of the three machine learning algorithms are compared with that of the traditional MOS method. All the three machine learning methods show a better performance in adjusting the wind speed than that of MOS, indicating great potential of the machine learning methods in improving local weather forecast.
文章编号:201805280237     中图分类号:    文献标志码:
孙全德,焦瑞莉,夏江江,严中伟,李昊辰,孙建华,王立志,梁钊明,0.[en_title][J].Meteor Mon,():-.
sunquande,jiaoruili,xiajiangjiang,yanzhongwei,lihaochen,sunjianhua,wanglizhi,liangzhaoming,0.Adjusting wind speed prediction of the numerical weather forecast model based on machine learning methods[J].Meteor Mon,():-.