摘要
以如东海上风电场升压站激光雷达测风资料为基础,提出了一种强风事件识别方法,设计并比较了三种预报强风事件识别方案。基于决策树和一元线性回归方法,分别开展了针对强风事件的订正方法研究。结果发现:三种预报强风事件识别方案中,等分位阈值方案明显更优,事件命中率达到76.1%,匹配时长命中率达到87.6%;采用消偏阈值方案和等分位阈值方案预报的强风事件时长会更接近观测强风事件时长;等分位阈值方案识别的事件基本可以覆盖到各次观测强风事件的全程;两个订正模型相对于模式预报都有一定提升与改进,其中决策树比一元线性回归模型更优,其平均绝对误差、相对误差和均方根误差明显更小。
Based on the wind speed observation data of Marine Booster Station in Rudong Wind Farm,this paper proposes a method of identifying gale event.Three identification schemes of gale event forecast are developed and compared through the determination of the crucial parameters.Then,based on the decision tree method and a single linear regression method,the correction methods for gale events are studied.The results show that the gale event forecast of equal cumulative frequency scheme is superior to other schemes,having the hit rate of 76.1%and the hit rate of matching duration of 87.6%.The duration of gale event forecast of eliminating deviation and equal cumulative frequency schemes are more in agreement with the observation data.Besides,the equal cumulative frequency scheme can cover the duration of every observation gale event,so it is good for proposing the beginning and end time of gale warning.The above-mentioned two correction methods can improve the forecast performance to a certain extent.However,the improvement done by the decision tree method is more obvious,for it can significantly reduce the MAE,RE,RMSE.
作者
韩乐琼
何晓凤
张雪松
肖擎曜
陈笑
HAN Leqiong;HE Xiaofeng;ZHANG Xuesong;XIAO Qingyao;CHEN Xiao(Beijing Jiutian Meteorology Science and Technology Co.,Ltd,Beijing 100081;Huafeng Meteorological Media Group Co.,Ltd.,Beijing 100081;China Guangdong Nuclear Wind Power Co.,Ltd.,Beijing 100070)
出处
《气象》
CSCD
北大核心
2023年第12期1542-1552,共11页
Meteorological Monthly
基金
国家重点研发计划(2018YFC1507804)
中广核尖峰计划项目(001-GN-A-2021-SN-0239)共同资助。
关键词
强风事件
数值预报
订正
决策树
一元线性回归
gale event
numerical weather prediction(NWP)
correction
decision tree
a single linear regression
作者简介
第一作者:韩乐琼,主要从事数值产品的订正应用.E-mail:641822468@qq.com;通讯作者:何晓凤,主要从事数值预报解释应用及专业气象服务.E-mail:hexf@cma.gov.cn。
