摘要
利用随机森林算法,基于历史地面实况观测数据,构建随机森林1~6 h风场预报模型,并用2018年的地面实况观测数据对预报模型进行检验分析.结果表明,随机森林算法在风场预报中有较好的泛化能力,对地面10 m风场有较好的预报水平,在1~6 h的预报中,预报风场与实况风场比较接近,各预报时效风速的年平均绝对误差为1.0 m/s,风向的年平均绝对误差为59°;随着预报时效的延长,随机森林风场预报模型对较大风速和较大风速时风向的预报能力逐渐减弱;利用A_Int检验方法对随机森林1~6 h风场预报结果进行检验,结果表明模型预报的准确率普遍较高,各预报时效2018年平均准确率为61.5%.
A random forest was used to build a 1-6 h wind field forecast model based on historical live weather data.The test results showed that the random forest algorithm had a good generalization ability in wind field forecasting and a good forecasting ability for the wind field,in the 1-6 h forecast,when the forecast wind field and the live wind field was relatively close;the wind speed annual average absolute error of each forecasted time was 1.0 m/s,and the annual average absolute error of the wind direction was 59°.With the increase in the forecasted time,the ability of the forecasting model gradually weakened,as to be able to predict the wind direction at larger wind speeds and larger wind speeds.The A_Int test method was used to test the 1-6 h wind field forecast results of the random forest.The results showed that the accuracy rate of the model forecast was generally high,and the average accuracy rate in 2018 was 61.5%.
作者
付旭东
王金艳
李龙燕
陈金车
苏士翔
常伟
王明
FU Xu-dong;WANG Jin-yan;LI Long-yan;CHEN Jin-che;SU Shi-xiang;CHANG Wei;WANG Ming(Key Laboratory of Arid Climatic Changes and Disaster Reduction of Gansu Province,College of Atmospheric Sciences,Lanzhou University,Lanzhou 730000,China;Troops 95994 of the People's Liberation Army of China,Jiuquan 735000,Gansu,China)
出处
《兰州大学学报（自然科学版）》
CAS
CSCD
北大核心
2021年第4期503-509,共7页
Journal of Lanzhou University(Natural Sciences)
基金
国家重点研发计划项目(2020YFA0608402)
国家自然科学基金面上项目(41575138)
关键词
机器学习
随机森林
风场
预报
machine learning
random forest
wind field
forecast
作者简介
通信联系人:王金艳(1977-),女,新疆沙湾人,副教授,博士,硕士研究生导师,e-mail:wangjny@lzu.edu.cn,研究方向为灾害性天气及其预报
