| 随机森林算法在人工增雨效果统计检验中的应用研究 |
| |
| 引用本文： | 王伟健,姚展予,贾烁,赵文慧,谭超,张沛,高亮书,祝晓芸.随机森林算法在人工增雨效果统计检验中的应用研究[J].气象与环境科学,2018,41(2):113-119. |
| |
| 作者姓名： | 王伟健 姚展予 贾烁 赵文慧 谭超 张沛 高亮书 祝晓芸 |
| |
| 作者单位： | 中国气象科学研究院中国气象局云雾物理重点开放实验室 |
| |
| 摘 要： | 评估人工影响天气的效果通常需通过一定的方法近似求出未作业时的降水量,传统的人工影响天气效果的统计检验方法以线性回归为主,这种方法能够较好地建立目标区和对比区的历史回归方程。但实际上,降水量受多种因素影响,通过历史回归方法无法解决降水量非线性的问题。为解决降水估算中存在的非线性问题,提高降水量估算精度,引入了当下流行的机器学习算法——随机森林算法。使用优选出的多个与降水量相关的预报因子,分别构建了月、日降水估算模型,然后利用江西省南昌市1961-2010年降水数据及环流指数数据,并结合2012-2014年南昌市地面增雨作业信息,验证所建模型估算降水的可行性和人工增雨作业效果统计检验的适用性。结果表明,随机森林模型估算精度较高、泛化能力强,能较准确地估算月、日降水量,对于稳定性降水的估算准确率达90%;但随机森林模型对汛期的月降水量和强对流天气影响下的日降水量估算结果较观测值偏低,有待进一步订正和研究。总体来说,随机森林模型具有精度高、稳定性好、收敛快、不用进行数据预处理、参数少、操作便捷等特性,在综合性能上具有一定优势和可推广性。
|
| 关 键 词： | 人工增雨 统计检验 随机森林 降水估算 |
Application Research on Random Forest Algorithm in the Statistical Test of Rainfall Enhancement Effect |
| |
| Wang Weijian,Yao Zhanyu,Jia Shuo,Zhao Wenhui,Tan Chao,Zhang Pei,Gao Liangshu,Zhu Xiaoyun.Application Research on Random Forest Algorithm in the Statistical Test of Rainfall Enhancement Effect[J].Meteorological and Environmental Sciences,2018,41(2):113-119. |
| |
| Authors: | Wang Weijian Yao Zhanyu Jia Shuo Zhao Wenhui Tan Chao Zhang Pei Gao Liangshu Zhu Xiaoyun |
| |
| Abstract: | The evaluation of rainfall enhancement operation needs to quantify the precipitation approximately if under the unseeded condition using statistical methods. Usually linear regression is applied which can greatly establish the relationship between target and control region. But in reality, precipitation is influenced by a variety of factors; this method can not deal well with non linear problem. To resolve this issue and promote the prediction of precipitation, one of the popular machine learning algorithm called random forestalgorithm is introduced. Models for predicting monthly and daily precipitation are constructed, respectively, using several factors related to precipitation. Nanchang precipitation of 1961-2010 and meteorological circulation index, as well as the rainfall enhancement information during 20122014, are used to verify the applicability of rainfall enhancement effect statistical test and feasibility of estimating precipitationof the model. The results show that the random forest model has high precision and good generalization. The accuracy rate of predicting stable precipitation can reach to 90%. Monthly precipitation in flood season and daily precipitation under severe convection weather are underestimated by this model, however, which needs more research. Generally speaking, random forest modelhas the advantages of high precision, good stability, fast convergence, without data preprocessing, less parameters and convenient operation. It has greatly overall performance and popularity. |
| |
| Keywords: | rainfall enhancement statistical test random forest precipitation estimation |
| 本文献已被 CNKI 等数据库收录！ |
| 点击此处可从《气象与环境科学》浏览原始摘要信息 |
| 点击此处可从《气象与环境科学》下载免费的PDF全文 |
|
