人工智能在短临降水预报中应用研究综述  

Survey on the application of artificial intelligence in precipitation nowcasting

作　　者：方巍[1,2] 庞林 王楠[3] 易伟楠 FANG Wei;PANG Lin;WANG Nan;YI Weinan(School of Computer and Software,Nanjing University of Information Science and Technology,Nanjing 210044;Provincial Key Laboratory for Computer Information Processing Technology,Soochow University,Suzhou 215006;Shaanxi Meteorological Observatory,Xi'an 710014)

机构地区：[1]南京信息工程大学计算机与软件学院,南京210044 [2]苏州大学江苏省计算机信息处理技术重点实验室,苏州215006 [3]陕西省气象台,西安710014

出　　处：《南京信息工程大学学报（自然科学版）》2020年第4期406-420,共15页Journal of Nanjing University of Information Science & Technology（Natural Science Edition）

基　　金：教育部天诚汇智创新促教科研创新基金(2018A03038);苏州大学江苏省计算机信息处理技术重点实验室开放课题(KJS1935)。

摘　　要：短临降水预报是一项重要且具有挑战性的世界性难题.研究人员曾尝试使用各种技术预报降水,但是由于降水本身具有高度非线性、随机性和复杂性的特性,使得降水预测精确度并不高.近年来,随着人工智能技术的迅猛发展,其日渐渗透到人们生活的方方面面,气象领域也因此得益.人工神经网络能够对非线性系统进行建模,因此相比于传统方法,如数值天气预报法和光流法等,人工智能方法使得降水预报的准确率大大提高.本文介绍了传统降水预报的方法,着重总结概括了用于短临降水预报的各种最新人工智能方法,并对各研究方向进行归纳分析,为各类研究人员研究提供有益参考和借鉴.Precipitation nowcasting is an important and challenging worldwide problem. Various techniques have been used to predict precipitation,but the accuracy of precipitation nowcasting is not high due to the highly nonlinear,random,and complex nature of precipitation.In recent years,with the rapid development of artificial intelligence technology,it has gradually penetrated into all aspects of people’s lives,and the meteorological fields are no exception.Compared to traditional methods,such as numerical weather forecasting and optical flow methods,artificial neural networks can model non-linear systems,which makes the accuracy of precipitation nowcasting being greatly improved.In this article,we review the traditional methods and summarize the latest artificial intelligence methods used for short-term precipitation forecasting,and analyze the research directions to provide useful references for various types of researchers.

关 键 词：人工智能 短临降水预报 雷达回波图 神经网络 机器学习 

分 类 号：TP183[自动化与计算机技术—控制理论与控制工程] P456.1[自动化与计算机技术—控制科学与工程]