基于随机森林的恩施地区降水预报订正方法研究  

Revision Method of Precipitation Forecast in Enshi Area Based on Random Forest

作　　者：卢松 高光平 张毅 纪洪岑 姚志武 尹章才[2] LU Song;GAO Guangping;ZHANG Yi;JI Hongcen;YAO Zhiwu;YIN Zhangcai(Guoneng Changyuan Enshi Hydropower Development Co.,Ltd.,Enshi 445000,Hubei,China;School of Resources and Environmental Engineering,Wuhan University of Technology,Wuhan 430070,Hubei,China;Changjiang Survey,Planning,Design and Research Co.,Ltd.,Wuhan 430010,Hubei,China)

机构地区：[1]国能长源恩施水电开发有限公司,湖北恩施445000 [2]武汉理工大学资源与环境工程学院,湖北武汉430070 [3]长江勘测规划设计研究有限责任公司,湖北武汉430010

出　　处：《水力发电》2024年第12期18-24,32,共8页Water Power

基　　金：国家自然科学基金资助项目(42171415)。

摘　　要：数值预报是目前中短期降水预报的主要方法,其输出的降水量往往存在偏差,因此,有必要对数值预报进行订正。恩施地区分布的水电站对降雨预报精度提出了更高要求,已有的订正方法具有区域针对性因而难以直接应用于恩施地区。为此,以恩施地区的板栗园雨量站为例,利用随机森林方法,建立数值预报模式输出的各种预报因子与实际降水量之间的非线性回归关系,形成数值预报产品的降雨订正模型,并分析其时间尺度效应,为恩施地区高精度订正模型选择提供了理论依据。应用2016年~2022年的7年历史数据,采用随机森林方法构建订正模型,并与支持向量机和长短时记忆网络等其他机器学习进行对比。实验结果表明:相较于数值预报的降雨预测,机器学习方法能显著提升降水量预报的准确性和可靠性;3种机器学习方法中,随机森林效果最好。Numerical forecasting is the main method for short-and medium-term precipitation forecast,and its output is often biased,so it is necessary to revise the numerical forecast.The wide spread of hydropower stations in Enshi area puts higher requirements for rainfall forecast accuracy,and the existing revision methods are region-specific and thus difficult to be directly applied to Enshi area.For this reason,this paper takes the Banli Yuan Rainfall Station in Enshi area as an example,and uses the random forest method to establish the nonlinear regression relationship between various forecast factor output from the numerical forecasting model and the actual precipitation,to form the rainfall revision model of the numerical forecasting products and analyze the time scale effect,which will provide the theoretical basis for the selection of the high-precision revision model for the Enshi area.Seven years of historical data from 2016-2022 are applied to construct the revised model using the random forest method,and compared with other machine learning such as support vector machine and long and short-term memory network.The experimental results show that the machine learning method can significantly improve the accuracy and reliability of precipitation forecasting compared with numerical forecasting for rainfall prediction,and among the three machine learning methods,the random forest has the best effect.

关 键 词：机器学习 降水预报 预报因子 偏差订正 随机森林 

分 类 号：TV331[水利工程—水工结构工程]