机器学习技术在曹娥江流域洪水预报中的应用  

Application of Machine Learning Technology in Flood Forecast of Cao’e River Basin

作　　者：孙小洪 赵兵 孙逸群[3] 石朋[3] SUN Xiaohong;ZHAO Bing;SUN Yiqun;SHI Peng(Yudong Sluice Operation Management Center of Shangyu District,Shaoxing 312300,Zhejiang,China;Administration Bureau of Nanwan Reservoir in Xinyang,Xinyang 464031,Henan,China;College of Hydrology and Water Resource,Hohai University,Nanjing 210098,Jiangsu,China)

机构地区：[1]上虞区虞东水闸运行管理中心,浙江绍兴312300 [2]信阳市南湾水库管理局,河南信阳464031 [3]河海大学水文水资源学院,江苏南京210098

出　　处：《浙江水利科技》2022年第2期83-87,共5页Zhejiang Hydrotechnics

摘　　要：以逐步线性回归、正则化技术、L曲线方法等机器学习技术为理论基础,提出基于机器学习方法的水位预报模型,在此基础上建立曹娥江中下游流域的水位预报方案。水位预报模型使用逐步回归方法以使用最少的预测变量数来实现最大化预测的能力,引入正则化方法及L曲线方法在应对复共线性问题的同时引入先验信息。研究实现多模型方案的对比,分析历年场次洪水特征和重点水利工程对洪水预报精度的影响,提出基于机器学习方法的水位预报模型,在台风“烟花”中进行试运行,结果表明技术可行,方案模型准确。并应用于曹娥江中下游流域,预报结果精准。In this study,a stage forecast model is proposed based on several machine learning methods,including the stepwise linear regression,the regularization method,and the L-curve method.Using the proposed stage forecast model,we establish a stage forecast system for the midstream and downstream of the Cao’e River.The stepwise regression method is utilized to maximize the forecasting ability with limited variables.The regularization method and the L-curve method are used to handle the collinearity and absorb the prior information.By comparing schemes of multiple models,analyzing the flood characteristics over the years,and the influence of major water projects on the accuracy of flood forecasting,the stage forecast model is proposed based on the machine learning methods and applied to forecast the“#1526 Super Typhoon In-fa”.The results show that the stage forecast model is suitable for stage forecasting of the midstream and downstream of the Cao’e River.

关 键 词：机器学习 水位预报 逐步回归 正则化技术 L曲线方法 曹娥江 

分 类 号：P338.9[天文地球—水文科学]