基于机器学习的来水预报多模型对比应用  

Multi-model Comparative Application of Inflow Forecasting Based on Machine Learning

作　　者：陶春华 时焱红 王甫志 TAO Chunhua;SHI Yanhong;WANG Fuzhi(CHN Energy Dadu River Big Data Service Co.,Ltd.,Chengdu 610041,Sichuan,China;Sichuan Xin Neng Project Consulting Co.,Ltd.,Chengdu 610000,Sichuan,China)

机构地区：[1]国能大渡河大数据服务有限公司,四川成都610041 [2]四川新能工程咨询有限公司,四川成都610000

出　　处：《水力发电》2021年第11期15-19,共5页Water Power

摘　　要：选用俄日河流域玉科、二楷水文站构建断面区间,对区间实测径流过程分析,确定出影响二楷水文站断面流量过程的影响因子,采用基于机器学习线性回归、BP神经网络、K邻近等经典算法,分别构建二楷水文站断面的来水预报模型,并与已建的时变线性汇流模型预报结果及断面实测过程进行对比。对比结果表明,基于机器学习的线性回归和K邻近算法模型预报效果均优于时变线性汇流模型,其中基于机器学习的线性模型整体预测精度最高。The Yuke and Erkai hydrological stations in the Eri River Basin are selected to construct the sectional interval,and based on the analyzing of runoff process in the interval,the impact factors affecting the cross-sectional flow of Erkai Hydrological Station are determined.Then the classical machine learning algorithms such as linear regression,BP neural network and K proximity are adopted to construct runoff forecasting model of Erkai Hydrological Station,and the forecasting results are compared with the result of established time-varying liner confluence model and the data of actual measured process.The comparison shows that the forecasting results of linear regression and K proximity are better than the time-varying liner confluence model.The overall forecasting accuracy of the linear model based on machine learning is the highest.

关 键 词：来水预报 机器学习 时变线性汇流模型 线性回归 BP神经网络 K邻近 俄日河流域 

分 类 号：TV122[水利工程—水文学及水资源]