基于机器学习的水文系列插补延长模型研究  

Research on Extension Model of Hydrological Interpolation Based on Machine Learning

作　　者：翁茂峰 刘莹莹[1] 寇思飞 梁曦[2] 刘蕊蕊 Weng Maofeng;Liu Yingying;Kou Sifei;Liang Xi;Liu Ruirui(PowerChina Northwest Engineering Corporation Limited, Xi'an 710065,China;Shaanxi Hydropower Survey and Design Research Institute, Xi'an 710001,China)

机构地区：[1]中国电建集团西北勘测设计研究院有限公司,西安710065 [2]陕西省水利电力勘测设计研究院,西安710001

出　　处：《西北水电》2021年第2期15-20,共6页Northwest Hydropower

摘　　要：传统的水文插补延长模型预测效果一般且计算较为繁琐,难以满足重要水文资料的插补延长。文章基于机器学习模型,依据安塞水文站和枣园水文站多年水文资料,利用水文数据间的非线性关系建立水文插补延长模型。结果表明,传统的水文插补延长线性模型具有一定的效果,但仍需进一步提升;机器学习模型能搜寻数据间的非线性关系,不同的模型效果具有一定的差异,其中支持向量机(SVR)最优,BP神经网络模型次之,XGBoot模型最差;SVR模型结合网格寻优可以得到较为有效的水文插补延长模型,可为水文插补延长提供参考。The traditional hydrological interpolation extension model has general prediction effect and cumbersome calculations,and it is difficult to meet the interpolation extension requirement of important hydrological data.Based on machine learning model,the author establishes a hydrological interpolation extension model with the nonlinear relationship between hydrological data on basis of the long-term hydrological data of Ansai Hydrological Station and Zaoyuan Hydrological Station.The results show that the traditional hydrological interpolation extended linear model has a certain effect,but it still needs further improvement;although the machine learning model can search for nonlinear relationships between data,the effects of different models have certain differences.Among them,SVR is the optimal;the BP neural network model is the second,and the XGBoot model is the least effective.The SVR model combined with grid optimization can get a more effective hydrological interpolation extension model,which provides a reference for hydrological interpolation extension.

关 键 词：水文系列 插补延长 机器学习模型 安塞水文站 枣园水文站 

分 类 号：TU443[建筑科学—岩土工程]