基于偏最小二乘回归的洪水预报方法  

A method of flood forecasting based on partial least squares regression

作　　者：丁胜祥 董增川[2] 张莉 李庆航 

机构地区：[1]长江水利委员会水文局,湖北武汉430010 [2]河海大学水文水资源与水利工程科学国家重点实验室,江苏南京210098 [3]长江勘测规划设计研究院,湖北武汉430010

出　　处：《河海大学学报（自然科学版）》2011年第4期367-371,共5页Journal of Hohai University(Natural Sciences)

基　　金：教育部科学技术研究重点项目(104197);水利部公益性行业科研专项(200901001;200901002)

摘　　要：针对常规多元回归模型无法克服预报因子间复相关性的问题,提出了基于偏最小二乘回归的洪水预报方法.通过结合遗传算法与偏最小二乘回归来进行因子筛选,对筛选后的因子采用Bootstrap方法进行检验,再建立基于偏最小二乘回归的预报模型.实例分析结果表明,该方法建立的洪水预报模型结构简单,能依据给定目标进行因子筛选,有助于克服水文要素间复杂的多重相关性,具有较高的预报精度.In order to solve the problem of multiple correlation among forecast factors which cannot be avoided in the conventional multiple regression model,a flood forecasting method based on the partial least squares regression was developed.This method combines partial least squares regression with a genetic algorithm to select variables.The selected variables were tested by the Bootstrap method.Case study results show that this method can effectively select variables by given objects and avoid multiple correlation among hydrological elements.The method has the advantages of simple structure and high accuracy.

关 键 词：洪水预报 偏最小二乘回归 遗传算法 因子筛选 

分 类 号：P338[天文地球—水文科学]