基于主成分分析的K均值聚类法在洪水预报中的应用  

Application of K-mean cluster method to flood forecasting based on principal component analysis

作　　者：刘可新[1] 包为民[1,2] 阙家骏 李佳佳[1] 束慧连 

机构地区：[1]河海大学水文水资源学院,江苏南京210098 [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

出　　处：《武汉大学学报（工学版）》2015年第4期447-450,458,共5页Engineering Journal of Wuhan University

基　　金：国家自然科学基金资助项目(编号:51279057;41371048);国家自然科学基金重大项目(编号:51190091);国家重点实验室专项基金(编号:2009586412)

摘　　要：洪水预报中影响因素很多,蕴含的信息也很复杂,如何从这些信息中获取有效信息是提高洪水预报精度的关键.考虑到洪水聚类时,指标间的相关性和信息冗余会严重影响分类效果,从而造成分类洪水预报精度不佳,应用主成分分析方法,力求提取历史洪水的有效信息,并以这些信息为基础,运用K均值聚类方法将历史洪水分类,对各类型洪水分别率定参数,通过计算洪水指标到各聚类中心的距离来判别即将发生洪水的归属类别,采用对应的模型参数进行预报.应用于实际流域,结果表明,基于主成分分析的分类洪水预报能够有效减小运算量,提高洪水预报精度.Flood forecasting is affected by so many factors that accessing to effective information makes the key to improve the flood forecasting accuracy.It is considered that there exits dependency among flood indexes which may affects the result of classification that is the key factor to forecasting accuracy;then the principal component analysis was applied to extract effective information from historical floods.The information was then used as the basis to classify the historical floods.After that,model parameter calibration,directed to different kinds of floods,was redone.Finally,the testing floods were forecasted by using the parameter that fitted them best.It turned out that the computational cost was reduced and the forecasting accuracy was improved by using the method.

关 键 词：主成分分析 K均值聚类 洪水预报 

分 类 号：P338[天文地球—水文科学]