基于MI-ANFIS的中长期水文预报方案优选研究  

Optimization selection of mid and long-term hydrologic forecast schemes based on MI-ANFIS

作　　者：路剑飞[1,2] 陈子燊[1] 王扬圣 

机构地区：[1]中山大学地理科学与规划学院水资源系,广州510275 [2]广州海洋地质调查局,广州510760 [3]三洲田铜锣径水库管理处,广东深圳518118

出　　处：《水力发电学报》2013年第2期48-53,共6页Journal of Hydroelectric Engineering

基　　金：2009年广东省水利创新研究项目(批准号:2009-41)

摘　　要：本文利用互信息法(MI)对广东省西江流域梧州至高要段的实测月径流和月降水时序的多个预报方案进行优选,结合自适应神经模糊推理系统(ANFIS)进行验证和进一步筛选,得到最优的预报方案。研究结果表明:1)当有多个输入变量可供选择时,如果每个输入变量与输出变量之间的广义相关指数相差不大,则单独采用任何一个输入变量建立预报模型时,得到的预报精度应该差不多;2)ANFIS方法可以有效的辅助互信息法对预报方案进行筛选,既避免了由于互信息计算方法的片面性导致的对预报准确率的低估情况,又能有效的适应由于实际水文系统的非恒定性导致的预报方案的理论预报准确率与实际值的较大出入。This study adopted mutual information(MI) to optimize and select forecast schemes for the monthly series of stream flow and rainfall in the Xi River reach from Wuzhou to Gaoyao in Guangdong province,and verified the results using an adaptive network-based fuzzy inference system(ANFIS).It is concluded that in the case of multiple input variables available,if their generalized correlation indices with the output variable are nearly the same,a forecast model that uses any one of them,will produce similar forecast precision.And ANFIS can be used as effective assistance to mutual information method in selection of forecast schemes to avoid underestimation of forecast precision caused by one-sidedness of mutual information calculations.With such assistance,the forecast model can accommodate the difference in theoretical and actual forecast accuracies caused by non-constant features of concrete hydrological systems.

关 键 词：水文学 中长期水文预报 预报方案优选 互信息法 ANFIS 

分 类 号：P338.2[天文地球—水文科学]