基于聚类分析的人工神经网络洪水预报模型研究_尹雄锐.pdfVIP

39 3 四川大 学 学 报 ( 工程 科 学版 ) V o.l 39 No. 3 2007 5 JOURNAL OF SICHUAN UNIVERSITY ( ENGINEER ING SC IENCE ED ITION ) M a 2007 : 1009-3087( 2007) 03-0034-07 1 1 1, 2 尹雄锐 , 张 翔, 夏 军 ( 1. , 430072; 2. , 100101) : C ( FCM )( SOM ), , (M FN)( FCMMFN SOMMFN ), , , , FCMMFN SOMMFN : C ; ; ; ; : TV124 : A Classification-based Flood ForecastingM odel Using ArtificialNeuralNetworks 1 1 1, 2 YIN X iong-rui , ZHAN G X iang , X IA Jun ( 1. State Ke Lab. ofW ater Resources and H dropow er Eng. Sci. , W uhan Un iv. , W uhan 430072, Ch in a; 2. Ke Lab. ofW ater C cle and Related Land Surface P rocesses, IGSNRR, Ch inese Academ of Sci. , Beij ing 100101, Ch ina) Abstract: Fuzz C M eans( FCM ) clustering method and Self-Organ izing Feature M ap ( SOM ) clustering m ethod w ere both emplo ed to decomposed the flow h drograph to several segments, and the situation of rain and runoffw as anal zed in each segment, then tw o h brid artificial neural newt orks ( FCMMFN SOMMFN ), based on Fuzz C M eans clustering method and Self-Organ izing Feature M ap clustering method separatel, w ere applied to smi u late the rainfall-runoff relationship. The case stud in W angjiachang Reservoir ind icated that wt o clustering m ethods have the abilit to decomposed the flow h drograph to several segments in wh ich the under-l ing mechan ism s of stre- amflow generation appear d ifferen.t Besides, the tw o classification-based artificial neural netw orks are both superior to the singlemulti-la er feedforw ard newt ork, furthermore, the performance of FCMM FN is better tha