2009-08-012024-05-18https://scholars.lib.ntu.edu.tw/handle/123456789/704447摘要：學習向量量化（Learning Vector Quantization, LVQ）方法為一種新式的分類演算法（classification algorithm），具有呈現可視化結果與萃取資料特徵的特性。此法已被廣泛地運用於資訊工程領域，然尚未見於水文領域中。本研究首度應用學習向量量化方法於分析降雨特性，建置研究區域降雨特性的空間變異性與地形地貌之關聯性，並將建置之關聯性用於推估未設站處之降雨特性（總量和時間分布）。在水資源與防洪工程之規劃與設計中，常需設計降雨資料。然而設計降雨建構過程需要雨量測站之實際量測降雨特性資料，而所建立之雨型在使用上也僅限於鄰近測站處。對於未設站或測站過於遙遠之地點，則無法有效地建立設計降雨。區域化方法常被使用來推估未設測站處之水文特性，但傳統的分析方法使用上卻存在無法得到客觀的分析結果以及缺乏效率的缺點。為解決此一問題，本研究將以學習向量量化方法架構一個有效且客觀的未設測站處降雨特性之推估模式，並將本研究所研發之模式與利用傳統群集分析方法所建立之推估模式進行比較，其成果將可提供相關設計單位做為推估未設測站處降雨特性之參考。<br> Abstract: Learning vector quantization (LVQ) is an effective algorithm for classification and has been widely used in computer science. However, LVQ is seldom applied to solve hydrological problems. In this study, the potential of use of LVQ for extracting features from hydrological data is investigated. In hydraulic engineering planning and design, the design hyetograph is usually used to represent the time distribution of total design rainfall depth corresponding to a duration and a return period. The influence of the design hyetograph on the shape and peak value of the resulting design runoff hydrograph is significant. The design hyetograph is built merely at a location which has rainfall gauge to record rainfall data. If a site for interest is ungauged, it is difficult to obtain its design hyetograph. To efficiently estimate the design hyetograph and the design rainfall depth of ungauged sites, an LVQ-based approach is proposed. The proposed approach is capable of extracting features from rainfall data and providing visual representations. Comparisons between the proposed and conventional approaches based on the K-means, Fuzzy C-means, and Ward’s methods are presented to demonstrate the advantages of the proposed approach.學習向量量化分類演算法設計降雨未設站處learning vector quantizationclassificationdesign rainfallungauged site