本研究係應用最小二乘支持向量機於降雨－逕流歷程之模擬，以期提高模擬之精確度。實際之降雨－逕流歷程為非線性關係，應用線性模式進行模擬，存在一定程度之誤差。本研究採用具有非
線性迴歸功能之最小二乘支持向量機，其係基於小樣本統計學習理論，採用結構化風險最小化歸納原則，能夠在小樣本學習的基礎上對其他樣本進行快速且準確之擬合預測。為驗證模式之合適性，將所倡議之方法應用於基隆河流域五堵上游集水區之降雨－逕流歷程分析，並與常用於降雨－逕流模擬之總逕流線性響應模式比較。研究結果顯示最小二乘支持向量機，可以有效地進行模式之檢定及驗證，適用於實際之非線性降雨－逕流歷程之模擬，臻以提供水資源規劃應用之參考

This paper applies least square support vector machine (LS-SVM) to describe a new framework that enhances the precision of the modeling of rainfall-runoff processes. Unlike linear models that produce certain error when used to model actual nonlinear rainfall-runoff processes, the S-SVM with nonlinear regression capability appls statistic learning theory of small samples cquire fast and exact simulation of other samples via adopting the rule of minimum structured risks. The researchers chose a case of Wu-Tu watershed of Kee-Lung River to verify the accuracy of this proposed model, whose result is compared with the result derived from a simple linear model (SLM). Such comparison shows that LS-SVM can enhance the precision of modeling and can be effectively used to model the cali L ie to a bration and validation. The result of comparison also indicates that LS-SVM is suitable for modeling actual nonlinear rainfall-runoff processes, and thereby offers useful references to water resources planning applications.