以共用臨域法預測旬流量

Shared Near Neighbors Modeling Approach to Streamflow Forecasting

陳彥璋(YEN-CHANG CHEN) ； 邱曉萍(HSIAO-PING CHIU) ； 張斐章(FI-JOHN CHANG)

《台灣水利》 49卷4期 (2001/12) Pp. 14-23

摘要

集水區降雨－逕流具複雜之非線性關係，本研究利用共用臨域法（Shared Near Neighbors，SNN）對雨量與流量進行分類，再利用線性迴歸來建構降雨－逕流之模式。而研究中所採用之共用臨域法爲非監督式學習，此法之優點包含可處理非球狀群集之問題、演算法則簡易、所需參數少及計算所需時間短等。經由此分類之觀念可將複雜之降雨－逕流關係降低，且可將資料分爲數個區間，並經由少數之線性關係之組合推估暴雨時間之逕流量。於模式實用性之測試方面，係利用大甲溪降雨逕流之旬資料來分析，並與時間序列ARMAX模式作比較，經研究證實，SNN模式所得分類與推估之結果於低流量上已有相當之準確性，且SNN模式比ARMAX模式所預測流量之結果佳。

關鍵字

流量推估 ； 非線性系統 ； 共用臨域法

並列摘要

The relation of rainfall and runoff in a catchment is extremely complex and highly nonlinear. In this study a nonparametric clustering technique, shared near neighbors (SNN), is presented and used to cluster the patterns of rainfall and runoff. Then the linear regression method is applied in each cluster to build the rainfall-runoff model. SNN can be seen as unsupervised learning. The most advantage of SNN is that it can deal with the nonglobular clusters. Clustering data in a nonparametric way, the computational elegance of the model, and saving computing time are the other advantages of SNN. By using SNN, the complex relation of rainfall and runoff can be reduced and clustered into a few patterns. The available rainfall and runoff data of every ten-day of the Dachia River, in central Taiwan, is used to evaluate the proposed model. A comparison of the results obtained by SNN and time series model, ARMAX, indicates the superiority of SNN.

並列關鍵字

Discharge estimation ； Nonlinear system ； Shared near neighbors