結合自組織映射圖網路與輻狀基底函數網路於地下水位預測之研究

Study on Groundwater Level Forecasting by Combining the Self-Organizing Map and the Radial Basis Function Network

陳儒賢(Lu-Hsien Chen) ； 陳清田(Ching-Tien Chen) ； 潘衍谷(Yan-Gu Pan) ； 林典蔚(Dian-Wei Lin)

《農業工程學報》 55卷2期 (2009/06) Pp. 42-55

https://doi.org/10.29974/JTAE.200906.0003

摘要

台灣受全球持續暖化所引發氣候變遷之衝擊，致水文環境有重大改變，且氣候變異現象亦加劇北澇南旱之趨勢，更凸顯水資源經營管理之重要性，而台灣於短期內難以開發新地表水資源。於有限水資源前提下，如何利用地下水資源實乃一重要課題。本研究結合自組織映射圖網路與輻狀基底函就網路之理論，建置一地下水位預測模式稱爲SOM-RBFN模式，並將此模式應用於雲林斗六地區六個地下水位站（自1997年8月至2003年12月）之月平均地下水位預測上。首先，先以SOM所得之二維密度圖解決傳統RBFN網路於訂定隱藏層神經元數目及中心點之不確定性問題，再者，利用RBFN之原理運算隱藏層至輸出層間神經元之權重值。而經本研究結果顯示，應用多站預測單站之月平均地下水位結果較單站預測單站之精確度爲高，然而多站之數量並非越多越好，其與模擬結果之精度並非成完全正比。此外，由於SOM-RBFN模式擁有較簡單之網路架構及簡易之學習演算法則，且多站預測模式相對地擁有更佳之精確度，因此建議可應用多站SOM-RBFN模式於地下水位之預測上。

關鍵字

地下水 ； 輻狀基底函數網路 ； 自組織映射圖

並列摘要

Taiwan is subjected to correlation influence of global warming and climatic changes causes the significant changing of hydrology environment. The warming current of Taiwan is the same as global's, and intensifies the tendency of north waterlogging and south drought. Thus, it demonstrates the importance of water resource management. However, it is difficult to search new surface water resource in the short term in Taiwan. Therefore, the groundwater use plays a decisive role under using the premise of limited water resource. In this paper, a groundwater level forecasting model is proposed by combing the theory of self-organizing map (SOM) and radial basis function network (RBFN). It is examined by simulated six groundwater stations's data in Douliou city, Yunlin county. Traditionally, the number of hidden units and the positioning of the radial basis centers are crucial problems for establishing RBFN. The result shows that proposed model can decide the number of RBFN's hidden units with using the two-dimensional feature map which is constructed by SOM, and then it can determine the positioning of the radial basis centers easily. Finally, the proposed model is applied to actual groundwater head data. It is found that the proposed model which with mult-stations' data can forecast more precisely than single. For groundwater level forecasting, the proposed model which with molt-stations' data is recommended as an alternative to the other method, because it has a simple structure and can produce more reasonable forecasts.

並列關鍵字

Groundwater ； Radial Basis Function Network ； Self-Organizing Map