地表及地下水聯合運用之良窳為水資源有效利用問題中重要之一環,倘能預測未來地下水位變化,則將有助於地表地下水資源運用上及早進行規劃,為精確又快速的求得地下水之水位變化。本研究旨在運用類神經網路預測濁水溪沖積扇之地下水位,在建立模式之過程中,比較倒傳遞類類神經網路(BPNN)、輻狀基底函數類神經函數(RBFNN)與回饋式類神經網路(RNN)運用在地下水水位預報之精準度。本研究結果,類神經網路可推估雲林地區三個地下水位站在未來一個月內之水位變化,以輻狀基底函數類神經網路表現最佳,倒傳遞類神經網路表現次之,誤差均在合理之範圍內,將可提供該地區地表地下水資源聯合運用、地下水位監控及地下水利用管制等議題之參考。
This paper presents a research study that predicts the groundwater level by Artificial Neural Networks (ANN) in Choushui Creek Alluvial Fan, Taiwan. The concept of conjunctive use is the key factor in water resources planning study. If we could predict groundwater level fluctuation in a systematic way, it is beneficial to increase the utility efficiency of water resources, especially in the severe land subsidence area. In the process of constructing the model, the research compares three kinds of ANNs, which are Back Propagation Neural Networks (BPNN), Radial Basis Function Neural Networks (RBFNN), and Recursive Neural Networks (RNN). The results show that RBFNN can reasonably predict the groundwater level in the Alluvial fan, and BPNN ranks second. Hence, the ANN model could actually predict the groundwater level fluctuation within the reasonable error, even if we apply several assumptions in Choushui Creek Alluvial Fan.