輻狀基底函數網路於水文系統之研究(1/3)

本年度為輻狀基底函數網路於水文系統之研究計畫第一年。本研究將嘗試應用全面監督 式訓練法則來建立輻狀基底函數網路之網路架構，作為洪水流量之預報模式。全面監督式學 習於訓練過程中，具有自動決定隱藏層神經元數目及訓練誤差隨著神經元增加而遞減之優 點，避免了使用複合式訓練法則時，必須先設定隱藏層神經元個數之困擾。再者，本研究於 網路訓練時加入了提早停止之條件，以避免網路過度配合現象的產生。最後，將其應用於翡 翠水庫集水區之水庫入流量預測上，經由四種評鑑指標計算之結果發現，應用全面監督式訓 練法則所建立之輻狀基底函數網路對於洪水流量之預測具有相當良好之效果。由於此模式具 有網路建構容易、學習速度快速以及對外在環境能快速適應等優點，因此在實際之洪水流量 預測上具有相當之優勢。This study is the first year’s work of a three-year project. In the project, the radial basis function network (RBFN) is used to construct a rainfall-runoff model, and the fully supervised learning algorithm is presented for the parametric estimation of the network. The number of hidden layer neurons can be constructed automatically and the training error decreases with increasing number of neurons. The fully supervised learning algorithm has advantages over the hybrid-learning algorithm that has the trouble of setting up the number of hidden layer neurons. Furthermore, early stopping technique is used to cease training, which can avoid over-fitting during the process of network construction. Finally, the proposed methodology is applied to the Fei-tsui Reservoir watershed to forecast the one-hour ahead inflow. The result shows that the RBFN can be applied to build the relation of rainfall and runoff successfully