以輻狀基底函數網路建立降雨－逕流模式

集水區降雨－逕流為複雜之非線性關係，本研究利用人工類神經網路來推求降雨－逕流模式。研究中所採用之輻狀基底函數類神經網路為混合式之學習網路，前階段之訓練為非監督式學習，以模糊最小最大分類法決定輻狀基底函數之中心及個數；後階段為監督式之學習，以最小平方法推求網路隱藏層與輸出層間之參數。由於利用模糊最小最大團塊法取代k-means法則對訓練範例進行分類，使得建構模式之過程中無需設定眾多之參數，僅需輸入團塊擴展係數；再加上引入統計推論的方式對模式之驗證進行有效性測試，故本研究可明確減少傳統輻狀基底函數類神經網路之建構時程。於模式實用性之測試方面，係利用蘭陽溪颱風暴雨及流量資料來建構出降雨－逕流模式並據以預測流量。經研究證實，此一方法可適切的應用於集水區複雜之非線性降雨－逕流模式的建立。

關鍵字

類神經網路；非線性系統；輻狀基底函數；降雨－逕流；模式有效性

並列摘要

The rainfall-runoff relation is notoriously nonlinear. In this study, the artificial neural network is used to construct a rainfall-runoff model. The radial basis function neural network (RBFNN) uses the hybrid unsupervised and supervised learning schemes. Training a RBFNN occurs in two phases. During the first phase, the unsupervised learning, fuzzy max-min clustering, is used to determine the center and the number of radial basis functions. During the second phase of the algorithm, the weights from the hidden layer to output layer are determined by the method of least squares. It is not necessary to setup a great of parameters. The only parameter needed is user-defined value θ. The rainfall-runoff model can be constructed automatically by the training data. The model is calibrated through training data and verified by the test of model validation. The typhoon data of the Lan-Young River is used to construct the rainfall-runoff model. The result shows that the RBFNN can be applied successfully to build the relation of rainfall and runoff. High prediction accuracy of runoff predicted is obtained.

並列關鍵字

Artificial neural network ； Nonlinear system ； Radial basis function ； Rainfall-runoff ； Model validation

主題瀏覽