在水文應用上,降雨與流量之關係通常由許多經驗或半經驗公式表示。經驗公式的概念簡單、平易、快速且實用,尤其在無流量紀錄流域或在工程設計上,透過少許的參數條件即可快速推出洪峰流量、稽延時間等。 類神經網路為眾多黑盒模型的一種,具有多變的數學結構、快速求解,能判別輸入資料與輸出資料間的非線性關係。近年來,許多文獻指出類神經網路能有效的推求雨量與流量之間的映射關係。因此本研究透過類神經網路建立林邊溪降雨逕流的模式。以不同階段的降雨時間與降雨量作為輸入資料,如0-25%、50%-75%、75%-100%等區間,推求流量歷線的重要參數,例如尖峰流量、基期、尖峰流量到達時間、尖峰流量到達50與75%的寬度、尖峰流量到達50與75%的時間。最後模式驗證4場颱洪事件,並與傳統的複迴歸方程與Snyder合成單位歷線作比較。結果顯示,類神經網路誤差在Qp與tb部份,誤差和Snyder合成歷線與複迴歸方程間差距不大。其餘tb、W50、W75、Wup50、Wup75的歷線參數,類神經網路則是在三種模式中,誤差最小的方法。因此未來在工程實務的應用上,只需要雨型的設計,導入類神經網路的模型,便可求得流域的流量歷線。
In hydrological applications, Empirical formula is conceptually simple, easy to use, commonly used in the compute relationship between Rainfall and Runoff . Particularly to no flow record region or engineering design, through the few parameters such formula can quickly estiniate a peak flow discharge and lag time. Artificial neural network (ANN) is one of a commonly used black-box model scheme with variables mathematical structure and can objectively judge the nonlinear relationship between input and output data. In recent years, many studies pointed out that the neural network can derived successfully the mapping relationship between ramfall and runoff. This study tries to applied ANN model on Linbian Creek rainfall runoff analysis. The model input data include different rainfall periods and amount in proportion to 0-25%, 25%-50%, 50%-75%, and 75%-100% stage. By using these input data, model will estimate the hydrograph parameters, such as peak discharge, base time, time to peal, time to 50% and 75% peak, and width of time of 50% and 75% peak. The final model verified by 4 flood events, and results compared to traditional multiple regression model and Snyder unit hydrograph showed that ANN model has more accurate estimation than other two methods on most of parameters. Therefore the future practical engineering application, only need thedesign hyetograph, as the input of the neural network model ehich can estimate the runoff discharge hydrograph fo the tiver baisn, hydrograph in Linbian Creek.