- 學位論文
類神經網路於颱風降雨與流量預報之研究
Artificial neural network for typhoon rainfall and flood forecasting
摘要
摘要 本研究探討颱風因子是否為影響颱洪流量預測的重要因子，文中以三種不同輸入項之洪水流量預報模式來進行分析。首先，本研究嘗試先以可測得之颱風特性資料及雨量、流量為輻狀基底函數網路之輸入值建立洪水流量預報模式，稱之為ANN1。進而再建立將颱風特性資料與雨量測站之降雨資訊先納入倒傳遞類神經網路與半變異元理論，所求得之降雨量預報值，再和觀測雨量、流量為輻狀基底函數網路之輸入值建立洪水流量預報模式，稱之為ANN2。另外建立只納入觀測雨量、流量為輻狀基底函數網路之輸入值之洪水流量預報模式，稱之為ANN3，以資比較。最後比較三種模式預報結果，發現ANN2模式對於颱風洪水流量之預報有更加精確之預報能力。此外，由本研究亦可得知颱風因子在預測流量方面確為可靠的重要因子。
並列摘要
Abstract The main objective of this thesis aims at clarifying whether typhoon is a vital factor to be considered when forecasting runoff. To advance the research, I construct, three different runoff forecast models that each consists of several distinct input values. For the first forecast model, denoted ANN1, I incorporate variables such as the measurable typhonic data, rainfall, and runoff to be the input values of the radial basis function network. In the second model, first I acquire the forecasted rainfall value through integrating the typhonic data and rainfall information into the back-propagation n network and Semivariogram thesis. Then I combine the outputted rainfall value with the runoff to assemble the second forecast model, named ANN2. For the third forecast model, marked ANN3, I plug in only the observed rainfall value and runoff as the input values of the radial basis function network. In conclusion, I compare three models on the basis of their predicting capability and accuracy level. The result reveals that ANN2 appears to be the most reliable model among others in forecasting the typhoon-related runoff. Moreover, the research goes on to attest the hypothesis that typhoon is indeed a crucial factor for forecasting runoff.