臺灣大學: 生物環境系統工程學研究所許銘熙林洙宏Lin, Shu-HorngShu-HorngLin2013-03-212018-06-292013-03-212018-06-292010http://ntur.lib.ntu.edu.tw//handle/246246/248526台灣位處亞熱帶地區，受季風氣候影響於夏秋兩季常因颱風之侵襲造成嚴重災損，若能有效結合水文監測資訊，提高洪水預報的精度並做為早期預警及防災之用，將可有效減低洪災損失。 本文以過去研究為基礎，建立結合類神經網路之洪水演算模式以進行河川洪水位預報。在雨量-水位模式部份，係將水文監測站(含雨量站及水位站)之歷史記錄，利用類神經網路模式預測水位站之短期(1-3小時)預報水位，提供做為河川洪水演算模式之邊界條件，透過河川洪水演算進而達成全河系縱向水位短期(1-3小時)預報之目的。洪水演算模式係基於動力波方程式並以四點有限差分法求解。本文以5場颱洪事件檢定雨量-水位模式，並以另3場颱洪事件進行驗證，結果顯示結合類神經網路之洪水演算模式確能精確提供全河系縱向水位短期(1-3小時)預報。 此外，本文亦建立單一河道洪水預報模式，直接採用類神經網路模式預測兩相鄰水位監測站短期(1-3小時)之預報水位作為單一河道洪水預報模式之上、下游邊界條件。模擬預報結果顯示單一河道洪水預報模式亦可提供未設站斷面之河川水位短期預報，且具有不錯之預報精度。Taiwan located at the sub-tropic monsoon climate area. Typhoon occurrences often cause huge damages in summers and autumns. An early warning system based on the accurate flood forecast with the real-time hydrological monitoring data can be used to reduce the flood damage effectively. A flash flood routing model with artificial neural networks (ANN) predictions was developed for stage profiles forecasting. At gauge stations in a river the artificial neural networks were used to predict the 1-3 hour lead time river stages, which were taken as interior boundaries in the flash flood routing model for the forecast of longitudinal stage profiles, including un-gauged sites of a whole river. The flash flood routing model was based on the dynamic wave equations with discretization processes of the four-point finite difference method. Five typhoon events were applied to calibrate the rainfall-stage model and other three events were simulated to verify the model’s capability. The results revealed that the flash flood river routing model incorporating with artificial neural networks can provide accurate river stages for flood forecasting. In addition, a single river segment flood forecasting model was developed for comparison. In the single segment model, the 1-3 hour lead time river stages predictions from the ANN at the two adjacent gauge stations are imposed as upstream and downstream boundaries, respectively. The results show that the single segment model can provide accurate 1-3 hour lead time stage forecast at un-gauged sites efficiently.2174037 bytesapplication/pdfen-US動力波模式變量流暴洪演算類神經網路河川水位預報Dynamic routing modelUnsteady flowFlash flood routingArtificial neural networkRiver stage forecastingthesishttp://ntur.lib.ntu.edu.tw/bitstream/246246/248526/1/ntu-99-D92622003-1.pdf