類神經網路在洪水演算及預報之應用

Application of Artificial Neural Networks on Flood Routing and Forecasting

鍾世豐(Shih-Feng Chung)

指導教授 ： 許銘熙

國立臺灣大學/生物資源暨農學院/生物環境系統工程學研究所/碩士(2007年)

https://doi.org/10.6342/NTU.2007.02843

摘要

台灣屬於海島型氣候國家，夏秋兩季因颱風之侵襲造成許多重大災害。有效利用氣象及水文資訊以提高洪水預報的精確度做為減災及應變之用就顯得相當重要。傳統洪水預報系統(降雨-逕流-洪水預報)常受限於較複雜的降雨-逕流預報的估算方法。故為能簡單快速的獲得邊界條件，直接採用類神經網路預報上游邊界及河段內部水文站之水位，並利用水文監測資料即時修正洪水預報之初始值，配合洪水演算模式做為河川各斷面之洪水位預報。為驗證模式的效能，本文針對淡水河系，以近年來紀錄較完整之三場颱洪事件進行模式測試，並探討動力波模式與初始值修正、及類神經網路之洪水演算之間的差異，從中探討預報未來一至三小時水位變化較精確之方式。
由模擬結果顯示動力波模式受限於上下游邊界預報值、側入流及抽水站流量影響，而初始值修正之動力波演算，可有效改善預報第一、二小時之精度，但第三小時之改善程度較有限，在類神經網路之洪水演算中，其側入流及抽水站影響較小，受水位站預報精度影響較大，但預報未來一至三小時可獲得較精確之水位值。由本文所建立淡水河之河川洪水預報模式測試結果，確可在颱洪期間提供合理及準確的河川洪水資訊，以供應變決策之參考。

關鍵字

動力波 ； 初始值修正 ； 整合類神經網路水位預報 ； 洪水位預報 ； 類神經網路

並列摘要

Taiwan located at the sub-tropic monsoon climate area. Typhoon occurrences often cause huge damages in summers and autumns. Flood forecasting model is a useful tool to capture precise meteorological and hydrological information for flood mitigations and responses emergency. Traditional flood forecasting model using the rainfall-runoff forecasting technique was limited to long time computation due to complicated approach. In order to obtain hydrological information in upriver rapidly, this study uses artificial neural networks (ANN) model to predict the stage of upriver in observed stations along a river. Moreover, for the sake of forecasting the stage of each section in the river, this study utilizes real-time observed stage to correct the initial stage with the incorporation of flood routing process for flood forecasting. Three typhoon events were simulated to confirm the accuracy of the forecasting model. The results are compared among dynamic routing, initial stage correction for forecasting, and integrated flood forecasting model with ANN, so as to obtain the suitable approach from three methods.
The results reveal that the accuracy of dynamic routing is seriously influenced by predicting stage in upriver, lateral inflow discharge, and pumping stations discharge. The initial stage of forecasting is able to well predict the stage for the lead time of two hours, but less performance after three hours. Moreover, the accuracy of predicting stage at stage stations has significant influence on integrated flood forecasting model with ANN, which the lateral inflow discharge and pumping stations discharge become less influence. By referencing the above forecasting information, better decisions can be made by the emergency operation agency during storm occurrences.

並列關鍵字

Dynamic routing model ； initial stage of forecasting ； integrated flood forecasting model with ANN ； Artificial Neural Network ； flood forecasting

參考文獻

18. 吳建均，2005，河川洪水位演算模式之研究，國立台灣大學生物環境系統工程學研究所碩士論文。

32. 陳信中，2006，蘭陽溪洪水預報模式之研究，國立台灣大學生物環境系統工程學研究所碩士論文。

38. 蔡亞欣，2003，模糊範例學習推論系統於水位預測之研究，國立台灣大學生物環境系統工程學研究所碩士論文。

1. Chiang, Y. M., Chang, L. C., and Chang, F. J., 2004, Comparison of static-feedforward and dynamic-feedback neural networks for rainfall-runoff modeling, Journal of Hydrology, volume 290, pp.297-311.

2. Chang, F. J., Chang, L. C., and Chiang, Y. M., 2005, Reply on ‘Comparison of static-feedforward and dynamic-feedback neural networks for rainfall-runoff modeling’ by Y.M. Chiang, L.C. Chang and F.J. Chang, 2004, Journal of Hydrology 290, 297-311., Journal of Hydrology, volume 314, pp.204-206.