2004-03-242024-05-14https://scholars.lib.ntu.edu.tw/handle/123456789/658075摘要：台灣地區近年遭逢包括賀伯、象神、潭美、桃芝、納莉等重大颱洪侵襲，造成嚴重災情。頻仍發生之颱洪事件，所導致之損失往往難以估計。因此，建置具即時及準確推估能力之智慧型防災預警系統實刻不容緩。本計畫將針對蘭陽溪及烏溪流域建置一智慧型之河川防災預警系統，預警系統可預報至少未來三小時之水位以作為決策支援。洪水預報的模式以具人工智慧(AI)之類神經網路(Artificial Neural Network)為主，並深入探討靜態與動態等多種不同類型之類神經網路模式架構於上述河川流域之精確性與實用性。最後，配合流域不同之水文水理特性、水文測站建置數量與位址、歷史資料收集長度、即時水文資訊接收品質，來架構各流域最適用之類神經網路模式，運用類神經網路強大的智慧型學習能力及動態調整的能力，學習各河川流域之特性及歷史事件資料之結構特性成為該流域之防洪預測之主要工具。此外，配合水文觀測現代化多工多埠傳輸系統，取得各水文站即時傳回之資料，並加強自動化減少人工輸入使智慧型系統即時預測的能力更加穩健。透過網際網路以Web型式展示即時觀測值及洪水預報功能，達成水文資訊之即時化，以擴大水文資訊之範疇與提昇服務的品質與精確度<br> Abstract: Taiwan has a subtropical climate where typhoons, usually coupled with heavy rainfall, hit the island around four times a year, causing downstream flooding within a few hours. Consequently, streamflow forecasting is crucial for flood warning system. In this study, the artificial neural networks (ANNs) will be used to model the multistep ahead rainfall-runoff processes and implemented in Lanyang and Wu watershed. In order to explore ANN models accuracy, stability and practicability, many types of ANN models, such as static ANN and dynamic ANN, will be discussed deeply. ANN’s powerful intelligent learning ability and the dynamic adaptive ability will be applied to construct the best forecasting model. For the practicable purpose, the forecasting model and the coming data are integrated to provide the flood information for the decision-maker through on-line facility, such as internet or intranet.類神經網路防洪預警系統多時刻預測Artificial Neural NetworkFood Warning SystemMultistep Ahead Forecasting