2012-08-012024-05-18https://scholars.lib.ntu.edu.tw/handle/123456789/700105摘要：颱風降水是颱風造成災害的主因，更是造成山崩、土石流、淹水…等相關災害的關鍵性因素，於災害預警系統中，降雨預報是首要需精進的技術。然而颱風降雨具有空間和時間的高度變異性，難以架構以物理為基礎的降雨預報模式。因此，為使相關決策單位能及早預警並做出適當應變措施以降低颱風災損，本計畫擬以三年為期，以一種新型人工智慧技術建置預報更準確、結果更可靠、預報速度更快的颱風降雨預報模式，並藉由此技術能整合複雜資訊的能力，首度將颱風因子納入模式輸入項，以改善模式於較長延時預報準確度。 由於採用的研究方法為一種新型技術，因此計畫第一年的研究重點在於研析新型人工智慧技術的理論架構，評估模式於降雨預報問題的適用性，以及建立模式的建置流程。第二年則以新型人工智慧技術建置整合颱風雨降雨資訊之預報模式，並與傳統慣用的倒傳遞類神經網路模式進行預報準確度、模式強健性及模式效率三方面的比較。許多研究僅考慮預報準確度，忽略了模式在強健性及建置效率兩方面的探討，但對於颱風降雨預報問題而言，這兩方面的表現都是很重要的。強健性影響結果可靠度，是能否被應用到實務上非常重要的考量；而效率直接影響可供災變反應的時間。因此，除了比較模式預報準確度外，本計畫更以實際案例量化兩種模式於強健性與效率兩方面的表現。第三年的研究重點則在於探討颱風因子對降雨預報的影響，以及探討哪幾項颱風因子是主要影響因素，藉此更進一步改善較長延時預報準確度，如此將有助於增加政府單位及民眾對災害的應變時間，使防止或降低災損的決策有更充裕的時間得以施行。 <br> Abstract: During typhoons, rainfall forecasting plays an essential role in almost all kinds of disaster warning systems. However, typhoon rainfall is one of the most difficult elements of the hydrologic cycle to forecast because of the high variability in space and time and the complex physical process. The highly nonlinear and extremely complex physical process of typhoon rainfall also leads to a lot of difficulties in constructing a physically-based mathematical model. An attractive alternative to the physically-based models is the neural networks, which is a kind of artificial intelligent techniques with great flexibility in modeling nonlinear processes. In this project, a novel kind of neural networks called support vector machines are used to integrate typhoon and rainfall information for developing a well-performed, robust and efficient rainfall forecasting model. The proposed model can yield 1- to 6-h ahead forecasts of rainfall. The forecasts will be compared with those resulting from the back propagation network based model. In addition, the influence of typhoon characteristics on the forecasts will be investigated. To asses the improvement in forecasting performances due to the addition of typhoon characteristics, two types of model input (with and without typhoon characteristics) are designed. Finally, the influence of each individual typhoon characteristic will also be investigated. The proposed modeling technique is expected to be helpful to support flood, landslide, debris flow and other disaster warning systems.颱風降雨預報新型人工智慧技術支援向量機颱風因子災害預警系統rainfall forecastingartificial intelligent techniquessupport vector machinestyphoon characteristicsdisaster warning systems