颱風是台灣重要的自然災害,颱風所夾帶的豪雨,造成巨大的經濟損失及人員傷亡。其中以「西北颱」為最容易造成重大災害的颱風。 由於形成颱風的因素眾多,而其產生的降雨量亦為颱風本身的特徵之一,故降雨量和颱風的其他特徵或颱風的成因之間應存在某些關聯。本論文主要的目的即在探討影響西北颱降雨量的主要因子,進而提供有效預報降雨量的方法。本研究先選取影響颱風降雨之因子,選用鐘型函數將中心最低氣壓、近中心最大風速、移動速度、暴風半徑、颱風中心距測站最短距離(以臺北測站為基準) 、測站濕度做為六個輸入變數,建立模糊隸屬函數。其建構流程為先將輸入變數值作模糊化轉換,利用MATLAB 模組ANFIS 軟體做類神經網路演算建立模糊資料庫,建立模糊資料庫之後,再將演算結果反模糊化轉換輸出,將其各測站之推估降雨量,與各測站實測降雨量作檢驗,分析評估其學習效果,至其誤差達最小。 颱風之模組建構以西元1950年至2004年中央氣象局颱風之基本資料,藉降雨資料作模糊演算及分析,計算結果可看出本文所述影響雨量六因子,對其學習階段的颱風所推求之累積雨量於臺灣北部各地及迎風面之中央山脈等地尚精確。預測階段的颱風預估累積總雨量上精度比學習階段雖較大但尚可接受,此結果可提供未來西北颱雨量預測參考。
Typhoon is the important calamity of Taiwan. The rainfall of typhoon causes enormous economic losses and casualties, especially for northwestwardly typhoons. There are many factors to form a typhoon and the rainfall from typhoon is also a symbol of typhoon itself. Therefore, rainfall and other characteristics of typhoon should exist some relationships. The purpose of this study is to investigate the major factors that affects rainfall in the northwestwardly typhoon, and then offers the effective method to predict rainfall. In this research at first the six factors affecting rainfall were chosen, as the input variables. They are the minimum atomospheric pressure, maximum wind velocity near typhoon center, move speed of typhoon center, the radius of typhoon, the shortest distance between typhoon center and Taipei monitoring station, station humidity. These variables were considered to have bell-shaped function distribution to set up fuzzy function. Fuzzification transfers the input data and creats the fuzzy database from the method of the neural nets and its computation program MATLAB/ANFIS. Then the resulting rainfalls were obtained. The estimated rainfall was compared with the observation rainfall of each station. From these comparisons, the learning effects were analyzed until the minimum errors obtained. Use the typhoon data from the Central Weather Bureau during the period from years 1950 to 2004, to make a fuzzy calculation and analysis by using rainfall data. For the results in the learning stage, the accuracies of the estimate rainfalls for each station were good, especially for the areas located in the north region of Taiwan and the windward region of the Central Mountain. Although the accuracies of rainfalls for the predicting stage were larger than that of the learning stage, their accuracies were acceptable. The results of this study can be used as a reference for the rainfall prediction for the northwestwardly typhoons.