本研究選用三角形函數、梯形函數、鐘形函數及高斯曲線函數等四種隸屬函數當輸入函數,將影響西北颱降雨量的五個氣象因子-平均相對濕度、暴風半徑、測站最低氣壓、颱風中心與測站之距離、颱風移動速度,作為五個輸入變數,進而建立模糊隸屬函數,來模擬推估西北颱的降雨量。其建構流程為先將輸入變數値做模糊化的轉換,利用MATLAB中之ANFIS軟體做模糊類神經網路演算建立模糊資料庫,建立資料庫後,將演算結果解模糊化轉換輸出,以推估各測站之降雨量,再與各測站之實際降雨量作檢驗,分析比較其學習模擬成果。 颱風之模組建構以西元1958年至2005年中央氣象局之基本颱風降雨資料,藉颱風降雨資料作模糊演算及分析,計算結果可看出四種隸屬函數於颱風降雨量之模擬及推估的整體趨勢相似,且誤差尚屬合理,四種隸屬函數之間的優劣在不同類型颱風狀況下各不相同,所得相關係數介於0.4至0.8之間,此結果希望可提供未來西北颱降雨預報參考。
In this research, the five factors affecting rainfall of northwestwardly typhoons were selected as the input membership of the fuzzy function. They are the average relative humidity , the radius of typhoons , the minimum atomospheric pressure at station, the distance between typhoon center and observed station , the typhoon velocity. These variables were considered to have triangle-shaped function ,trapezoid-shaped function , bell-shaped function and Gauss-shaped function distribution from which the membership of the fuzzy function were set up to predict the rainfalls of northwestwardly typhoons. Fuzzification transfers the input data and creats the fuzzy database by using the method of the neural nets and its computation program MATLAB/ANFIS. After the database was set up,the predicted rainfalls can be obtained by the antifuzzification. The estimated rainfalls were compared with the observed rainfalls of each station. From these comparisons , the learning effects were analyzed until the minimum errors were obtained. Using the typhoon data from the Central Weather Bureau during the period from years 1958 to 2005, to make a fuzzy calculation and analysis by using rainfall data. From the results in the learning stage of the four membership functions, the over-all trends of the estimated rainfalls for each station were similar, and the accuracies were feasible. Each membership function has its advantage in different typhoon’s type. The results of this study can be used as a reference for the rainfall prediction for the northwestwardly typhoons in the furture.
為了持續優化網站功能與使用者體驗,本網站將Cookies分析技術用於網站營運、分析和個人化服務之目的。
若您繼續瀏覽本網站,即表示您同意本網站使用Cookies。