頁籤選單縮合
題 名 | 龍洞海域颱風波浪之類神經網路模式=Artificial Neural Network Models for Typhoon Waves at the Longdong Coast in Northeastern Taiwan |
---|---|
作 者 | 蔡仁智; 邱方瑜; 蔡政翰; | 書刊名 | 海洋工程學刊 |
卷 期 | 12:2 2012.12[民101.12] |
頁 次 | 頁235-256 |
分類號 | 443.1 |
關鍵詞 | 颱風波浪; 倒傳遞類神經網路; 模糊理論; 轉換函數; Typhoon waves; Back-propagation neural network; Fuzzy theory; Transfer function; |
語 文 | 中文(Chinese) |
中文摘要 | 本研究使用倒傳遞類神經網路(BPNN)建置一個台灣東北角龍洞海域的颱風波浪模式。我們收集了2002-2007年間之氣象、波浪以及颱風相關資料,在此期間總共有15個有效颱風資料。本文依不同的風場參數分別建立3種颱風波浪預報模式。研究中為考慮颱風登陸後受地形的影響,我們藉由模糊歸屬函數修正颱風中心位置與測站間距離及颱風方位角,並考慮時間因子對波浪的影響,故將每一輸入參數往更早時間推至2-6小時前之資料做為輸入,進而推算出每下兩小時颱風波浪的波高。經測試結果,我們發現最佳模式是以中央氣象局所發佈的颱風路徑,並配合美軍關島聯合颱風警報中心(JTWC)提供的颱風暴風半徑四個象限大小,所計算的暴風半徑涵蓋面積做為風場輸入資料,及颱風中心相對測站移動速度、颱風中心與觀測站距離和颱風方位角做為影響測站未來波高的輸入參數。推算結果顯示最佳推算值與實測值波高之相關係數CC為0.839;均方根誤差RMSE為76公分;3個測試颱風波高峰值誤差分別為5%、9%與2.9%;峰值時間誤差分別為0小時、0小時與-2小時(即模式比實測早2小時),故結果顯示本文所建置的倒傳遞類神經網路模式能有效地模擬龍洞海域的颱風波高。 |
英文摘要 | This study presents a back-propagating neural network model for typhoon waves at the Longdong coast in northeastern Taiwan. We collected meteorological, typhoon, and wave information measured at the Longdong station between 2002 and 2007. Among these data, we have found information of 15 typhoons that can be used for this study. We built three neural network wave models based on various wind and wave parameters. Considering the effects of land on the typhoon wind field, we incorporated membership functions to represent the distance and bearing of the typhoon relative to the wave station. After data trainings of our models, they were tested with data of three other typhoons to see how the model would hindcast the wave height. It was found that the model with best results used typhoon tracks provided by the Central Weather Bureau, four quadrants radii from the Joint Typhoon Warning Center, typhoon forward speed relative to the wave station, distances and the bearings of the typhoon as input parameters. Our test results showed that the best correlation coefficient between the model and measured wave height was 0.839 with root mean square error of 76 cm. The errors of peak significant wave height during the typhoon passage for the three tested typhoon were 5%, 9% and 2.9% respectively. The errors in the time of peak wave height occurrence were 0, 0 and -2 hour (model peak wave height ahead of the measured by 2 hours), respectively. This shows that our neural network can model the typhoon waves reasonably well. |
本系統中英文摘要資訊取自各篇刊載內容。