自組非線性系統應用於颱風波高預測

Typhoon Wave Height Forecasting by Using Self-Organization Algorithm Model

顏沛華(Pei-Hwa Yen) ； 黃暉翔(Huei-Siang Huang) ； 張志賢(Jhih-Sian Chang)

《海洋工程學刊》 13卷3期 (2013/09) Pp. 215-237

https://doi.org/10.6266/JCOE.2013.1303.03

摘要

本研究以GMDH（Group Method of Data Handling）演算法為基本架構，蒐集2006至2009年間所發生15個颱風事件於新竹資料浮標測得之波高（H）、風速（V）及由中央氣象局每6小時發布颱風中心坐標計算所得距浮標之距離（L），及對應之方位角（θ）等四個參數建立浮標所處海域前置時間6小時之「颱風波高預測模式」；而加入若干即時更新數據取代舊有資料之遞迴GMDH程序，可使模式具時變性而能自我調整，達到持續且精確預測的效果。另以雙對數方式可建立「逐時波高預測模式」。應用該二模式於新竹海域模擬預測8～14組颱風事件結果顯示，其平均誤差均方根（RMSE）介於20.37～41.34 cm間，相關係數（CC）介於80.09～90.76％間，平均誤差尺度百分比（RMSE/H1/3-ave）則介於17.32～26.41％間，證實本研究發展之模式可應用於颱風波高之預測實務。

關鍵字

颱風波高預測 ； 自組非線性系統模式

並列摘要

A forecasting model of typhoon wave height based on the GMDH structure with four parameters of wave height (H), wind speed (V), distance (L) and azimuth (θ) in between the target location and typhoon center is proposed in this paper. Data observed at Hsin Chu data buoy and obtained from CWB were be used to construct the prior 6 hrs forecasting model with 15 typhoon events data during 2006 to 2009. A recursive GMDH model could match the time variant properties to improve the predict accuracy by using update data to substitute prior data. A prior 1 hr forecasting model could be set up with the log-logarithm correlation between progressive every 6 hr's and every hour's measured data. Both prior 6 hrs and 1 hr forecasting with data of 8-14 typhoon events result reasonable predicting accuracies with the average RMSE, CC and error scale ratio in between 20.37 ~ 41.34 cm, 80.09 ~ 90.76% and 17.32 ~ 26.41% respectively and make both models possessing the practical usage of typhoon wave height forecasting at the specific surroundings.

並列關鍵字

typhoon wave height forecasting ； self-organization algorithm model