颱風期間花蓮港船隻動態的ANN預警模式之建立

Abstract

本文利用花蓮港港務局所提供85年~92年之船隻動態記錄，分析其船隻動態之時間與颱風規模及位置之關係。經由往昔之文獻選取影響波高之可能影響因子，並將其分類，進而利用類神經網路建立花蓮港船隻異動指數的預警模式。為得到最佳之預警模式，類神經網路測試的參數，包含隱藏層神經元數、隱藏層轉換函數及隱藏層層數，輸入之影響因子為颱風路徑、颱風規模、颱風經緯度、颱風風速以及颱風中心與花蓮港之方位角。經由模式測試結果可知，最佳花蓮港船隻預警模式為單層隱藏層其神經元數為27，且轉換函數為正切轉換函數，此模式可精準地預測出颱風來臨時之船隻異動指數，使未來花蓮港船隻在颱風來襲之時，能經由此模式模擬出之船隻異動指數，以判定船隻是否有出港避湧之需要。

The paper develops an ANN alert model of ship escape from Hualien harbor during typhoon period. The conditions of official annual ship escape reports from 1996 to 2003 are classified into four grades and related to the corresponding typhoon position and scale. Typhoon’s path, scale, maximum wind speed and position are the dominant affecting factors to be the input parameters in the present ANN model. The present ANN model with 27 neurons in only one hidden layer and with hyperbolic tangent sigmoid transfer function is examined to provide the fittest and most precise alert system for ship escape from Hualien harbor during typhoon period. The construction of the present ANN model can be extended to the other harbor if the ship escape report is detailed and long enough.