嵌入天文潮於神經網路之潮汐推算模式研究

Abstract

潮汐是一種海水水位週期運動的一種現象。潮汐對於海岸活動、海岸結構的設計高程及海港船 隻進出安全有相當影響。因此，如何正確預測潮汐是相當重要的。往昔利用潮位測站實際量測水位， 以瞭解水位高低，再利用調和分析法分析組合潮位的各分潮的特性。但是調和分析法的缺點為需要 長時間的資料才能達到預測精度，而遺缺的資料，便會影響預測的品質。另一個應用在工程上的問 題為，若欲求解某地之潮汐均必需有觀測資料方能分析各分潮或預測未來的潮汐水位。因此，本研 究將應用神經網路可快速計算及自我學習等解決複雜問題之優點，建構一個台灣地區潮位推算模 式。形成潮汐之主要引力來源，大部分來自月球與太陽對地球的引力作用，本研究將建立相關天體 引力之參數為神經網路模式之輸入參數與實測潮汐為輸出參數之關係，在未來，本模式將只需要天 文引力之參數，即可推求長期的潮汐水位。另外，本模式將推廣應用於學習測站以外無潮汐資料之 推算，以解決實際工程上無潮汐觀測站之潮汐推算問題。最後，本模式將與NAO99b 之數值潮汐模 式比較推算潮汐之能力。

Tide is a natural phenomenon of water level varying periodically with time. The induced tidal current affects the decision of water level for marine structures, navigation safety in harbors. It is of importance to accurately calculate tidal level for marine engineering and navigation. Harmonic analysis is very popular method of distinguishing the possible components from a long duration of tidal data. If tidal data is short or absent, harmonic analysis is not applicable. Neural networks have high speed computation and optimally self-adapting to solve some complex problems. Thus neural networks have become increasingly popular for wide applications. As we know the tides on the earth are primarily induced by astronomical forces of the moon and the sun, a theoretical tide theory can be applicable to calculate tidal level with rather accuracy. The difference between calculated tidal levels and measured ones can be remedied from neural networks. The present study will develop a neural network to connect the relation of input theoretical tidal level to the real tidal levels measured. The relationship can be decided by a weight matrix and a bias matrix through optimization process in the chosen Back-propagation neural networks. The developed neural network tidal model can be used to calculate the long-term prediction of tidal level. The present model can be applicable to calculate tidal level at a point where no real tidal data are available, thus the model avoids disadvantages of harmonic analysis. The present model will be examined the capability of predicting tidal level at any point and in any time by a well-developed numerical NAO99b model. The present study will provide a powerful neural network tide model to calculate tidal levels along the Taiwan coast to solve some problems of applying harmonic analysis to calculate tidal level.