船舶運動推算台灣海域波浪場模式及系統模擬

Abstract

近年來，類神經網路(Artificial Neural Network； ANN)模式已大量應用在海洋工程及船舶工程的領域之中，這些研究大多將ANN模式應用於時間域的預測上，而ANN模式之預測結果其精度也多能被工程應用上所接受，因此除了傳統的預測模式以外，ANN模式提供了另一種可能的選擇。本研究即利用ANN模式之模式辨識(Pattern Recognition)能力以建構船體運動反算波浪之類神經網路模式；並將ANN模式應用於重建波浪場於空間域的分佈情況。最後運用以上所述兩個ANN模式建構出一個屬『面』的台灣海域波浪測報系統，藉由模擬探討此系統之有效性。 本研究之目的在於利用遍佈台灣海域巡航之海巡署所屬船舶(約50~70艘)做為波浪之行動感測器(mobile agent)，透過收集各船艦之運動反應狀態，並各別代入船體運動反算波浪之類神經網路模式中，反算出各船艦所在位置之示性波高、示性波週期、波向角以及各別之船速。最後再將這些空間上分散之波浪資料(50~70組)代入重建波浪場空間分佈之類神經網路模式中，最終取得該時段台灣周圍海域之波浪場資料。研究中引用TaiCOMS之中尺度台灣近海波浪模擬資料，做為建構此系統分析方法與模式建構和驗證之依據。為了探討本研究中所提出系統之有效性，研究中利用MATLAB及Neural Solution分別做為模擬船體在短波頂不規則波中運動量分析及建構類神經網路模式之工具，進而模擬整個系統之流程，並估算出該系統之精度及評估此系統之有效性。 研究顯示所提出之船體運動反算波浪類神經網路模式能良好掌握船體運動之特性，進而反算精度良好的結果；而重建波浪場空間分佈類神經網路模式亦能掌握台灣週圍海域之波浪場分佈特性。整體來說所提出之台灣附近海域波浪測報系統是有效的，未來也能進一步對此系統之架構微調，以獲得更加準確的精度與更好的適用性。

In recent years, artificial neural network (ANN) has been applied on the field of ocean engineering and naval architecture. Most of these researches are used to do prediction in time domain, and the accuracy is acceptable for engineering applications. In the present study, the pattern recognition ability of ANN is applied to develop a model for evaluating wave characteristics basing on ship motions. In addition, ANN is also applied to reconstruct the spatial distribution of wave field around Taiwan sea area by using limited numbers of sampling wave data. In the present study, maybe 50~70 ships of the fleet of Taiwan Coast Guard are considered as the mobile agents of a sensor network for capturing the wave field around Taiwan sea area. By sensing the motion responses of each ship at the same time and putting them into ANN model, the significant wave height、wave period、incident angle and ship speed at the location of each ship can be evaluated. Then puting these scattered wave data (about 50~70 sets) into another ANN model, the spatial distribution of wave field around Taiwan sea area can be reconstructed. In this study, SWAN data of the TaiCOMS are applied for simulating the proposed ship-bone wave monitoring system and verifying the accuracy and feasibility of this system. MATLAB is used to develop the software to simulate the ship motions in short crested irregular waves, and Neural Solution is applied to develop the ANN models. ANN models have a good performance on evaluating wave characteristics basing on ship motions and reconstructing the spatial distribution of wave field around Taiwan sea area by using limited numbers of sampling wave data. The result of this research shows that the present ship-bone wave monitoring system is feasible with acceptable accuracy for engineering use.