English  |  正體中文  |  简体中文  |  Items with full text/Total items : 28611/40649
Visitors : 618070      Online Users : 73
RC Version 4.0 © Powered By DSPACE, MIT. Enhanced by NTU Library IR team.
Scope Adv. Search

Please use this identifier to cite or link to this item: http://ntour.ntou.edu.tw:8080/ir/handle/987654321/52875

Title: 以深度學習法之深度神經網路預測近岸颱風風浪之研究
Nearshore wind-wave predictions using deep neural networks of deep learning techniques during typhoons
Authors: Cheng, Ju-Yueh
Contributors: NTOU:Department of Marine Environmental Informatics
Keywords: 颱風風浪;類神經網路;深度神經網路
typhoon wind height;Artificial Neural Network;Deep Neural Networks
Date: 2018
Issue Date: 2020-01-20T06:20:34Z
Abstract: 颱風多生成於太平洋西側的間熱帶輻合區,又受到副熱帶高壓的外圍氣流影響,多數往西、西北方向移動,而臺灣本島地理位置約為北緯21°54' 至25°18'、東經120°至122°之間,屬副熱帶氣候且位於西太平洋颱風的主要路徑上,兩者互相影響,易導致極端氣候的產生,強風不但挾帶豪雨,也助長巨大風浪的形成。本研究為求準確掌握下時刻的風浪資料,以深度神經網路(DNN)建立模式,達到預測之功能,降低威脅。 本研究利用深度神經網路建立二種不同模式方案,並以龍洞測站及龜山島測站為例預測示性波高。資料來源為中央氣象局2002年至2017年的颱風警報資料、海象資料及氣象資料。根據案例設計篩選相關之資料屬性,再將資料分成三組,訓練資料(2002年-2008年)、驗證資料(2009年-2013年)及測試資料(2016年-2017年),訓練資料用來建立深度學習模式架構,驗證資料則用來找出最佳模式參數,測試資料為另一組獨立數據,用來評量模式預測優劣。本研究所設計的兩種模式方案,分別為:Case 1的直接預測浪高(簡稱一階段式預測方案)、及Case 2為先預測風速、再預測浪高(簡稱二階段式預測方案)。本研究實驗區域為臺灣東北海域的龍洞測站及龜山島測站,研究結果顯示未來1小時的示性波高的預測結果,兩測站皆以二階段式預測結果較一階段式為佳。另外,未來2至6小時的預測評估結果,亦以二階段式預測結果較一階段式為佳;再比較兩測站準確性結果顯示,龜山島測站較龍洞測站為準,推測其原因可能為龍洞測站受地形較為複雜之影響,因此龍洞測站預測誤差值較龜山島測站大。
Typhoons are mostly generated in the inter-tropical convergence zone on the western side of the Pacific Ocean, and are also affected by the peripheral airflow of the subtropical high, mostly moving to the west or northwest. Taiwan is located at a distance of 21°54 to 25°18 north latitude and 120° to 122° east longitude. It is a subtropical climate and is located on the main path of the typhoons in the western Pacific. The two reasons influence to each other and are prone to extreme weather. The strong wind of the typhoons not only brought heavy rain, but also led to the formation of huge waves. This study uses the deep neural network to establish a model, in order to accurately grasp the wind speed and wave hight at the next moment, to achieve the function of prediction and reduce the threat of life. In this study, three different models were established using the deep neural network (DNN), and the indicative wave height was predicted at the Long-dong station and the Guishan Island station. The source of the research data was selected from the typhoon warning, walrus data and meteorological data from 2002 to 2017 by the Central Meteorological Administration. According to the relevant materials selected by the case designs, the data is divided into three groups: training data set, verification data set and testing data set. The training data set is used to establish a deep learning model, and the verification data set is used to find the best model parameters. The testing data set is another set of independent data, which is used to evaluate the quality of the model. The model is designed in two ways, directly predicting the wave height, called the one-stage prediction (Case 1), and first predicting the wind speed and then predicting the wave height, which is called the two-stage prediction (Case 2). This study area is the Long-Dong Station and the Guishan Island Station in the northeastern waters of Taiwan. This study shows that the wave height predictions of both stations are the result of Case 2, which is better than the case 1.In addition, the observation of the lead time of 1 to 6 hours is the result of the Case 2, which is better than the Case 1.Therefore, it is speculated that the Long-Dong station may be affected by the terrain, resulting in a larger error value than the Guishan Island station.
URI: http://ethesys.lib.ntou.edu.tw/cgi-bin/gs32/gsweb.cgi?o=dstdcdr&s=G004054E004.id
Appears in Collections:[海洋環境資訊系] 博碩士論文

Files in This Item:

File SizeFormat

All items in NTOUR are protected by copyright, with all rights reserved.


著作權政策宣告: 本網站之內容為國立臺灣海洋大學所收錄之機構典藏,無償提供學術研究與公眾教育等公益性使用,請合理使用本網站之內容,以尊重著作權人之權益。
網站維護: 海大圖資處 圖書系統組
DSpace Software Copyright © 2002-2004  MIT &  Hewlett-Packard  /   Enhanced by   NTU Library IR team Copyright ©   - Feedback