請用此 Handle URI 來引用此文件:
http://tdr.lib.ntu.edu.tw/jspui/handle/123456789/91477
標題: | 運用 HDCE 方法於流量特性變化之探討 Investigation of Streamflow Pattern Alterations Using the HDCE Method |
作者: | 温浚達 Chun-Ta Wen |
指導教授: | 游景雲 Gene J-Y You |
關鍵字: | 逕流特性,時間序列聚類,階層式分群法,動態時間規整,轉折點測試,總體經驗模態分解法, Streamflow patterns,Time series clustering,Hierarchical clustering,Dynamic Time Warping,Change point detection,Ensemble Empirical Mode Decomposition, |
出版年 : | 2023 |
學位: | 碩士 |
摘要: | 逕流模式流量特性的改變是水資源管理的重要影響因素。近年來,氣候變化不僅導致極端水文事件的強度和頻率增加,也導致人類社會受到負面影響和經濟損失,然而,多數研究著重於極值的趨勢,本研究著重於水文時序特徵的變化。由於降雨的時空分佈不均,台灣的水資源問題更為嚴重。這項研究旨在發現台灣逕流模式的變化和逕流中乾濕時間點變化。本研究提出一個由層次聚類、動態時間扭曲、變化點檢測和總體經驗模態分解法組成的 Hierarchical clustering, Dynamic time warping, Change-point detection, Ensemble empirical mode decomposition (HDCE) 框架 。在此框架下,本研究會先比較不同距離矩陣的層次聚類,透過凝聚指數可得到最優聚類數和最佳層次聚類方法。而動態時間規整在研究中被用來計算時間序列的相似度,此相似度即為距離矩陣的輸入值。在確定此歷史資料集的最優聚類數和最佳層次聚類方法後,因為基於概念假設每年的乾濕變化會造成最大的流量差距,所以選擇使用AMOC(最多一個變化)來分析歷史水文逕流時間序列的結構。對每個水文時間序列使用AMOC檢查最大變化的時間點。最後採用總體經驗模態分解法來判斷轉折時間點的趨勢。本研究利用所提出的框架,將這些方案應用於台灣主要的入流觀測站。在結果討論中本研究將討論翡翠、石門、曾文的逕流模式變化及各站點在轉折時間點變化的趨勢結果。根據本研究結果發現,台灣的逕流模式主要受到各站點的水文條件影響,而轉折點在時間上的趨勢會根據各站而有些許不同。基於本研究的概念假設,轉折點結果表明近年來翡翠的乾濕變化點有提前的趨勢,石門則是有延後的現象,曾文則發現有些許提前趨勢。本研究的結果不僅可用以檢視乾旱資訊,也可望有助於制定抗旱策略。 The analysis of streamflow patterns is crucial for effective water resources management. In recent years, climate change has resulted in more frequent and intense hydrological events, leading to unexpected consequences and economic losses. However, limited attention has been addressed to studying the changes in hydrological temporal patterns. This issue is particularly significant in Taiwan due to the uneven distribution of rainfall across different regions and time periods. To address this research gap, an HDCE (Hierarchical Clustering, Dynamic Time Warping, Change Point Detection, and Ensemble Empirical Mode Decomposition) framework is proposed. The framework incorporates various analytical techniques to examine streamflow patterns and identify turning points in hydrological time series data. Hierarchical clustering is utilized, comparing different distance matrices, to determine the optimal number of clusters. The agglomerative coefficient is used to guide this selection process. Dynamic time warping is employed to measure the similarity between time series, providing input values for the distance matrix. Once the optimal clusters are determined for the historical dataset, the AMOC (At Most One Change) approach is applied to analyze the structure of the hydrological streamflow time series. The AMOC assumes that the largest flow difference occurs during wet-dry changes each year. By applying the AMOC to each hydrological time series, the time points of the maximum change are identified. Last, empirical mode decomposition is utilized to identify turning points and determine their trends. The proposed framework is applied to major inflow observation stations in Taiwan, specifically focusing on Feitsui, Shihmen, and Tsengwen. The results and discussions center around streamflow patterns and trends of turning points at these locations. The findings suggest that streamflow patterns in Taiwan are primarily influenced by the hydrological conditions at each station, and the trends of turning points exhibit slight variations among stations. Based on the conceptual assumptions of this study, the identified turning points indicate an advancing trend in wet-dry changes at Feitsui, a lagging phenomenon at Shihmen, and a slight advancement at Tsengwen. These findings not only contribute to drought examination but also have the potential to inform the development of drought mitigation strategies. |
URI: | http://tdr.lib.ntu.edu.tw/jspui/handle/123456789/91477 |
DOI: | 10.6342/NTU202302035 |
全文授權: | 同意授權(限校園內公開) |
顯示於系所單位: | 土木工程學系 |
文件中的檔案:
檔案 | 大小 | 格式 | |
---|---|---|---|
ntu-111-2.pdf 目前未授權公開取用 | 6.95 MB | Adobe PDF | 檢視/開啟 |
系統中的文件,除了特別指名其著作權條款之外,均受到著作權保護,並且保留所有的權利。