請用此 Handle URI 來引用此文件:
http://tdr.lib.ntu.edu.tw/jspui/handle/123456789/8158
標題: | 應用支援向量回歸於河川洪水預警之研究 —— 以朴子溪爲例 Application of Support Vector Regression in Flood Forecast —— Taking the Puzi River as an Example |
作者: | TianYi Xue 薛天一 |
指導教授: | 駱尚廉(Shang-Lien Lo) |
關鍵字: | 洪水預警,水位預測,支援向量回歸,灰色模式,朴子溪, Flood Warning,Water Stage Forecast,Support Vector Regression,Gray Model,Puzi River, |
出版年 : | 2020 |
學位: | 碩士 |
摘要: | 臺灣屬於副熱帶溼潤氣候,夏季雨量充沛,河川逕流量較大。在2018年的823水災中,部分低窪地區遭遇了洪水及淹水問題,帶來負面影響,其中便包括了朴子溪流域的嘉義縣市。因此,本研究利用嘉義朴子溪流域過往之水位及降雨資料,建立支援向量回歸、灰色水位、線性回歸三種水位預測模式,並嘗試在不處理原始資料雜訊的前提下進行河川水位預測,以期判斷各模式是否可用於快速提供當地洪水預警。 本研究使用朴子溪流域牛稠溪橋(1)、竹崎、番路三個測站2017年一整年之水位、降雨資料率定模式,使用2018年三場降雨事件分別作驗證,其中包括了823水災。研究先確定洪峰延滯時間為3小時,並以此作爲預警時間。而後建立支援向量回歸、灰色水位、線性回歸三種模式分別預測三場降雨事件,以事件一爲例,支援向量回歸的決定係數、均方根誤差、最大絕對誤差、尖峰水位誤差均爲最佳,分別為0.9902、0.1754、0.9341、0.5713。再根據三場預測得到各糢式之總體決定係數,分別爲0.9741、0.9434、0.9560。經由對比三種模式後得到,各模式都可在一定程度上預測河川水位高度,其中以支援向量回歸為最優,其健壯性(Robustness)相對另外兩種模式較佳。 Due to the humid subtropical, Taiwan receives abundant precipitation throughout the summer, which result a high river discharge. In the 823 flood disaster, many low-lying areas experience flooding during heavy rains, including Chiayi county in Puzi River Basin. In this study, three models are utilized for forecasting: Support Vector Regression, Gray Model, Linear Regression. Values of input data was not denoised, in order to provide flood warning as fast as possible. The purpose of this study is to construct a quickly water stage forecasting model at Puzi River. In this study, water stage and precipitation of the 1550H017, C0M700, C0M720 station during 2017 is used to establish the models, and three rainfall events from 2018 is used to verification, including the 823 flood disaster. At first, determine lag time of 3 hours as early warning time. Then, establishing Support Vector Regression, Gray Model, Linear Regression model, forecasting three rainfall events. Take Event One as an example, the R^2 (Coefficient of determination), RMSE(Root mean square error), MAE (Maximum absolute error) and EPWL (Error of peak water level) of Support Vector Regression are the best, which are 0.9902, 0.1754, 0.9341 and 0.5713, respectively. In addition, the overall R^2 of each formula are 0.9741, 0.9434 and 0.9560, respectively. By comparing the three models, the water stage prediction model of Support Vector Regression is more effective in predicting river water level, and its robustness is stronger than the other two models. |
URI: | http://tdr.lib.ntu.edu.tw/jspui/handle/123456789/8158 |
DOI: | 10.6342/NTU202003785 |
全文授權: | 同意授權(全球公開) |
顯示於系所單位: | 環境工程學研究所 |
文件中的檔案:
檔案 | 大小 | 格式 | |
---|---|---|---|
U0001-1708202016053600.pdf | 2.51 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。