透過您的圖書館登入
IP:18.205.67.119
  • 學位論文

利用支撐向量迴歸配合自組織網路分類與昇降二元分類於逕流量預測之研究

Streamflow Prediction Using Support Vector Regression Combined with SOM and Binary Classification

指導教授 : 陳儒賢
共同指導教授 : 林國峰(Gwo-Fong Lin)

摘要


臺灣地區由於地狹人稠,自然環境儲水空間有限,加之多山少平原的島嶼環境,即使位於多雨的東亞季風氣候地區,主要用水均來自降雨,只是降雨型態卻在時間、空間上的有極不均勻的分佈情形。因此,每年若能分別在短、中、長期時間長度內有效預測降雨,實為一活化臺灣地區水資源調度的課題。 本研究為有效強化長、中、短期降雨預測能力,分別就臺灣地區月流量及旬流量,選擇不同地區的5個(內灣站、彰雲橋站、崇德橋站、蘭陽大橋站及瑞穗大橋站)水位站,依次期流量昇降情形進行二元性的分類,使用支撐向量迴歸(Support Vector Regression)分別建構上昇、下降型態適用的月流量及旬流量強化預測模式,期能作為中、長期河川流量調度用水之參考。而對於短延時的時流量部份,由於臺灣地區的用水主要來源大多是颱風所挾帶豪大雨,經水庫有效蓄留後所用,因此本研究另以自組織映射圖網路學習配合支撐向量機,以翡翠水庫為測試地點,就不同颱風事件期間降雨進行預測能力強化。 其結果顯示經由二元性分類配合支撐向量迴歸模式所建構的月流量,確實均較傳統的自迴歸移動平均(ARIMA)模式及單純建構的支撐向量模式能有效掌握流量的昇降變化,整體在效率上的表現也確實較佳。而旬流量則承繼月流量模式所建構二元分類,加入影響各水位站的雨量資料後,以支撐向量迴歸進行預測模式建構,其成果不僅明顯較傳統的多變量分析優於相位差不易出現,且預測效率亦較直接由支撐向量迴歸為佳。期以本研究針對不同時間尺度的流量預測強化模式,提供往後水資源防災、蓄水、保水及開發參考。

並列摘要


The water in Taiwan are is mainly supplied from precipitation. However, the uneven spatial and temporal distribution of rainfall in Taiwan due to the mountainous terrain makes the rainfall difficult to predict. Therefore, it is an important task to make rainfall predictions at different time scales. This study tries to consider a binary classification between rise and decline of flow and then generate the flow predictions using Support Vector Regression (SVR) to improve both the monthly and ten-day flow predictions for five water level stations in Taiwan. Moreover, the self-organizing map (SOM) is combined with the SVR to improve the Feitsui reservoir inflow predictions during Typhoon periods. The results show that the monthly flow predictions generated form the combined SVR and binary classification can effectively capture the rise and decline of flow. Moreover, it is found that the combined SVR and binary classification could be used for ten-day flow predictions. The binary classification for SVR can yield better performance than ARIMA models, especially for the phase difference in ten-day predictions. In addition, the SVR combining with SOM can improve the reservoir inflow predictions. The proposed model is able to predict flows at different time scales, and it is recommended as an alternative to existing methods for water resources policy, flood prevention and decision making.

參考文獻


47.Yu, X. Y., Liong, S. Y. & Babovic, V. (2004). “EC-SVM approach for real-time hydrologic forecasting.” Journal of Hydroinformatics, 6(3), 209-223.
1.Alvarez-Guerra, M., González-Piñuela, C., Andrés, A., Galán, B. & Viguri, J. R. (2008). “Assessment of Self-Organizing Map artificial neural networks for the classification of sediment quality,” Environment International, 34(6), 782-790.
2.Box, G. E. P., and Jenkins, G. M. (1976). “Time Series Analysis: Forecasting and Control.” Holden-Day, San Francisco.
3.Baylar, A., Batan & M. (2010). “Usage of artificial intelligence methods in free flowing gated closed conduits for estimation of oxygen transfer efficiency,” Advances in Engineering Software, 41(5), 729-736.
4.Bovolo, F., Bruzzone, L. & Marconcini, M. (2008). “A novel approach to unsupervised change detection based on a semisupervised SVM and a similarity measure.” IEEE Transactions on Geoscience and Remote Sensing, 46(7), 2070-2082.

被引用紀錄


鄧學謙(2017)。不同型式滯洪池探討〔碩士論文,國立臺灣大學〕。華藝線上圖書館。https://doi.org/10.6342%2fNTU201701428

延伸閱讀