支持向量機於颱洪時期流量預報之研究

Flood Forecasting during Typhoon Periods Using Support Vector Machines

黃珮瑜(Pei-Yu Huang)

指導教授 ： 林國峰

國立臺灣大學/工學院/土木工程學研究所/博士(2009年)

https://doi.org/10.6342/NTU.2009.01421

摘要

於颱風襲台期間，流量預報模式是洪水預警工作中相當重要的一環，如何快速地建立一個準確且穩定的流量預報模式，以及如何增加較長延時預報的準確度以增加防汛的反應時間，一直以來都是研究的重點項目。為達到此一目標，本研究應用一種新的類神經網路—支持向量機(support vector machine, SVM)建立預報未來一至六小時的流量預報模式，並與一般研究中最常被應用的倒傳遞類神經網路(back-propagation network, BPN)進行比較。基於統計理論，SVM主要有三項優點，第一，SVM有較佳的歸納衍生(generalization)能力，一般能得到較好的結果；第二，SVM的參數及架構是以求解一個二次規劃問題而得，具唯一最佳解；第三，SVM的訓練速度快。
本研究藉由BPN與SVM於預報準確度、模式強健性以及架構模式的時間(即效率)等三個方面的比較，清楚呈現SVM的三項優點。結果顯示，在準確度部分，SVM不論預報時間長短，其預報準確度均高於BPN，顯示SVM的歸納衍生能力較佳；在強健性部分，BPN需使用試誤法挑選最佳的隱藏層神經元個數，且以迭代法求最佳解的方式會受初始權重影響而影響結果表現， SVM的參數及架構則具最佳且唯一性，沒有上述問題，因此SVM模式的強健性高於BPN；在效率的部分，經過實際測試，SVM架構最佳模式的時間僅有BPN所需時間的1/10000，大大提高了架構模式的效率。而經此三方面比較，證實SVM不論在模式強健性、效率或是預報準確度上，其表現均優於BPN。因此，本研究建議以SVM取代BPN，成為架構流量預報模式的一種更好選擇。
除了以SVM替代BPN之外，本研究進一步利用有效颱風因子作為模式輸入項，以提昇長期預報的效果。比較輸入項包含有效颱風因子與未包含有效颱風因子的模式，顯示BPN無法從有效颱風因子中萃取出對洪流預報有幫助的資訊，而SVM萃取資訊的能力較佳，能經由增加有效颱風因子提升預報準確度，且提昇的幅度有隨著預報延時越長，提昇越多的趨勢，證實有效颱風因子能幫助提昇較長延時預報準確度。此外，在缺少雨量資料的情況下，顯示於模式中加入有效颱風因子，能使提昇的幅度更大，說明當資料有所缺乏時，更需要有效颱風因子使替代模式維持一定的預報準確度。綜上所述，可使用SVM快速地建立一個準確且穩定的流量預報模式，並可利用有效颱風因子增加較長延時預報的準確度，使颱洪期間能有一個預報更準確，能提供更長的延時預報資訊的預報模式，以期提供防洪預警更多有利的訊息。

關鍵字

颱洪預報 ； 支持向量機 ； 颱風因子 ； 強健性與效率 ； 交替驗證

並列摘要

In this paper, effective flood forecasting models based on the support vector machine (SVM), which is a novel kind of neural networks, are proposed. Based on statistical learning theory, the SVM has three advantages over back-propagation network (BPN), which is the most frequently used neural network. Firstly, SVM has better generalization ability. Secondly, the architecture and the weights of the SVM are guaranteed to be unique and globally optimal. Finally, SVM is trained much more rapidly. An application is conducted to clearly demonstrate these three advantages. The results indicate that the proposed SVM-based models are more well-performed, robust and efficient than the existing BPN-based models. In addition to using SVM instead of BPN, typhoon characteristics, which are seldom regarded as key input for flood forecasting, are added to the proposed models to further improve the long lead-time forecasting during typhoons. A comparison between models with and without typhoon characteristics is also presented to confirm that the addition of typhoon characteristics significantly improves the forecasting performance and the improvement increases with increasing lead-time, especially when the rainfall data are not available. In conclusion, the typhoon characteristics should be used as input to the flood forecasting. The proposed SVM-based models are recommended as an alternative to the existing models because of their accuracy, robustness and efficiency. The proposed modeling technique is expected to be useful to improve the flood forecasting.

並列關鍵字

flood forecasting ； support vector machines ； typhoon characteristics ； robustness and efficiency ； cross-validation

參考文獻

Abrahart R.J., See L.M., “Neural network modelling of non-linear hydrological relationships”, Hydrology and Earth System Science, 11(5): 1563-1579, 2007.

ASCE Task Committee on Application of Artificial Neural Networks in Hydrology, “Artificial neural networks in hydrology. I: preliminary concepts”, Journal of Hydrologic Engineering, 5(2): 115-123, 2000a.

ASCE Task Committee on Application of Artificial Neural Networks in Hydrology, “Artificial neural networks in hydrology. II: hydrologic applications”, Journal of Hydrologic Engineering, 5(2): 124-137, 2000b.

Bae, D.H., Jeong, D.M., Kim, G., “Monthly dam inflow forecasts using weather forecasting information and neuro-fuzzy technique”, Hydrological Sciences Journal, 52(1): 99-113, 2007.

Cameron, D., Kneale P., See, L., “An evaluation of a traditional and a neural net modeling approach to flood forecasting for an upland catchment”, Hydrological Processes, 16(5): 1033–1046, 2002.