應用支援向量機於颱風雨量及洪水預報

Typhoon Rainfall and Flood Forecasting Using Support Vector Machine

周揚敬(Yang-Ching Chou)

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(2013年)

https://doi.org/10.6342/NTU.2013.02408

摘要

本論文的主要目標為將支援向量機應用於洪災消減及災害預警上，主要可分為以下兩個部分：
當颱風來襲時，雨量預報在大部分災害預警系統中皆扮演了非常關鍵的角色。為了能更快速的得到準確的降雨預報，各個防災單位總是積極研發各種新式的預報模式。本研究提出一種稱為支援向量機(support vector machine, SVM)的類神經網路，並以此為基礎架構有效的颱風時雨量預報模式。相較於傳統上較常被使用的倒傳遞類神經網路，基於統計學習理論的支援向量機具有三項優勢。第一、支援向量機具備了更佳的學習能力(generalization ability)，第二、支援向量機在架構和權重的決定上保證有唯一解並且為全域最佳解，最後、支援向量機大量減少架構模式所需的訓練時間。本研究以實際案例來說明支援向量機所具備的優勢。研究結果顯示支援向量機相較於倒傳遞類神經網路不但能得到更加準確的預報結果，並且有更佳的強健性，其中最大的優勢是能大幅的縮短架構模式所需的時間。除了模式間的比較，為了能進一步提升長期預報的準確度，本研究更是新增了颱風因子做為降雨預報模式的輸入項，並與沒加入颱風因子做為輸入項的模式進行比較，以探討颱風因子對於雨量預報的影響。研究結果也證明了颱風因子可以有效提升中長期預報的準確度。總結來說，本研究提出以支援向量機為基礎納入颱風因子做為輸入項的預報模式確實能提升颱風時期雨量預報的準確度。而本模式亦預期能為洪水預報、土石流警戒等災害預警系提供幫助。
對於洪水預警來說準確的流量預報是非常重要的關鍵。因此在第二階段的研究中提出一個以支援向量機為基礎，整合型的洪水預報模式來提升洪水預報的準確度。整合型的洪水預報模式可以分為兩個部分，雨量預報單元及流量預報單元。在第一階段，將以雨量及颱風因子作為輸入項發展雨量預報單元。接著將預報雨量及觀測流量作為輸入項發展流量預報單元。為了驗證整合型洪水預報模式的能力，本研究另外架構了直接納入觀測流量、雨量及颱風因子的洪水預報模式進行比較。並以實際發生的颱風事件作為研究案例並預報未來1至6小時的流量。研究結果顯示第一階段的雨量預報單元可以得到合理的預報結果。而將此預報雨量納入輸入項的整合型洪水預報模式，相較於直接納入各因子的洪水預報模式能得到更為準確的預報結果，甚至連尖峰流量亦有顯著的改善。值得注意的是，本研究提出的模式更是顯著的提升了中長延時的預報準確度。歸納結果，本研究提出的模式有效的減少了輸入項和輸出項間，隨著預報時間延長所帶來的負面影響，因此才能在中長延時仍能維持一定的準確度。而此一優勢將對於提升颱風時期洪水預警的反應時間有所幫助。

關鍵字

雨量預報 ； 洪水預報 ； 支援向量機 ； 颱風因子 ； 災害預警系統

並列摘要

The objective of this dissertation is to apply support vector machine for flood mitigation and disaster warning. There are two major parts in this paper, which are summarized in the following manner.
Typhoon rainfall forecasting plays a critical role in almost all kinds of disaster warning systems during typhoons. To obtain more effective forecasts of hourly typhoon rainfall, novel models with better ability are desired. Based on support vector machines (SVMs), which is a kind of neural networks (NNs), effective hourly typhoon rainfall forecasting models are constructed. As compared with back-propagation networks (BPNs) which are the most frequently used conventional NNs, SVMs have three advantages: (1) SVMs have better generalization ability; (2) the architectures and the weights of the SVMs are guaranteed to be unique and globally optimal; (3) 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. To further improve the long lead-time forecasting, typhoon characteristics are added as key input to the proposed models. The comparison between SVM-based models with and without typhoon characteristics confirms the significant improvement in forecasting performance due to the addition of typhoon characteristics for long lead-time forecasting. The proposed SVM-based models are recommended as an alternative to the existing models. The proposed modeling technique is also expected to be useful to support reservoir operation systems and flood, landslide, debris flow, and other disaster warning systems.
Accurate runoff forecasts are required to provide early warning of impending floods. In this part, an integrated flood forecasting model based on the support vector machine (SVM) is proposed to improve the flood forecasting performance. In the first stage, the observed typhoon characteristics and rainfall are used to produce rainfall forecasts. Then the forecasted rainfall and observed runoff are used to yield runoff forecasts. An actual application is performed to yield 1- to 6-h lead time runoff forecasts. The results show that the rainfall forecasting in the first stage can generate reliable rainfall forecasts, and the proposed model can provide accurate runoff forecasts, especially for the peak values. It is worth noting that the proposed model can significantly improve the 4- to 6-h lead time flood forecasting performance. In conclusion, the proposed model effectively mitigates the negative impact of increasing forecast lead time and is useful to improve the long lead time flood forecasting during periods of typhoon.

並列關鍵字

rainfall forecasting ； flood forecasting ； support vector machines ； typhoon characteristics ； disaster warning systems

參考文獻

ASCE Task Committee on Application of Artificial Neural Networks in Hydrology, 2000a. Artificial neural networks in hydrology. I: preliminary concepts. J. Hydrol. Eng. 5 (2), 115-123.

ASCE Task Committee on Application of Artificial Neural Networks in Hydrology, 2000b. Artificial neural networks in hydrology. II: hydrologic applications. J. Hydrol. Eng. 5 (2), 124–137.

Asefa, T., Kemblowski, M., Lall, U., Urroz, G., 2005. Support vector machines for nonlinear state space reconstruction: Application to the Great Salt Lake time series. Water Resour. Res. 41, W12422, doi: 10.1029/2004WR003785.

Chang, F.J., Chang, K.Y., Chang, L.C., 2008. Counterpropagation fuzzy-neural network for city flood control system. J. Hydrol. 358 (1-2), 24-34.

Chang, F.J., Chiang, Y.M., Chang, L.C., 2007. Multi-step-ahead neural networks for flood forecasting. Hydrol. Sci. J.-J. Sci. Hydrol. 52 (1), 114–130.