結合支援向量機與自組織映射於區域淹水預報

即時淹水預報是都市災害預警系統最重要的技術之一。然而，現行的區域淹水預報仰賴物理模型，如SOBEK、TOPMODEL及HecRAS等等，這些物理模式運算時間過長，無法即時預報淹水結果，因此無法直接應用於災害預警系統中。為了要解決這個問題，本研究提出定量降雨映射區域淹水模式，其結合SOBEK模擬成果、預報降雨、支援向量機與自組織映射，在颱風及暴雨期間可即時預報區域淹水區域。核心之概念為建立淹水拓撲圖以及根據預報之總淹水體積更新所選之淹水分布。研究區域選自台灣東北部每逢豪大雨或颱風就極易淹水之宜蘭縣，。方法包含五個主要之方向：(1)使用降雨繁衍之方法將15場歷史降雨事件繁衍為3000場降雨情境，(2)利用各區域特有的淹水特性將研究區域劃分為9個子區域，(3)使用自組織映射將大量的區域淹水圖分類成有著各自特性之拓撲圖，(4)建立多步階預報之支援向量機來預報各子區域中的總淹水體積，(5)基於預報之總淹水體積調整自組織映射中被選擇神經元之權重而後獲得及時修正之區域淹水圖。本研究提出之模式利用大量的淹水資料及降雨資料進行訓練及測試，且使用了方均根誤差、平均絕對誤差、相關係數、效率係數、召回率作為評鑑指標。結果顯示，(1)本研究所提出之預報模式能穩定地預報至少46小時後各個子區域之淹水總體積。除了第一子區域外，其他子區域在訓練階段以及測試階段相關係數皆不低於0.9。(2)本研究所使用的聚類演算法能有效地分類出不同的淹水型態，且在各子區域中平均淹水深度的均方根誤差及絕對平均誤差值皆不大於0.12 m以及0.03 m，(3)本研究所提出之模式能有效地模擬出淹水範圍，且在降雨較強的事件中表現更為優異；與淹水災情發生時之回報資料比較後可以發現，可發現到本模式平均上可以掌握實際有淹水區域的73 %。在未來面臨豪雨時，本研究提出之模型可以做為決策的參考，預防災害以及減少人命傷亡的損失。

關鍵字

區域淹水深度預報；機器學習；多步階預報；支援向量機；自組織映射

並列摘要

The real-time inundation forecasting is the most essential technique for the urban disaster early warning system. However, existing models for the regional flooding forecasting rely on physic-based models, such as SOBEK, TOPMODEL and HecRAS, which might cost too much operating time to be directly adopted in the early forecasting use. To overcome this problem, this study proposed Quantitative Precipitation Forecast mapping Regional-Inundation Forecasting model, which integrates simulation results of SOBEK, rainfall forecasting, Support Vector Machine and Self-organizing map. This model can forecast regional inundation area immediately during storms and typhoon periods. The core concept is to construct a topological map as distributions of inundation and switch distributions of inundation according to the forecasting result of total inundation volume. The study area is Yilan County in northeastern Taiwan, where floods usually occur during storms and typhoon periods. This study includes five stages: (1) using15 historical rainfall events to reproduce 3000 sets of simulated rainfall, (2) dividing the study area into 9 sub-regions according to its characteristic of inundation, (3) using Self-organizing map to cluster regional inundation map, (4) using Support Vector Machine Multi-steps forecasting to forecast total inundation volume in each sub-region and (5) adjusting the weights of neurons in Self-organizing map according to the forecast of total inundation volume and then adjusting the regional inundation map. Moreover, Root Mean Square Error (RMSE), Mean Absolute Error (MAE), Coefficient of Efficiency (CE), Coefficient of Correlation (CC) and True-Positive Rate (TPR) are determined to evaluate the model performance. The results first indicate that the proposed model can effectively forecast total inundation volume in each sub-region at least 46 hours ahead. Except the first sub-region, value of CC is higher than 0.9 during training phase and test phase in the rest of sub-region. Second, the cluster algorithm used in this study can effectively classify inundation maps into different types of inundation, and the values of RMSE and MAE in mean inundation depth are less than 0.12 m and 0.03 m, respectively. Third, the proposed model can efficiently simulate the area of inundation, especially in heavier rainfall events. Comparing with the flooding from public reports, this model can capture about 73% of real flooding area on average. In the future, the proposed model can be adopted as a reference for disaster prevention.

並列關鍵字

regional inundation depth forecast ； machine learning ； multi-step forecasting ； support vector machine ； self-organizing map

參考文獻

1. Aggarwal, S.K., Goel, A., Singh, V.P., 2012. Stage and discharge forecasting by SVM and ANN techniques. Water Resources Management 26, 3705–3724. doi:10.1007/s11269-012-0098-x

Google Scholar

2. Alfieri, L., Burek, P., Dutra, E., Krzeminski, B., Muraro, D., Thielen, J., Pappenberger, F., 2013. GloFAS – global ensemble streamflow forecasting and flood early warning. Hydrology and Earth System Sciences 17, 1161–1175. doi:10.5194/hess-17-1161-2013

Google Scholar

3. Bates, P.D., Stewart, M.D., Siggers, G.B., Smith, C.N., Hervouet, J.M., Sellin, R.H.J., 1998. Internal and external validation of a two-dimensional finite element code for river flood simulations. Proceedings of the Institution of Civil Engineers - Water Maritime and Energy 130, 127–141. doi:10.1680/iwtme.1998.30972

Google Scholar

4. Bates, P.D., Horritt, M.S., Fewtrell, T.J., 2010. A simple inertial formulation of the shallow water equations for efficient two-dimensional flood inundation modelling. Journal of Hydrology 387, 33–45. doi:10.1016/j.jhydrol.2010.03.027

Google Scholar

5. Bender, M.A., Tuleya, R.E., Kurihara, Y., 1987. A numerical study of the effect of island terrain on tropical cyclones. Monthly Weather Review 115, 130–155. doi:10.1175/1520-0493(1987)115<0130:ansote>2.0.co;2

Google Scholar

國際替代計量

結合支援向量機與自組織映射於區域淹水預報

未授權

主題瀏覽