- 期刊
結合自組織映射圖網路與支撐向量機於颱風期間水庫入流量預測之研究
Reservoir Inflow Forecasting During Typhoon Periods by Combining Self-Organizing Map with Support Vector Regression
摘要
臺灣由於地狹人稠，水資源一直都仰賴水庫供給，而水庫每年主要水量均是來自梅雨、颱風等降雨集中事件，尤其颱風降雨為在短延時中動輒降下數百釐米雨量，對於水庫儲水雖有其利，亦可能因庫容難以負荷，反而造成災害，所以在即使是短時間的入流預測，對於水庫入流及放流的掌握也是極其重要。本研究以翡翠水庫集水區入流量為例，進行25 場颱風降雨事件短延時水庫入流量預測。為能加強預測能力，運用非監督式的自組織映射圖網路(Self-Organizing Map，SOM)先行針對已降下的前期雨量與水庫入流量，以不同水庫入流歷程進行分類後，再利用支撐向量機(Support Vector Machines, SVM)於廻歸能力上的優勢，進行未來1 到3 小時水庫入流量的預測。結果顯示本研究所建立，先行經過分類後再行預測的SOM-SVM 模式，不僅在短延時確實造成25 場事件整體均方根誤差均下降，預測效率則使25 場事件有24 場都達到0.9 以上。而在1 到3 小時平均效率係數則分別提昇17.5%、22.7%及23.2%。研究成果有助於未來颱風期間水庫入流量之預測。
並列摘要
The reservoir inflow forecasting during typhoon periods is always an important task for water resources management and disaster mitigation. For improving the short leadtime flood forecasting performance, a reservoir inflow forecasting model is proposed in this paper. The observed inflows of Feitsui reservoir for 25 typhoon events are adopted. Firstly, the Self-Organizing Map (SOM) is used to classify the data into different regions relative for different inflow process. After the inputs are arranged into different regions, the Support Vector Regression (SVR) is implemented for determining the 1- to 3-h ahead inflow forecasts. The results show that the performance of the proposed model can provide improved forecasts of hourly inflows. In conclusion, the proposed model, which considers higher related inputs instead of all inputs, can generate better forecasts in different clusters. This proposed model is recommended as an alternative to the original SVR model because of its accuracy and efficiency.