本研究之目的在於研發一套水庫防洪最佳即時操作模式,以供水庫操作人員在颱洪來臨時,決定水庫最佳即時放水量,以減少水庫下游發生水患之機會,並於颱風過後貯蓄充足之水量以供未來之用。 本研究之水庫防洪最佳即時操作模式,主要係由類神經網路之BP水庫入流量預報模式及ANFIS水庫放水量決策模式所構成。其中ANFIS水庫放水量決策模式係以水庫防洪操作優選模式所優選出之水庫放水歷線做為輸出項,並透過無母數統計分析方法進行相關性分析,選取相關性較高者做為輸入項,以進行ANFIS模式之訓練和驗證工作。 本研究將所建立之模式應用於曾文水庫系統的兩個颱風,結果顯示本模式於防洪操作結束後相較歷史操作之蓄水體積更接近水庫運轉規線之上限,並且颱洪期間下游控制點之洪峰流量小於歷史之洪峰流量,由此可知本研究所發展出的水庫防洪最佳即時操作,相較於歷史操作確實有較佳之成果。
The purpose of this study is to invent an optimal real-time operation model for reservoir flood control. This model can help the reservoir operator to decide the optimal releasing water at once for reducing the opportunity of flood loss, and retains sufficient water in the reservoir for supplying people after typhoon period. This optimal real-time operation model is constructed from BP inflow forecasting model and ANFIS outflow deciding model. For training and verifying ANFIS model, the input layer vectors are influent factors determined from higher Spearman’s Rank correlation coefficient on outflow, and the output layer vectors are the outflow hydrograph values decided by reservoir flood control optimization model. The optimal real-time operation model is applied to Zengwun Reservoir system with two typhoon events. Compare the ending water level of reservoir after typhoon period and peak discharge of downstream control point with the historical results, the water level and peak discharge of optimal real-time operation model are closer and smaller than historical results, respectively. Therefore the deciding process of optimal real-time operation model is better than historical operation results.