曾文水庫即時最佳防洪操作之研究

Abstract

台灣每年遭受颱風侵襲之機率相當高，而其中洪水災害更是主要的災損之一，而大型水庫可降低洪峰減少洪災，惟其有賴於適當的防洪操作。本研究目的乃以曾文水庫為研究區域，整合遺傳演算法、河道模擬與類神經網路等模式，發展水庫即時最佳防洪操作模式。前述之即時最佳防洪操作模式包含水庫最佳防洪操作模式與水庫入流量預測模式兩部份;其中水庫最佳防洪操作模式，乃在入流量已知之前提下，兼顧水庫壩體安全、降低下游地區淹水損失，並考量水庫防洪運用要點的最佳防洪操作模式;水庫入流量預測部份，則發展新的入流量預測模式，其入流量預測，乃以當下時刻前之觀測入流量，從颱風案例資料庫中，選取最吻合之歷史颱風事件，並以此歷史事件之未來時刻流量，作為未來入流量之預測值，因此每一時刻皆隨新的觀測資料的加入，而更新未來之入流量預測。 本研究目前已收集40場歷史颱風入流量資料，並選定2007年之聖帕、科羅莎颱風，以及2008年之卡玫基、鳳凰、辛樂克、薔蜜颱風等，進行即時操作模擬。由結果發現本研究所發展之水庫即時最佳防洪操作模式，相對於歷史操作結果，對減少下游地區之淹水延時較為顯著，以各場颱風為例，對科羅莎颱風可減少下游地區之溢堤延時4小時、對辛樂克颱風可減少3小時溢堤延時。整體而言，模擬結果顯示本研究所發展之即時最佳防洪操作模式具相當之實用性，可做為水庫即時防洪操作之輔助。

Taiwan has high risk of flood damage due to frequent typhoon events. A reservoir with proper reservoir operations can reduce the peak flow during a flood event. This study develops a real-time optimal flood control model using Genetic algorithms, a river simulation model, and an Artificial Neural Network model. This model includes two parts: a reservoir inflow forecast model and a reservoir optimal flood control model. The forecast model predicts the reservoir inflow based on the real-time discharge observations and the historical record of typhoon events. Whenever the discharge observation is available, the inflow forecast is updated using the observation and the data of the best fit historical typhoon event. The real-time flood control model optimizes the flood control operation of the reservoir using the forecast inflow and the Genetic algorithms. The objective function is to minimize the loss of the flood damage at the downstream area and subject to the dam safety and the reservoir operation guidelines during the flood events. This study applies the developed methodology to Tseng-wun Reservoir. Forty typhoon events are collected as the historical database. Six typhoon events are used to verify the proposed model. These typhoons include Typhoon SEPAT and Typhoon KORSA in 2007 and Typhoon KALMAEGI, Typhoon FUNG-WONG, Typhoon SINLAKU and Typhoon JANGMI in 2008. The results show that the proposed model can reduce the flood duration at the downstream area. For example, the real-time flood control model can reduce the flood duration by 4 hours, 3 hours for Typhoon KORSA, Typhoon SINLAKU respectively. This result shows that the developed model can be a very useful tool for real-time flood control operation of reservoirs.