應用決策樹於水庫防洪減淤 最佳操作規則之研究-以曾文水庫為例

The Application of Decision Tree to the Optimal Operation Rules of Reservoir Flood Control and Sediment Sluicing– A case of Tseng Wen Reservoir

張瀚文(Han-Wen Chang)

指導教授 ： 張良正

國立交通大學/工學院/土木工程系所/碩士(2014年)

https://doi.org/10.6842/NCTU.2014.00035

摘要

颱風所帶來的降雨是台灣主要的用水來源，但大量的降雨常帶來洪水災害，且因台灣山區地形高陡及地質結構鬆散，大量洪水除易造成土石流災害外，大量泥砂進入庫區，造成淤積而縮短水庫壽命。因此如何在颱洪時期藉由適當的操作手段，在維持水庫安全的前提下，除考量減洪以及供水外，盡量提高水庫的排砂量，乃為重要的議題。
有鑒於此，本研究目的乃以曾文水庫為例，整合遺傳演算法、水量與泥沙模擬及水庫操作規定等，發展水庫防洪減淤操作最佳規劃模式，並以此為基礎，進一步應用決策樹歸納出同時考量減洪、供水及排砂等之操作原則。其中水庫防洪減淤操作最佳規劃模式，乃在維護水庫壩體安全之前提下，遵守水庫防洪運用要點，並兼顧水資源供應及排砂等目標下，求得水庫在不同颱風場次下之最佳放水歷線。為進一步了解防洪放水原則及方便應用，本研究進一步合成多場颱風降雨，並考量不同的水庫起始和目標水位下之多種情境，應用最佳規劃模式計算出多組最佳防洪排砂放水歷線，再配合情境分類，歸納出多種類別之決策樹放水規則。
應用最佳操作規則與最佳規劃模式操作結果比較，以2007年的柯羅莎(Krosa)颱風為例，排砂量相較規劃模式操作增加1.4%，蓄水量約少640萬噸；以2008年的薔蜜(Jangmi)颱風為例，排砂量相較規劃模式操作少2.2%，而蓄水量約少2054萬噸。另外最佳操作規則與歷史實際操作比較中，以柯羅莎颱風為例，排砂量相較歷史操作可額外增加15.4%，蓄水量約少838萬噸；以薔蜜颱風為例，排砂量可額外增加24.4%，而蓄水量約少758萬噸。整體而言，本研究發展之決策樹防洪減淤最佳操作規則，可以快速判斷適當的水庫放流量外，亦能有效的提升水庫排砂量，延長水庫壽命以達永續使用之目的。

關鍵字

水庫防洪操作 ； 決策樹 ； 遺傳演算法

並列摘要

Typhoon events are major sources of water in Taiwan. However, extreme storm events could cause floods and damages due to fragile geological conditions and mountainous area with steep slope of terrain. The storm events not only cause debris flow but also bring huge amount of sediment into reservoirs which would reduce the life of reservoirs. Therefore, how to operate a reservoir during storm events considering multiple objectives, flood control, water supply, and sediment sluicing, is a critical issue.
This study develops an optimization planning model for reservoir operation during storm events considering flood control and sediment sluicing. In this optimization model, genetic algorithms, water and sediment simulation model, and reservoir operation rules are incorporated for realistic results. The result shows the optimal reservoir release hydrograph considering dam safety, operation rules, water supply, and sediment sluicing. In addition, decision tree is used to provide operation principals considering predefined scenarios which include various reservoir's initial storage, target storage, and synthesis storm events.
Compare the optimization planning model (OPM) with the decision tree operation rules (DTOR), the DTOR model produces 1.4% more of the sediment removal but 6.40×106 m3 less of reservoir storage than the OM model output for the event of Krosa Typhoon in 2007. For the event of Jangmi Typhoon in 2008, the DTOR model produces 2.2% less of the sediment removal and 20.54×106 m3 less of reservoir storage than the OPM model output. Compare the DTOR model with the historical record, the DTOR model produces 15.4% more of the sediment removal but 8.38×106 m3 less of reservoir storage than the historical record of the Krosa Typhoon; and 24.4% more of the sediment removal but 7.58×106 m3 less of reservoir storage than the historical record of Jangmi Typhoon. The results show that the proposed DTOR model can increase the sediment sluicing and the life of the reservoir.

並列關鍵字

Reservoir Flood Control ； Decision Tree ； Genetic Algorithm

參考文獻

25. 王國威，「運用懲罰機制遺傳演算法於水庫颱洪操作之規劃」，淡江大學水資源及環境工程學系，碩士論文，民國90年。

39. 張大元，「類神經網路在水庫放流對河川水位增量之研究」，中原大學土木學系，碩士論文，民國92年。

40. 張淑勤，「模糊決策在防洪之應用」，中原大學土木工程學系，碩士論文，民國95年。

56. 魏志強，「多水庫系統最佳防洪操作之研究」，台灣大學土木工程學系，博士論文，民國95年。

3. Ali Khosronejad “Optimization of the Sefid-Roud Dam desiltation process using a sophisticated one-dimensional numerical model”, International Journal of Sediment Research Volume 24, Issue 2, June 2009, Pages 189–200