透過您的圖書館登入
IP:44.204.65.189
  • 學位論文

颱風期間考慮取水濁度限制下水庫防洪與防淤之最佳即時操作

Real-time Reservoir Optimal Operation for Flood and Sedimentation Control Considering Turbidity Constraints during Typhoon Invasion

指導教授 : 徐年盛

摘要


本研究目的為研發水庫多層放水系統聯合操作決策模式,以決定颱風期間考量下游取水濁度限制、水庫防洪與防淤條件下之最佳即時放水量。本研究可分為決策模式建構與即時操作模擬兩大部分,決策模式建構由模擬分析法與操作歷線優選所組成;模擬分析之目的為應用流體動力泥砂濃度模擬模式(USEPA WASP-based)配合本研究所設計理想放水操作型式之萃取方法,以搜尋出優選時之最佳初始解,並可做為建構嵌入優選模式中之泥砂濃度模擬替代模式(ANFIS-based 與RTRLNN-based)之用;本研究優選模式採用禁忌演算法求解,優選所得到之最佳操作歷線將做為建構決策模式之用。短時距與總入庫流量預報模式(ANFIS-based與RTRLNN-based)、總雨量預報模式為即時操作決策模式所包含之預報模式,上述預報資訊輸入決策模式後,配合氣象局所預報未來颱風移動路徑來判斷目前至各防洪階段之延時,以進行水庫多層放水之即時推估,最後模擬操作成效以評判操作模式之優劣。 本研究將所發展之方法應用於石門水庫後可獲致下列結果:(1)本研究所研發總入庫流量預報模式在輸入項擁有未來總降雨量與颱洪延時兩關鍵性輸入項,並配合其餘因子之綜合模擬下,RTRLNN可相對於ANFIS預報得到較佳準確性與穩定性之總入庫流量;(2) ANFIS對於誤差容忍與調適的能力較高,可優於RTRLNN地模擬出泥砂傳輸之機制與特性;(3) RTRLNN由於擁有即時回饋式的定率演算機制,其對於少樣本的資訊與機制推衍能力較強,因此RTRLNN在多層放水操作推估之成效較ANFIS為佳;(4)本研究耦合所研發之預報模式於智慧型操作決策模式中以進行鳳凰颱風與莫拉克颱風時石門水庫之即時操作測試,輸入項包括短時距及總入庫流量預報資訊、目前蓄水量至最大蓄水量及目標滿水量之差值、目前時刻至各防洪階段之洪水延時、即時觀測到之降雨量、入庫流量、水庫各層放水口及下游鳶山堰之泥砂濃度、下游河道三鶯橋水位、上時刻水庫各層之放水量,輸出項則為水庫各層之即時放水量,結果顯示:應用本研發模式操作下,整場颱洪事件之評比指標:鳶山堰最高泥砂濃度、排砂比、三鶯橋最高水位以及水庫最終蓄水位皆較歷史操作為佳。

並列摘要


This study develops two multi-layer reservoir conjunctive release operation models (RTRLNN-based FSC model and ANFIS-based FSC model) to determine the optimal real-time releases during typhoon invasions for the Shihmen Reservoir basin in Taiwan, taking into consideration turbidity constraints, flood control, and sedimentation control. The study can be divided into model construction and real-time operational simulation. The models consist of a forecast model component, an optimization model component, and an operational decision model component. The decision model is composed of the developed simulation-analysis approach and optimization of the operation hydrograph. The purpose of the simulation-analysis approach is to apply the US EPA WASP-based fluid dynamic sediment concentration simulation model with the developed extracting method of ideal releasing practice to search for the initial solution of optimization and construct the sediment concentration simulation models (ANFIS-based and RTRLNN-based) embedded in the optimization model. The optimization model is solved by tabu search and the optimized releasing operation hydrograph is used for the construction of the decision model. The developed short lead-time and total reservoir inflow forecast models (ANFIS-based and RTRLNN-based) and the real-time revised quantitative precipitation forecast model based on typhoon central location (RTR-TCL-QPF) are embedded in the operational decision model. The forecasted typhoon track information and the duration between the current time and each flood control stage from the Taiwan Central Weather Bureau (CWB) are entered into the operation model for release decision-making. Finally, this study assesses the quality of the decision model according to the outcome of real-time operation. After applying the operational decision model on the Shihmen Reservoir basin, the results can be concluded as follows: (1) Under the synthesized simulation using key input of future total precipitation and flooding duration with other hydrometeorological factors, the accuracy and stability of the RTRLNN-based total reservoir inflow forecast model is better than the ANFIS-based model; (2) Error toleration and adaption are superior in ANFIS, therefore, the ANFIS-based sediment concentration model better simulates the mechanisms and characteristics of sediment transport than RTRLNN; (3) RTRLNN owns a real-time recurrent deterministic routing mechanism, allowing for better simulation under fewer data samples. Hence, the multi-layer release operation outcome evaluated by RTRLNN-based FSC model is better than ANFIS-FSC model; (4) This study couples the developed forecast model into the decision model to proceed with real-time operation of the Shihmen reservoir during Typhoon Fung-Huang and Typhoon Morakot. The inputs of the decision model include short lead-time, total reservoir inflow, the difference between current storage and maximum storage, the difference between current storage and target full storage, flooding duration between current time and each flood control stage, and real-time observed precipitation, reservoir inflow, sediment concentration of each releasing outlet and Yuan-Shan weir, downstream channel water level at Shan-Yin bridge, and multi-layer reservoir release. The model outputs are real-time evaluated multi-layer reservoir release. Results show that the operational outcome of the developed decision model is better than the historical operation according to the four assessment index: maximum sediment concentration of Yuan-Shan weir, sediment removed ratio, highest water level at Shan-Yin Bridge, and final stored reservoir water level.

參考文獻


蕭閔鍾,「類神經網路應用於颱風時期集水區流量預測之研究」,碩士論文,淡江大學水資源及環境工程學研究所,台北(2012)。
許弘政,「模糊控制模式於颱洪時期水庫即時操作之研究」,碩士論文,淡江大學水資源及環境工程研究所,台北(2007)。
鄭家豪,「改良式倒傳遞類神經網路於水庫入流量預報之研究」,碩士論文,國立台灣大學土木工程學研究所,台北(2008)。
張家銓,「改良式自組織映射線性輸出模式於水庫入流量預報之研究」,碩士論文,國立台灣大學土木工程學研究所,台北(2009)。
林桂宏,「探討石門集水區降雨-逕流機制與類神經網路洪水預報模式」,碩士論文,國立台灣大學生物環境系統工程學研究所,台北(2012)。

延伸閱讀