應用遺傳演算法於大甲溪及大安溪水資源最佳聯合調配模式

Abstract

本研究可分為兩個階段，第一階段在整合遺傳演算法（Genetic Algorithm）、線性規劃（Linear Programming）及水庫模擬等方法，發展一新的水資源最佳調配模式，並將其應用於大甲溪及大安溪之水資源最佳聯合調配模式，第二階段在以第一階段之調配模式為基礎，進一步加入可考量長、短期水文狀況的複合式規線之探討。 本研究之水資源最佳調配模式的演算核心為先以水庫模擬的精神決定各時刻系統所需之參數值，每個時刻再以線性規劃所建構的水源調配核心模式進行水量調配，系統考量了不同的供水標的，如農業、公共給水，以及河川生態基流量與農業用水迴歸水等，各標的用水的先後順序則以權重大小來衡量，此種整合方式不但保留了線性規劃的彈性及效率，並可模擬多水庫聯合規線操作，模式的最外層再以遺傳演算法進行規線的優選。在優選複合式操作規線時，模式先優選出一考量長期水文狀況的傳統操作規線，再以此傳統操作規線為基準，加入短期水文狀況（如枯水期等）的影響因子，另優選出一可同時考慮長、短期水文狀況之複合式規線。 本研究發展的模式並應用於大甲及大安溪兩流域的聯合調配，水源調配的研究成果充分顯示本模式的彈性及高效率，為一相當實用而一般化的水資源調配規劃及評估模式。另由複合式規線之探討可發現，同時考慮長、短期水文狀況之複合式規線相較於傳統的單一規線，其缺水指數較低，此說明本研究建議之複合式規線較傳統的單一規線確可更進一步降低缺水的風險。

This work develops a novel optimal planning model for the conjunctive use of water resources by integrating genetic algorithm, linear programming and reservoir simulation and, then, applying it to the Da-Chia and Da-An River. Hybrid rule curves are also developed for the reservoir operation while considering both the normal and low stream flow periods. The proposed planning model largely consists of a GA embedded with a simulation model. The simulation model computes the water distribution sequentially by using linear programming. This model can also consider multiple water demands, stream flow requirement and irrigation return flow. Weightings determine the priority of the water demands. In addition to simulating the operation of multiple reservoirs based on rule curves, the simulation model can also benefit the resilience and efficiency of the linear programming. On top of the simulation model, a genetic algorithm is applied to determine the optimal rule curves. Besides applying the model to study the conjunctive use of Da-Chia and Da-An Rivers, the planning model is also adopted to investigate the feasibility of applying a hybrid rule curves to reservoir operation. The hybrid rule curves extend the conventional rule and consider both the long and short periods of hydrologic conditions. Results of this study indicate that the novel planning model is not only reliable and efficient, but can also be applied to a large water resources system. Simulation results further demonstrate that the proposed hybrid rule curves can effectively decrease the shortage risk comparing to a conventional rule curve.