Coupled Global-Local Optimization Scheme for a Conjunctive-Use Project in a Desert Environment
論文發表人: 邱永嘉 (加州大學洛杉磯分校土木與環境工程研究所博士班)
為減低因地方住戶化糞池產生的硝酸鹽入滲污染地下水系統，加州Yucca Valley的 Hi-Desert Water District (HDWD) 計畫建造一座新的污水處理廠用以解決此問題。經過污水處理廠處理過之廢水，將會注入一個人工濕地用以補注地下水。此設計歸類為地表水地下水交互使用之策略規劃。HDWD 希望可以尋找出最佳之地表水地下水交互使用策略，寄望有效控制地下水位，找出最佳的抽水井位置，同時符合加州州政府對於地下水補注之法規。
此研究，我們研發連結優化模型與數值模型的一個整合模型，此模型可找出最佳之地表水地下水交互使用的最佳策略。最佳策略的目標函數為最低之水資源供應成本，而限制條件則為，新抽水井之位置，抽水量之大小，可使用之地下水補注量，公共用水之供應量，以及地下水位高程之控制。根據設計，我們的優化模型將是一個非線性整數混合規劃模型。傳統的數值解法分為兩種，一為梯度法，二為全面性演算法。梯度法的計算速度非常快速，但容易陷入局部的最佳解，全面性演算法可以找出真正的最佳解，但計算速則度非常換緩慢。此研究，我們將連結全面性演算法與梯度法來找出真正的最佳解。連結全面性演算法與梯度法中，全面性演算法將使用基因演算法 (Genetic Algorithm) 找出接近真正最佳解的近似解，而後，再根據此近似解，用梯度法找出真正的最佳解。
To reduce the quantity of septic-tank nitrate from reaching the ground-water system, Hi-Desert Water District (HDWD) in Yucca Valley, CA is planning to construct a wastewater treatment plant. The treated wastewater will be recharged to the ground-water basin via percolation ponds as part of a larger conjunctive-use strategy, subject to state regulations (e.g. minimum distance, travel time). HDWD wishes to identify conjunctive-use strategies that control ground-water levels, meet regulations, and identify new production-well locations.
In this study we develop a simulation-optimization model that identifies optimal conjunctive-use strategies that minimizes water-delivery costs subject to constraints including pump capacities, available recharge water, water-supply demand, water-level constraints, and new-well locations. As formulated, the optimization problem is a mixed-integer, non-linear programming problem. Conventional optimization-solution methods include gradient-search and global-optimization schemes. Gradient-search schemes are computationally fast; however, they may only identify local optima. Global-optimization schemes identify the global optimum; however, they are computationally slow. In this work, we couple a global-optimization scheme (genetic algorithm or GA) with a gradient-search. The GA identifies a near-optimal solution and the gradient-search scheme uses the near optimum to identify the global optimum. Results indicate the global-local scheme is faster than GA alone and the global optimum can be identified.
In this study we develop a simulation-optimization model that identifies optimal conjunctive-use strategies that minimizes water-delivery costs subject to constraints including pump capacities, available recharge water, water-supply demand, water-level constraints, and new-well locations. As formulated, the optimization problem is a mixed-integer, non-linear programming problem. Conventional optimization-solution methods include gradient-search and global-optimization schemes. Gradient-search schemes are computationally fast; however, they may only identify local optima. Global-optimization schemes identify the global optimum; however, they are computationally slow. In this work, we couple a global-optimization scheme (genetic algorithm or GA) with a gradient-search. The GA identifies a near-optimal solution and the gradient-search scheme uses the near optimum to identify the global optimum. Results indicate the global-local scheme is faster than GA alone and the global optimum can be identified.
張貼者：駐洛杉磯台北經濟文化辦事處教育組 於
上午10:23
