應用混合啟發式演算法推估地下水污染物暫態釋放問題

Abstract

當污染源的釋放歷程會隨時間改變時，污染源鑑定與歷程重建的工作將會變得較困難。隨著未知變數的增加，逆推的困難度也隨之增加。序的最佳化(Ordinal optimization)適用於解決複雜的最佳化問題，本研究利用此方法，結合模擬退火演算法、禁忌演算法、及旋轉輪盤法的優點，來處理地下水污染源暫態釋放的問題。本研究利用一個假設的污染廠址對發展方法的應用性作測試，分別探討七個及十五個未知數的問題：位置的三維座標X、Y、及Z、兩段及六段的釋放歷程與釋放濃度。利用MODFLOW-GWT地下水污染傳輸模式，可模擬監測井中污染物的濃度分布。在進行污染源鑑定與歷程重建的工作時，首先選取一污染源與鄰近格網所構成的可疑範圍，將範圍內的每一個格網視為候選污染源位置。利用禁忌演算法於候選區域中產生不同的候選污染源位置，再配合模擬退火演算法所產生的一系列釋放時間與釋放濃度的試誤解，可算得監測井中的汙染物濃度值。透過序的最佳化，可從候選格網中篩選出最好的前百分之五可能污染源位置，以縮小搜尋範圍；接著使用旋轉輪盤法，於其中挑選出下次迭代運算的污染源位置。目標函數設定為模擬濃度與實際濃度差值的平方和，當所得結果滿足收斂條件時，即視為得到最佳解。由所提出案例研究的分析結果顯示，本研究所提出的方法即使在污染物為暫態釋放的問題中仍然可得到很好的結果。

As the release of groundwater contamination source is a function of time, it will be very difficult to determine the source information such as the source location and source release history simultaneously. A method based on the ordinal optimization algorithm (OOA), simulated annealing (SA), tabu search (TS), roulette wheel approach, and MODFLOW-GWT is developed to determine the source release problem which contains at least fifteen unknowns including the location of three coordinates and six or more release periods with different concentrations. A hypothetic case for a contamination site is designed to test the applicability of the present method. In the identification process, the TS is first used to generate a candidate location within the block and SA is used to generate serial trial solutions with different release periods and concentrations. The plume concentrations at the monitoring wells can then be simulated and compared with the observed concentrations. To reduce the size of feasible solution space, the OOA is used to sift the top 5% candidate locations. Then the next location is chosen from them by the roulette wheel method. The optimal solution is obtained when the new result in the identification process satisfies the stopping criterion. The result of case study indicates that the proposed method is capable of estimating the source information even if the source release is in transient state.