This paper focuses on the multi-objective optimization of the groundwater extraction scheme in the Bouein-Myandasht aquifer (Iran) in order to reduce the concentration of nitrate, originating from agricultural activities and wastewater absorbent wells. A simulation-optimization model coupling an artificial neural network (ANN) as the simulator with the non-dominated sorting genetic algorithm-type II (NSGA-II) as the optimizer, are employed. The simulator is trained by help of data generated by process-based simulation models for groundwater flow (MODFLOW) and solute transport (MT3D). The optimization objectives include (1) minimizing the contamination concentration and (2) maximizing the net benefit of the agricultural activities. The outcome of the simulation-optimization model is an optimized management strategy formed by the optimal values of the optimization parameters searched and obtained consisting of (1) seasonal groundwater extraction volume; (2) the ratio of the wastewater which should be treated before being leached into the groundwater through the absorbent wells; (3) the ratio of the fertilizers consumption; and (4) the cultivated area for each of the main crops in the study area. The results of the model suggest a groundwater extraction policy fulfilling the objectives of the optimization. The optimal operating policy also indicates that a partly conflicting relation exists between minimizing the risk of groundwater contamination and maximizing the net benefits of the agricultural activities. Hence, the focus of this paper is at finding the better and better Pareto-fronts in the objective space while dealing with the parts of the objective functions with less conflict to reach the optimal Pareto-front on which the full conflict between the objectives is held. Then, an entropy-based trade-off reflected in designating a couple of weights assigned to the couple of objectives calculated for each solution in the bi-objective space is held over the solutions lying on the optimal Pareto-front and finally, the favorite solution minimizing the weighted-distance to the ideal point in the objective space is achieved using the TOPSIS method. With this policy the regional nitrate concentration will be decreased by 36.7%, 20.45% and 21.6% in the first, second and third study sub-areas, respectively, as compared to those in the actual operation. Furthermore, the model suggests 15%, 12% and 9% wastewater treatment and also 9%, 6% and 7% decrease in the fertilizer use in the first, second, and third study sub-areas, respectively.

本文重点研究 Bouein-Myandasht 含水层（伊朗）地下水提取方案的多目标优化，以降低源自农业活动和废水吸收井的硝酸盐浓度。模拟优化模型耦合人工神经网络(ANN) 作为模拟器与非支配排序遗传算法II (NSGA-II) 作为优化器，被采用。模拟器是在基于过程的地下水流 (MODFLOW) 和溶质运移模拟模型生成的数据的帮助下进行训练的(MT3D)。优化目标包括（1）最小化污染浓度和（2）最大化农业活动的净收益。模拟优化模型的结果是由搜索和获得的优化参数的最优值形成的优化管理策略，包括（1）季节性地下水抽取量；(2) 废水经吸水井浸入地下水前应处理的比例；（三）化肥用量比例；(4) 研究区主要农作物的种植面积。该模型的结果表明了满足优化目标的地下水提取策略。并使农业活动的净收益最大化。因此，本文的重点是在目标空间中寻找越来越好的帕累托前沿，同时处理目标函数中冲突较少的部分，以达到目标之间完全冲突的最优帕累托前沿。 . 然后，基于熵的权衡反映在指定分配给为双目标空间中每个解决方案计算的几个目标的几个权重上，这些权重被保持在位于最佳帕累托前沿的解决方案上，最后是最喜欢的解决方案使用 TOPSIS 方法可以最小化到目标空间中理想点的加权距离。实施该政策后，第一、二、三研究分区的区域硝酸盐浓度将分别下降36.7%、20.45%和21.6%，分别与实际操作中的比较。此外，该模型建议在第一、第二和第三个研究子区域分别减少 15%、12% 和 9% 的废水处理以及 9%、6% 和 7% 的肥料使用量。