| This dissertation explores the optimal design of groundwater quality management systems through an optimization-simulation approach. The optimization method used is a genetic algorithm (GA), which is a stochastic search method, based on “survival of the fittest.” Four issues relevant to groundwater remediation design were investigated: development of a robust constraint-handling technique for GAs, evaluating effects of constraint violations on optimal designs, investigation of various dynamic formulations, and development of a new GA for remediation design under uncertainty.; In exploring constraint-handling, two methods were evaluated for their effectiveness in finding feasible solutions to pump-and-treat design problems. Applying the Additive Penalty Method resulted in infeasible solutions, in which the effectiveness was strongly dependent on constraint weight values. However, the Multiplicative Penalty Method identified a set of feasible and near-optimal policies for both examples while using a range of constraint weight values.; Constraint violation sensitivity was investigated by solving two remediation examples with various remediation goals. The cost sensitivities to small constraint violations is of similar magnitude to the cost sensitivity to the ultimate water quality goal. Therefore, significant trade-offs which can occur between cost and water quality need to be considered. Dynamic formulations of optimal remediation design problems were compared. First, a static and two dynamic formulations of a bioremediation problem were compared. The simultaneous dynamic management policy had the lowest design cost. However, sequential dynamic policies had costs that were sensitive to the choice of intermediate remediation goal and were greater than static design costs. In another formulation, lengths of management periods were included as decision variables. Results indicated that using a dynamic policy, as opposed to a static policy, provided more significant cost savings than using flexible-length management periods in place of fixed-length periods.; Lastly, a GA was developed to take into account uncertainty in the hydraulic conductivity when determining robust remediation policies. The modified GA was evaluated by solving pump-and-treat examples with different heterogeneities. The method generated a set of non-dominated policies and demonstrated trade-offs between costs and clean-up goals. This work showed that using a deterministic description of the aquifer can result in significant under-design, with poor reliability. |
