A surrogate model for simulation-optimization of aquifer systems subjected to seawater intrusion

This study presents the application of Evolutionary Polynomial Regression (EPR) as a pattern recognition system to predicate the behavior of nonlinear and computationally complex aquifer systems subjected to seawater intrusion (SWI). The developed EPR models are integrated with a multi objective genetic algorithm to examine the efficiency of different arrangements of hydraulic barriers in controlling SWI. The objective of the optimization is to minimize the economic and environmental costs. The developed EPR model is trained and tested for different control scenarios, on sets of data including different pumping patterns as inputs and the corresponding set of numerically calculated outputs. The results are compared with those obtained by direct linking of the numerical simulation model with the optimization tool. The results of the two above-mentioned simulation-optimization (S/O) strategies are in excellent agreement. Three management scenarios are considered involving simultaneous use of abstraction and recharge to control SWI. Minimization of cost of the management process and the salinity levels in the aquifer are the two objective functions used for evaluating the efficiency of each management scenario. By considering the effects of the unsaturated zone, a subsurface pond is used to collect the water and artificially recharge the aquifer. The distinguished feature of EPR emerges in its application as the metamodel in the S/O process where it significantly reduces the overall computational complexity and time. The results also suggest that the application of other sources of water such as treated waste water (TWW) and/or storm water, coupled with continuous abstraction of brackish water and its desalination and use is the most cost effective method to control SWI. A sensitivity analysis is conducted to investigate the effects of different external sources of recharge water and different recovery ratios of desalination plant on the optimal results. (C) 2015 Elsevier B.V. All rights reserved.