Comparison of Multi-Objective Evolutionary Algorithms for Long-Term Monitoring Design
Joshua B. Kollat, Patrick M. Reed · 2005
Long-term groundwater monitoring (LTM) design is an extremely challenging problem, which requires that engineers capture an impacted system's governing processes, elucidate human and ecologic risks, limit monitoring costs, and satisfy the interests of multiple stakeholders (e.g., site owners, regulators, and public advocates). Prior studies have shown that evolutionary multi-objective optimization (EMO) tools can aid decision makers by providing rapid assessments of the tradeoffs between conflicting design objectives (e.g., minimizing sampling costs and minimizing uncertainty). This study compares the performance of several EMO algorithms (the NSGAII, the ϵ-NSGAII, and the ϵMOEA) for supporting LTM design. The EMO algorithms are used to quantify tradeoffs for a four-objective LTM test case. Optimization objectives include: (i) minimize sampling cost, (ii) maximize mapping accuracy while (iii) minimizing uncertainty, and (iv) minimize contaminant mass estimation error. A 25-well LTM test case has been enumerated to provide a reference Pareto-optimal solution set to facilitate rigorous testing of the EMO algorithms. The performances of the three algorithms are assessed and compared using three published performance metrics (convergence, diversity, and ϵ-performance). Results of the analysis indicate that the ϵ-NSGAII performs the most reliably and ultimately attains the highest performance on this application. In addition, the ϵ-NSGAII's ability to adaptively size its population and automatically terminated its run make it an appealing algorithm to the water resources practitioner by simplifying EMO search.