Diagnostic Assessment of Preference Constraints for Simulation Optimization in Water Resources

Abstract

Simulation-optimization frameworks, such as multiobjective evolutionary algorithms (MOEAs), are increasingly used for real-world water resources problems. Constraints in MOEA optimization commonly represent decision maker preference, which differs from their role in classical optimization. As a result, constraints are often considered an optional aspect of the problem formulation. However, the impact of including constraints on optimization search has not been rigorously examined. This study explores how constraints impact the effectiveness, efficiency, and consistency of MOEA optimization for two water resources problems. For each problem, algorithm performance metrics are compared for two cases: (1) with constraints included during search, eliminating solutions that do not meet preference requirements, and (2) with constraints applied a posteriori to filter the full set of solutions. Results show that constraints aid in the search process by favoring solutions that meet decision maker preferences, despite the increased difficulty of finding feasible solutions. This study highlights the importance of constraints in the problem formulation for simulation-optimization applications in water resources, balancing the performance of search algorithms with the decision relevance of the solution set.