How close simple EAs’ optimal solutions can approach global optima: experience from water distribution system design problems

Abstract

Abstract An issue regarding near-optimal solutions identified by evolutionary algorithms (EAs) is that their absolute deviations from the global optima are often unknown, and hence an EA's performance in handling real-world problems remains unclear. To this end, this paper investigates how close optimal solutions from simple EAs can approach the global optimal for water distribution system (WDS) design problems through an experiment with the number of decision variables ranging from 21 to 3,400. Three simple EAs are considered: the standard differential evolution, the standard genetic algorithm and the creeping genetic algorithm (CGA). The CGA consistently identifies optimal solutions with deviations lower than 50% to the global optimal, even for the WDS with 3,400 decision variables, but the performance of the other two EAs is heavily case study dependent. Results obtained build knowledge regarding these simple EAs’ ability in handling WDS design problems with different sizes. We must acknowledge that these results are conditioned on the WDSs and the parameterization strategies used, and future studies should focus on generalizing the findings obtained in this paper.