Archiving strategies for evolutionary multi-objective optimization algorithms (Q2218696)

This monograph concentrates on improving stochastic population-based methods developed for multiobjective optimization, in particular, evolutionary algorithms. Here, solving multiobjective optimization problems is understood as generating a representative set of Pareto optimal solutions. The selection of which solutions to save during the solution process of a stochastic population-based method is called archiving. The main focus of the monograph is in developing archiving strategies, also called archivers, for computing discretizations of Pareto optimal solutions with certain properties. They also consider nearly optimal solutions. Thus, both representing Pareto optimal and so-called epsilon-approximate Pareto solutions is of interest. Finally, some evolutionary multiobjective optimization methods are coupled with proposed achivers and encouraging results of the strenths of such a combination are reported. In this, one representative method of the three types of evolutionary multiobjective optimization methods is considered, that is, of dominance-based, indicator-based and decomposition-based methods. In addition, a fourth method is considered that aims to approximate the set of epsilon-Pareto optimal solutions. Many of the examples considered have only two objective functions, which enables visualizing the solutions. Some attention is also paid to benefits of achiving in single objective optimization.