Preference-Based Multi-Objective Particle Swarm Optimization Using Desirabilities

摘要

The integration of experts' preferences is an important aspect in multi-objective optimization. Usually, one out of a set of Pareto optimal solutions has to be chosen based on expert knowledge. A combination of multi-objective particle swarm optimization (MOPSO) with the desirability concept is introduced to efficiently focus on desired and relevant regions of the true Pareto front of the optimization problem which facilitates the solution selection process. Desirability functions of the objectives are optimized, and the desirability index is used for selecting the global best particle in each iteration. The resulting MOPSO variant DF-MOPSO in most cases exclusively generates solutions in the desired area of the Pareto front. Approximations of the whole Pareto front result in cases of misspecified desired regions.