Automated solution selection in multi-objective optimisation

Abstract

This paper proposes an approach to the solution of multi-objective optimisation problems that delivers a single, preferred solution. A conventional, population-based, multi-objective optimisation method is used to provide a set of solutions approximating the Pareto front. As the set of solutions evolves, an approximation to the Pareto front is derived using a Kriging method. This approximate surface is traversed using a single objective optimisation method, driven by a simple, aggregated objective function that expresses design preferences. The approach is demonstrated using a combination of multi-objective particle swarm optimisation (MOPSO) and the Simplex method of Nelder and Mead, applied to several, standard, multi-objective test problems. Good, compromise solutions meeting user-defined design preferences are delivered without manual intervention.