Multi-objective Optimisation using S-metric Selection: Application to three-dimensional Solution Spaces

Authors:
  • BNB. Naujoks
    B. Naujoks
  • NBN. Beume
    N. Beume
  • MEM. Emmerich
    M. Emmerich

Abstract
The S-metric or hypervolume measure is a distinguished quality measure for solution sets in Pareto optimisation. Once the aim to reach a high S-metric value is appointed, it seems to be promising to directly incorporate it in the optimisation algorithm. This idea has been implemented in the SMS-EMOA, an evolution- ary multi-objective optimisation algorithm (EMOA) us- ing the hypervolume measure within its selection oper- ator. Solutions are rated according to their contribu- tion to the dominated hypervolume of the current pop- ulation. Up to now, the SMS-EMOA has only been applied to functions with two objectives. The work at hand extends these studies, by surveying the behaviour of the algorithm on three-objective problems. Addition- ally, a new efficient algorithm for the computation of the contributions to the dominated hypervolume in three- dimensional solution spaces is presented. Different vari- ants of selection operators are proposed. Among these, a new one is presented that rates a solution concerning the number of solutions dominating it. So, solutions in less explored regions are preferred. This rating is an efficient alternative to theS-metric criterion whenever a selec- tion among dominated solutions has to be made. Com- parative studies on standard benchmark problems show that the SMS-EMOA clearly outperforms other well es- tablished EMOA. First results on a challenging real- world problem have been obtained, namely the multi- point design of an airfoil involving three objectives and nonlinear constraints. Not only a clear improvement of the baseline design, but a good coverage of the Pareto front with a small, limited number of points has been achieved.