Design and tuning of an evolutionary multiobjective optimisation algorithm

Abstract

In this cumulative thesis an approach to multiobjective evolutionary optimisation using the hypervolume or the S-metric, respectively for selection is presented. This algorithm is tested and compared to standard techniques on two-, three and more dimensional objective spaces. To decide upon the right time when to stop a stochastic optimisation run, the method called online convergence detection is developed. This as well as the framework of sequential parameter optimisation for evolutionary multiobjective optimisation algo- rithms are general frameworks for different kinds of optimisation approaches. Both are successfully coupled with the presented algorithm on different test cases, even industrial ones from aerodynamics. A chapter on diversity in decision and objective space completes this thesis, which ends with an outlook on interesting research directions for the future.