Decision space diversity can be essential for solving multiobjective real-world problems

Abstract

It has recently been argued that standard multiobjective algorithms like NSGA-II, SPEA2, and SMS-EMOA, are not well suited for solving problems with symmetries and/or multimodal single objective functions due to their concentration onto one Pareto set part. We here deliver a real-world application that shows such properties and is thus hard to solve by standard approaches. As direct tuning of the algorithms is too costly, we attempt it via constructive modeling (algorithm-based validation), but succeed only partly in improving performance, which emphasizes the need to integrate special operators for boosting decision space diversity in future algorithms. © Springer Physica-Verlag Berlin Heidelberg 2010.