Analysing the Robustness of Multiobjectivisation Approaches Applied to Large Scale Optimisation Problems


Abstract

Multiobjectivisation transforms a mono-objective problem into a multiobjective one. The main aim of multiobjectivisation is to avoid stagnation in local optima, by changing the landscape of the original fitness function. In this contribution, an analysis of different multiobjectivisation approaches has been performed. It has been carried out with a set of scalable mono-objective benchmark problems. The experimental evaluation has demonstrated the advantages of multiobjectivisation, both in terms of quality and saved resources. However, it has been revealed that it produces a negative effect in some cases. Some multiobjectivisation schemes require the specification of additional parameters which must be adapted for dealing with different problems. Multiobjectivisation with parameters has been proposed as a method to improve the performance of the whole optimisation scheme. Nevertheless, the parameter setting of an optimisation scheme which considers multiobjectivisation with parameters is usually more complex. In this work, a new model based on the usage of hyperheuristics to facilitate the application of multiobjectivisation with parameters has been proposed. Experimental evaluation has shown that this model has increased the robustness of the whole optimisation scheme.