An approach for the ordering of evaluation of objectives in multiobjective optimization

Abstract

The computational complexity of the multiobjective optimization (MOO) increases drastically in the presence of the large number of objectives. It is desirable to lower the complexity of the existing MOO algorithms. In this work we present an algorithm which periodically rearranges the objectives in the objective set such that the conflicting objectives are evaluated and compared earlier than non-conflicting ones. Differential Evolution (DE) is used as the underlying search technique. DE is designed especially for the real optimization problems. We have studied the reduction in the number of function computations and timing requirement achieved with the proposed technique. Remarkably, it is found to be much reduced as compared to the traditional approach. The variation of the gain in the number of objective computations vis-a-vis the number of objectives is demonstrated for a large number of benchmark MOO problems. Additionally, the relationship between the frequency of reordering the objectives and the number of objective computations is also established experimentally. © 2013 Springer.