Discussion of Search Strategy for Multi-objective Genetic Algorithm with Consideration of Accuracy and Broadness of Pareto Optimal Solutions

Abstract

Abstract. In multi-objective optimization, it is important that the obtained solutions are high quality regarding accuracy, uniform distribution, and broadness. Of these qualities, we focused on accuracy and broadness of the solutions and proposed a search strategy. Since it is difficult to improve both convergence and broadness of the solutions at the same time in a multi-objective GA search, we considered to converge the solutions first and then broaden them in the proposed search strategy by dividing the search into two search stages. The first stage is to improve convergence of the solutions, and a reference point specified by a decision maker is adopted in this search. In the second stage, the solutions are broadened using the Distributed Cooperation Scheme. From the results of the numerical experiment, we found that the proposed search strategy is capable of deriving broader solutions than conventional multi-objective GA with equivalent accuracy.

The anticipated mean shift and cluster registration in mixture-based EDAs for multi-objective optimization

“…A combination of MO optimization and SO optimization has been proposed before [10]. There, m+1 equal-sized populations are used.…”

“…In this way, given enough clusters, the rate of convergence in each cluster is expected to be similar, resulting in better-aligned support of the SO optimizers in terms of convergence. Furthermore, in [10] solutions are migrated from the SO populations to the MO population and vice versa. Assuming competent SO optimizers however, this may only reduce the effectiveness of the SO optimizers.…”