Loser-out multi metaheuristic framework for multi-objective optimization

Abstract

This paper proposes a multi rnetaheuristic framework consisting of four multi-objective optimization (MOO) algorithms in which they compete with each other along four phases to be surviving in the next phases. Likewise, it is assumed that number of phases is equal to the number of metaheuristics. The proposed method, named as Loser-Out-Framework (LOF) from this point on, runs in consecutive sessions so that a session starts with dividing global population into several subpopulations. Thereafter in the first phase, entire set of metaheuristics is assigned to each subpopulation and then rnetaheuristics are performed over subpopulations to modify and improve them. in continuation of each phase, non-dominated solutions extracted by all metaheuristic sets are stored in global archive, and then the most ineffective rnetaheuristic of each subpopulation is eliminated. The proposed method is evaluated and tested over the well-known DTLZ and WFG benchmarks. Comparative evaluations against several state-of-the-art algorithms exhibits that the proposed framework outperforms others in terms of extracted Pareto front quality.