Leadership succession inspired adaptive operator selection mechanism for multi-objective optimization

Abstract

Dynamic selection of representative operators shows great promise for multi-objective optimization, but existing methods suffer from difficulties in balancing fair comparison of operators with dynamic adaptation of evolutionary states, and inaccurate evaluation of operator quality. This paper proposes a leadership succession inspired adaptive operator selection mechanism (LS-AOS), aiming to enhance dynamic matching with time-varying evolutionary states while ensuring fair operator comparisons. In LS-AOS, a new campaign-incumbency rule is designed to be implemented iteratively to allow operators to undergo a fair campaign process, thus identifying optimal operators for generating offspring. Additionally, a two-layer oversight strategy is proposed to make real-time adjustments to operator selection and pool configuration based on operator performance and evolutionary state, with the aim of satisfying the diverse requirements for exploration and exploitation during the evolutionary process. To refine and improve the evaluation of operator quality, the novel Election Campaign Indicator (ECI) is designed that uniquely integrates measures of population diversity and convergence, and effectively extends the applicability of LS-AOS. The experimental results on 23 test problems indicate that LS-AOS possesses feasibility and can effectively improve the performance of benchmark algorithms. Compared with the existing state-of-the-art algorithms, the proposed LS-AOS exhibits sufficient competitiveness and advancement.