Pre-DEMO: Preference-Inspired Differential Evolution for Multi/Many-Objective Optimization

Abstract

Differential evolution (DE) has emerged as an effective technique for single-objective optimization problems (SOPs). Due to its efficient and straightforward framework, it has been extended further to address multiobjective optimization problems (MOPs). However, the existing multiobjective DE (MODE) algorithms focus on developing control strategies of mutation operators and parameters for a given population at every iteration, regardless of whether the population has insufficient distributions in objective space. Furthermore, several technical challenges exist when extending MODE approaches to deal with many-objective optimization problems (MaOPs). To break through such limitations, in this article, we propose a preference-inspired DE for multi and many-objective optimization (Pre-DEMO), which effectively and efficiently deals with a wide range of MOPs and MaOPs. First, a preference-inspired mutation operator is developed to generate individuals with good convergence and distribution properties. The local knee points are obtained among the nondominated individuals to articulate preferences in the mutation operator. Also, an adaptive strategy based on a clustering method is proposed to determine the local knee points. Second, a two-stage environmental selection is suggested in Pre-DEMO to preserve promising individuals for the next generations. Experimental results demonstrate that the Pre-DEMO approach outperforms the eight state-of-the-art algorithms on 35 benchmark problems.