A Dynamic Exemplars Selection-based Differential Evolution Algorithm for Constrained Multi-objective Optimization

Constrained multi-objective optimization problems (CMOPs) pose a significant challenge due to the presence of multiple conflicting objective functions requiring optimization, alongside numerous complex constraints that must be satisfied. When employing evolutionary algorithms (EAs) to address such problems, it is imperative to simultaneously consider the diversity, convergence, and feasibility of the population. To address this concern, this paper introduces a differential evolution algorithm based on dynamic learning exemplars, named DESDE, wherein the selection of learning exemplars for each individual is determined through evolutionary algebra. Specifically, to generate offspring with diversity, individuals have a plethora of learning exemplars to choose from during the early stage of evolution, thus more promising regions can be explored. Moreover, to expedite population convergence in the later stage, only individuals with superior performance are chosen as learning exemplars. The experimental results, obtained from evaluating the algorithm on 29 test problems alongside five classic constrained multi-objective evolutionary algorithms (CMOEAs), demonstrate the high competitiveness of the proposed method.