A Knowledge Archive-Based Framework for Multitask Multimodal Optimization Problems

Evolutionary computation algorithms have been extensively studied to locate multiple global peaks in multimodal optimization problems (MMOPs). However, multitask multimodal optimization, which aims to deal with multiple MMOPs simultaneously, is not sufficiently considered in the literature. In this article, a knowledge archive-based framework (KAF) is proposed to extract and transfer knowledge among different tasks to help locate more peaks in each task, which is generic and can be combined with existing multimodal optimization algorithms to solve multitask MMOPs (MTMMOPs). In KAF, two novel strategies are designed. Firstly, a transferred individual generation strategy (TIGS) is designed to construct knowledge archives based on found global peaks and to generate transferred individuals who carry elite knowledge. Secondly, an adaptive competitor matching strategy (ACMS) is designed to adaptively conduct knowledge transfer between a transferred individual and its nearest individual or the worst individual in the population. In the experiments, a benchmark set of MTMMOPs is constructed. KAF is combined with five existing multimodal optimization algorithms and experimental results validate the effectiveness of KAF.