Collaborative knowledge transfer-based multiobjective multitask particle swarm optimization

Abstract

Evolutionary multitask optimization (EMTO) has been an emerging optimization paradigm to handle several different optimization problems in parallel by utilizing knowledge transfer. However, most existing EMTO algorithms focus only on facilitating knowledge transfer in the search space to deal with multiple optimization tasks, while ignoring the potential relationship problem in the objective space, which may lead to the degradation of knowledge transfer performance, especially for multiobjective EMTO. To address this problem, a collaborative knowledge transfer-based multiobjective multitask particle swarm optimization (CKT-MMPSO) is designed in this paper. First, a CKT-MMPSO scheme is introduced to comprehensively exploit the knowledge from different spaces to solve multiple optimization problems. Then, the knowledge transfer can be effectively implemented to improve the quality of solutions. Second, a bi-space knowledge reasoning method is developed to make full use of population distribution information in the search space and particle evolutionary information in the objective space. Then, the search space knowledge and the objective space knowledge can be acquired to assist in the knowledge transfer. Third, an information entropy-based collaborative knowledge transfer mechanism is designed to balance convergence and diversity. Then, knowledge transfer patterns can be adaptively performed in different evolutionary stages to generate promising solutions. Finally, CKT-MMPSO is applied to some benchmark problems to verify its effectiveness. Furthermore, compared with other state-of-the-art algorithms, several experiments demonstrate that CKT-MMPSO can achieve the desirable performance.