A two-stage evolutionary algorithm assisted by multi-archives for constrained multi-objective optimization

• Published in: Applied soft computing 2024-09, Vol.162, p.111840, Article 111840
• Main Authors: Zhang, Wenjuan, Liu, Jianchang, Zhang, Wei, Liu, Yuanchao, Tan, Shubin
• Format: Article
• Language: English
• Subjects: Constrained multi-objective optimization; Evolutionary algorithm; Multi-archives; Two-stage framework
• ISSN: 1568-4946; 1872-9681
• DOI: 10.1016/j.asoc.2024.111840

Due to the widespread existence of constrained multi-objective optimization problems (CMOPs) in real life, many researchers start to research the constrained multi-objective evolutionary algorithms (CMOEAs). Therefore, a variety of CMOEAs have emerged. However, some of them still suffer from great difficulties when coping with CMOPs with complex feasible regions. To solve the issue, this article puts forward a two-stage evolutionary algorithm assisted by multi-archives for constrained multi-objective optimization, called MA-TSEA. In MA-TSEA, the evolution process is divided into two stages. In Stage 1, non-dominated solutions obtained by non-dominated sorting based on (M+1) objectives ( i.e., M objectives and the constraint violation degree (CV) ) are stored in a temporary archive and merged with the parent population to improve the exploration ability of population. Then, MA-TSEA drives the populations to evolve from diverse directions by a multi-objective evolutionary algorithm, while the feasible solutions are saved in a formal archive. Next, the formal archive is updated to improve convergence and diversity. In Stage 2, the main population and archive population cooperate to evolve towards the constrained Pareto front (CPF), where the formal archive and population of Stage 1 are respectively used as the main population and archive population. The experimental studies on five benchmark test suites and five real-world applications demonstrate the superiority of MA-TSEA over the other seven state-of-art CMOEAs. •A MA-TSEA method is develop for CMOPs.•Two stages are used in MA-TSEA with different purposes.•Multiple archives assist the MA-TSEA find the complete constraint Pareto front.•MA-TSEA can achieve better results compared with its seven competitors.