Constrained multi-objective state transition algorithm via adaptive bidirectional coevolution

• Published in: Expert systems with applications 2025-03, Vol.266, p.126073, Article 126073
• Main Authors: Sun, Yan, Zhou, Xiaojun, Yang, Chunhua, Huang, Tingwen
• Format: Article
• Language: English
• Subjects: Adaptive bidirectional coevolution; Constrained multi-objective optimization; Dynamic ɛ-constraint boundary; State transition algorithm
• ISSN: 0957-4174
• DOI: 10.1016/j.eswa.2024.126073

Constrained multi-objective optimization problems (CMOPs) involve optimizing multiple conflicting objectives subject to at least one constraint. These constraints often divide the search space into various infeasible regions and narrow or disconnected feasible regions. Most existing constrained multi-objective evolutionary algorithms struggle with imbalanced exploration between feasible and infeasible regions and exhibit poor search capabilities, resulting in populations becoming trapped in locally optimal feasible or infeasible areas. To overcome this limitation, we propose a novel constrained multi-objective state transition algorithm via adaptive bidirectional coevolution (CMOSTA). This algorithm comprises a main population (MP) and a cooperative population (CP), facilitating balanced exploration of both feasible and infeasible regions. CMOSTA adapts environmental selection strategies based on the proportion and distribution of feasible solutions within the MP, promoting efficient information sharing and avoiding unnecessary searches. Additionally, a dynamic ɛ-constraint relaxation strategy is put forward for the MP to prevent stagnation in locally feasible areas. A mating selection approach combining binary tournament and dynamic ɛ-constrained dominance is developed, followed by state transformation operators to generate candidate solutions with both global and local search capabilities. The effectiveness of CMOSTA is verified through 62 benchmark tests and an industrial case study on optimal copper removal, showing superior performance compared to ten well-established constrained multi-objective evolutionary algorithms. [Display omitted] •Propose a coevolution-based constrained multi-objective state transition algorithm.•Adapt environmental strategies based on real-time status of feasible solutions.•Present a dynamic ɛ-relaxation strategy to prevent local stagnation.•Employ binary tournament with ɛ-dominance for mating selection.•Adopt state transformation operators to generate high-quality candidate solutions.