Coevolutionary Multitasking for Constrained Multiobjective Optimization

作     者：Liu, Songbai Wang, Zeyi Lin, Qiuzhen Chen, Jianyong 

作者机构：College of Computer Science and Software Engineering Shenzhen University Shenzhen518060 China 

出 版 物：《SSRN》 

年 卷 期：2024年

核心收录：

主　　题：Multitasking 

摘      要：Exploring the complex landscape of constrained multiobjective optimization problems (CMOPs) using evolutionary algorithms presents challenges of harmonizing constraints with optimization objectives. Coevolutionary multitasking (CEMT) emerges as a promising strategy to grapple with these complexities, capitalizing on the potential synergy achievable from distinct yet complementary tasks. Continuing in this research direction, this paper proposes an adaptive CEMT framework, referred to as ACEMT. The goal of ACEMT is to solve CMOPs effeciently by facilitating knowledge sharing from the two adaptive auxiliary tasks that exhibit complementarity to the main task (i.e., the target CMOP). One of these tasks adapts to the evolving landscape by gradually narrowing its constraint boundaries, thus effectively exploring regions with smaller feasible spaces. The second auxiliary task is intentionally structured to focus exclusively on specific constraints, undergoing continuous adaptation throughout the evolutionary process to expedite convergence and uncover potential regions. Moreover, to boost their efficiency, an adaptive constraint relaxation and an adaptive constraint selection strategy are customized for these two auxiliary tasks, respectively. To validate ACEMT s performance, a comprehensive series of experiments is conducted, encompassing three benchmark suites and real-world applications. The experimental findings underscore ACEMT s superiority, as it consistently outperforms or rivals other state-of-the-art related constrained evolutionary algorithms. © 2024, The Authors. All rights reserved.