文献类型：期刊文献

中文题名：面向复杂可行域约束多目标优化问题的双种群协同进化算法

英文题名：Two-population Co-evolutionary Algorithm for Constrained Multi-objective Optimization Problems in Complex Feasible Domains

作者：丁炜超[1];孙立烨[1];罗飞[1];顾春华[1];董文波[1]

机构：[1]华东理工大学信息科学与工程学院,上海200237

年份：2025

卷号：51

期号：9

起止页码：2037

中文期刊名：自动化学报

外文期刊名：Acta Automatica Sinica

收录：;EI(收录号：20253919230234);北大核心:【北大核心2023】；

基金：国家自然科学基金(62506130,62403201);上海市自然科学基金(24ZR1415200,23ZR1414900,22ZR1416500);上海市基础研究特区计划(22TQ1400100-16)资助。

语种：中文

中文关键词：约束多目标优化;复杂可行域;协同进化;粒子群优化;向量群引导

外文关键词：Constrained multi-objective optimization;complex feasible domains;co-evolutionary;particle swarm optimization;vector swarm guidance

摘要：约束多目标优化问题主要考虑如何在复杂约束条件下同时优化多个相互冲突的目标,其广泛存在于工程实践中.解决多目标优化问题的关键在于约束满足和目标优化之间的平衡.然而,当问题具有复杂可行域时,现有算法往往存在选择压力大小的矛盾:若算法的选择压力较大,种群容易陷入局部最优;若算法的选择压力较小,种群则难以搜索到完整的约束前沿.针对此,提出一种双种群协同进化约束多目标优化算法.所提算法采用双种群协同进化框架,引入粒子群和向量群以实现种群间的信息共享和优势互补.其中粒子群使用带有辅助档案的粒子群优化器,通过粒子间的相互学习实现快速收敛,而辅助档案则借助逃逸机制帮助粒子群跳出局部最优.同时,设计一种新的ε-约束技术,动态调整约束松弛因子,使种群在进化初期注重不可行解的遗传信息,跨越不可行区域.向量群使用不考虑约束的参考向量法引导种群进化,使种群均匀分布于前沿面,有效维护了种群的多样性.在当前基准测试集和真实世界73个问题上的实验结果表明,所提出的算法超越对比算法,能够在保持种群多样性的同时快速收敛到约束前沿.
Constrained multi-objective optimization problems mainly consider simultaneously optimizing multiple conflicting objectives under complex constraint conditions,which are widely encountered in engineering practice.The key to solving multi-objective optimization problems lies in balancing constraint satisfaction with objective optimization.However,when tackling problems with complex feasible domains,existing algorithms often face a tradeoff in selection pressure:Excessive pressure may lead to population convergence to local optima;While insufficient pressure can hinder the discovery of the complete constraint front.To address the above issues,this paper proposes a two-population co-evolutionary constrained multi-objective optimization algorithm(TCCMOA).The proposed algorithm adopts a two-population co-evolutionary framework and introduces particle swarm and vector swarm to achieve information sharing and complementary advantages between populations.The particle swarm uses a particle swarm optimizer with auxiliary archive to achieve fast convergence through mutual learning between particles,while the auxiliary archive helps the particle swarm escape from local optima through escape mechanisms.Meanwhile,a new-constraint technology is designed to dynamically adjust the constraint relaxation factor,so that the population pays attention to the infeasible genetic information in the early stage of evolution and crosses the infeasible domains.Vector swarm employs the unconstrained reference vector method to guide population evolution,resulting in a uniform distribution of the population on the frontier and effectively maintaining population diversity.The experimental results on 73 problems in benchmark and real-world sets show that TCCMOA outperforms the comparative algorithms and can quickly converge to the constrained front while maintaining the diversity of the population.