首页|多群多策略差分大规模多目标优化算法

多群多策略差分大规模多目标优化算法

扫码查看
原文链接
  • NETL
  • NSTL
  • 万方数据
针对差分进化算法在解决大规模多目标优化问题时，出现优化后期多样性不足、收敛速度慢等问题，提出一种多群多策略差分大规模多目标优化算法。根据个体特性不同，将种群分为3个等级不同的子群，利用多群策略的优势维持种群多样性。为减少种群陷入局部最优的概率，在不同等级的子群中引入多个变异策略以较好地平衡子群个体的多样性和收敛性。为保证不同子群间信息得到有效交换，根据3个子群的进化状态确定重新分群时机，既保证个体在本群内得到充分进化，又保证个体在一定的条件下进行信息交换。为利用更多的信息生成优秀的子代，将更新后的子群与其父代子群合并，选出下一代子群。为验证所提出算法的有效性，在一组大规模基准测试问题上评估算法的性能，实验结果表明，所提出算法在两个常用测试指标IGD和HV上明显优于其他对比算法。
A large-scale multi-objective optimization based on multi-population and multi-strategy differential algorithm
In the middle or late evolutionary stage of large-scale multi-objective optimization problems(LSMOPs),the differential evolution(DE)algorithm has problems such as diversity shortage and slow convergence.A large-scale multi-objective optimization based on multi-population and multi-strategy differential evolution(LMOMMDE)is proposed.According to the characteristics of individuals in the population,the population is divided into three subpopulations with different levels,and the advantages of the multi-population strategy are used to maintain the diversity of the population.To reduce the probability that the population will fall into local optimum,multiple mutation strategies are introduced for subpopulations on different levels,this operation better balances the diversity and convergence of individual in subpopulations.To ensure the effective exchange of information among different subpopulations,this paper determines the timing of regrouping according to the evolutionary status of the three subpopulations,the individual can fully evolve within the population,and the individual can effectively exchange information under certain conditions.To use more information to generate excellent offspring,the updated subpopulations and their parent subpopulations are combined and generate the next generation subpopulations.To verify the effectiveness of the LMOMMDE,the performance of this algorithm is evaluated on a set of large-scale benchmark problems.The experimental results show that LMOMMDE is significantly better than the comparison algorithms in the two commonly used test indicators IGD and HV.

large-scale optimizationmulti-objective optimizationdifference evolutionmultiple population strategymutation strategy

葛媛媛、陈得宝、邹锋

展开 >

淮北师范大学计算机科学与技术学院,安徽淮北 235000

安徽省认知行为智能计算工程研究中心,安徽淮北 235000

淮北师范大学物理与电子信息学院,安徽淮北 235000

大规模多目标优化 多目标优化 差分进化 多种群策略 变异策略

国家自然科学基金项目安徽省学术和技术带头人后备人选科研活动经费项目安徽省高校学科(专业)拔尖人才学术项目

619761012021H264gxbjZD2022021

2024

10.13195/j.kzyjc.2022.1154

控制与决策
东北大学

控制与决策

CSTPCD北大核心
影响因子：1.227
ISSN：1001-0920
年,卷(期)：2024.39(2)
  • 25