首页|一种采用混合策略的大规模多目标进化算法

一种采用混合策略的大规模多目标进化算法

扫码查看
原文链接
  • NETL
  • NSTL
  • 万方数据
现实中存在大量的大规模多目标优化问题,这些问题所固有的目标函数间冲突性、巨大的搜索空间以及决策变量可能存在的交互等特征对传统的多目标进化算法构成了巨大的挑战.研究者根据此类问题的特点基于不同的视角提出了多种大规模多目标进化算法,但它们在解题的质量和效率方面尚存较大的提升空间.基于此,提出一种采用混合策略的大规模多目标进化算法LSMOEA/HS.该算法提出的一种黄金分层分组方法将大规模决策变量分成收敛性组和多样性组,然后对收敛性变量组执行基于变量组的相关性检测操作,将收敛性变量组划分成若干更小规模的子组,最后算法采用不同的优化策略分别优化收敛性变量组和多样性变量组以获得最终的解题结果.为验证LSMOEA/HS的有效性,将其与五种新近提出的高效的大规模多目标进化算法一同在决策变量维度为200、500、1000、2000和5000的2-目标和3-目标的LSMOP系列测试实例上进行IGD和HV性能测试,实验结果表明LSMOEA/HS具有显著较优的收敛性和多样性.由此表明,LSMOEA/HS是一种颇具前景的大规模多目标进化算法.
A Large Scale Multi-Objective Evolutionary Algorithm Adopting Hybrid Strategies
There exist a number of large-scale multiobjective optimization problems(LSMOPs)in reality.These LSMOPs admit some inherent characteristics such as the inherent conflicts among their objective functions,huge search space(decision variable space),and possible interaction existing in decision variables,which raise great challenges to the classical multiobjective evolu-tionary algorithms(MOEAs).Based on the characteristics of the LSMOPs,researchers have proposed a variety of large-scale multiobjective evolutionary algorithms(LSMOEAs)from different perspectives.However,these LSMOEAs are still far from satisfactory in terms of problem-solving quality and efficiency,leading an urgent need to improve significantly both the quality and efficiency.Motivated by these observations,a large-scale multiobjective evolutionary algorithm based a hybrid strategy,termed as LSMOEA/HS,is proposed in this paper.The LSMOEA/HS proposes a golden grouping method to divide large-scale decision variables into two groups,the convergence group and the diversity group.For the convergence group,a correlation detection operator based group is proposed to divide the convergence group into several subgroups of smaller size.Finally,in order to obtain the final solution set,the algorithm adopts different optimization strategies to optimize the convergent variable group and the diversity variable group respectively.In order to verify the effectiveness of LSMOEA/HS,an empirical comparison with five other newly developed efficient large-scale multiobjective evolutionary algorithms is performed in this paper.In this experimental comparison,both 2-objective and 3-objective benchmark LSMOP suite with decision variable dimensions of 200,500,1000,2000 and 5000 are used as test instances,and the IGD and HV are selected as performance indicators.The empirical results show that LSMOEA/HS has achieved significantly better results in terms of both the convergence and diversity.It shows that LSMOEA/HS is a promising large-scale multi-objective evolutionary algorithm with potential to solve large-scale multiobjective optimization problems.

large-scale multi-objective optimization problemsvariable groupingevolutionary algorithmconvergencediversitylarge-scale multi-objective evolutionary algorithm

谢承旺、潘嘉敏、郭华、王冬梅、付世炜

展开 >

华南师范大学数据科学与工程学院 广东汕尾 516600

南宁师范大学计算机与信息工程学院 南宁 530100

成都职业技术学院软件学院 成都 610041

大规模多目标优化问题 变量分组 进化算法 收敛性 多样性 大规模多目标进化算法

国家自然科学基金项目广西自然科学基金项目

617630102021GXNSFAA075011

2024

10.11897/SP.J.1016.2024.00069

计算机学报
中国计算机学会 中国科学院计算技术研究所

计算机学报

CSTPCD北大核心
影响因子：3.18
ISSN：0254-4164
年,卷(期)：2024.47(1)
  • 3