摘要
大规模多目标优化问题(Large-Scale Multi-objective Optimization Problem,LSMOP)固有的性质给多目标进化算法(Multi-Objective Evolutionary Algorithm,MOEA)带来挑战。目前大多数大规模多目标进化算法(Large-Scale Multi-Objective Evolutionary Algorithm,LSMOEA)需要耗费较多的计算资源对大规模决策变量进行分组,使得用于优化问题解的计算资源相对不足,影响了算法效率和解题性能。基于此,本研究提出一种基于变量两阶段分组的多目标进化算法(Large-Scale Multi-Objective Evolutionary Algorithm adopting two-stage variable grouping,LSMOEA/2s)。新算法首先利用基于变量组的相关性检测方法快速识别独立变量,然后利用高频次随机分组方法将非独立变量划分成若干子组,最后利用MOEA/D算法优化所有的独立变量和非独立变量子组。将所提算法与当前4种代表性算法(MOEA/D、CCGDE3、RVEA、S3-CMA-ES)一同在LSMOP系列测试问题上进行反转世代距离(Inverted Generational Distance,IGD)性能测试,结果表明,LSMOEA/2s较其他算法具有显著的性能优势。
The inherent properties of Large-Scale Multi-objective Optimization Problem(LSMOP)bring chal-lenges to Multi-Objective Evolutionary Algorithm(MOEA).At present,most Large-Scale Multi-Objective Evolutionary Algorithm(LSMOEA)need to consume more computational resources to group large-scale decision variables,which makes the computational resources used to optimize the problem solution relatively in-sufficient,affecting the efficiency and performance of the algorithm.Base on this,a Large-Scale Multi-Objec-tive Evolutionary Algorithm adopting two-stage variable grouping(LSMOEA/2s)is proposed in this study.Firstly,the new algorithm is conducive to the correlation detection method based on variable group to quickly identify independent variables.Then,the high-frequency random grouping method is used to divide the non-independent variables into several subgroups.Finally,the MOEA/D algorithm is used to optimize all inde-pendent variables and non-independent variable subgroups.The proposed algorithm is combined with the current four representative algorithms(MOEA/D,CCGDE3,RVEA,S3-CMA-ES)to perform the inverted gen-erational distance Inverted Generational Distance(IGD)performance test on the LSMOP series test prob-lems.The results show that LSMOEA/2s has significant performance advantages over other algorithms.
作者
谢承旺
潘嘉敏
付世炜
廖剑平
XIE Chengwang;PAN Jiamin;FU Shiwei;LIAO Jianping(School of Computer and Information Engineering,Nanning Normal University,Nanning,Guangxi,530000,China;School of Data Science and Engineering,South China Normal University,Shanwei,Guangdong,516600,China)
出处
《广西科学》
CAS
北大核心
2023年第2期413-420,共8页
Guangxi Sciences
基金
国家自然科学基金项目(61763010)
广西自然科学基金项目(2021GXNSFAA075011)资助。
关键词
大规模决策变量
多目标优化问题
大规模多目标进化算法
两阶段分组
收敛性
多样性
large-scale decision variables
multi-objective optimization problem
Large-Scale Multi-Objective Evolutionary Algorithm
two-stage variable grouping
convergence
diversity