融合动态模糊扰动和特征引导的稀疏大规模多目标优化算法
打开文本图片集
中图分类号：TP301.6 文献标志码：A 文章编号：1001-3695（2025）12-016-3651-09
doi：10.19734/j. issn.1001-3695.2025.04.0113
Sparse large-scale multi-objective optimization algorithm incorporating dynamic fuzzy perturbation and feature guidance
Li Tongla.1b，Gu Qinghuala.b†，Wang Qiana.b，Luo Jialeb，1c，Wang Jianguo² （1.ColgeKbeltf Management，Xi’anUniverstyfArchitectureandTechnology，Xi’an7o55，China；2.Hami HexiangIndustryandTrdeCo，d， Hami Xinjiang 839200，China）
Abstract：Existing multi-objectiveevolutionaryalgorithmsfacechalenges including weaksparsitycontrol，porbalance inesolving objectiveconflicts，andsusceptibilitytolocaloptima whensolving large-scalesparseulti-objectiveoptimizationproblems.Thispaperproposedamulti-objectiveevolutionaryalgorithmincorporating dynamicfuzzyperturbationand featureguidedadaptivecrosover（MOEA-FA）toaddresstheseissues.Itfirstlyemployedadynamic fuzyperturbation strategywith dual-phase parameteradaptationtoguidetheevolutionarypopulation towardsspecificpreferenceregions forobtainingdecisionmaker-desiredsolutions，alancingojectiveconflictsandpreventingprematureonvergene.Additionally，tealgoritadopted afeature-gudedadaptivecrossoverstrategyforbinaryvariables，dentifyingcriticalfeaturesbasedonvariableactivationfre quencyandprioritizing theretentionofhigh-contributionvariables tomaintainsolutionsparsity.Tovalidate MOEA-FA'seffectiveness，the paper evaluated itagainst sixstate-of-the-artalgorithmson eightbenchmark problems（SMOP）and porfoliooptimization problems.Experimental results show MOEA-FA achieves the best IGD values on 80% of the test problems and the best HV values on 82.86% of the test problems. These results demonstrate MOEA-FA offers superior performance for solving large-scale sparse multi-objective optimization problems.
Keywords：large-scalesparsemulti-objectiveoptimization；dynamic fuzzyperturbation；feature-guidedadaptivecrossover; two-stage parameter self-adaptive control； bimodal membership competitive decision-making
0 引言
现实问题中常存在多个相互冲突的目标，因此将工程问题建模为多目标优化问题具有直观意义[1]。（剩余20832字）
