摘要
为了提高用于更新代理模型的解集的多样性和收敛性以提高代理模型准确度,提出一种基于行列式点过程(determinantal point process,DPP)的代理模型辅助多目标进化算法(surrogate-assisted evolutionary algorithm,SAEA)。首先,提出一种基于行列式点过程的模型管理方法,从非支配解集基于行列式点过程选取子集并用真实目标函数评估,再从所有经真实目标函数评估的解中选取子集用于更新代理模型。另一方面,提出一种基于自适应行列式点过程的环境选择方法,在进化过程的早期侧重于提高种群的收敛性,在进化过程的后期侧重于提高种群的多样性。最后,基于DTLZ、WFG、MAF测试问题验证算法的有效性。将所提算法与K-RVEA、KTA2、CSEA等常用算法进行比较,使用IGD+指标进行评估。实验结果显示所提出的算法能得到更优的解集,从而证明了其高计算代价多目标优化问题上的有效性。
To enhance the diversity and convergence of the solution set used for updating the surrogate models and thereby improve the accuracy of the surrogate models,this paper proposed an SAEA based on DPP.Firstly,this paper proposed a surrogate management method based on DPP.The method selected a subset from the non-dominated solution set using DPPs and evaluated solutions in the subset with the real objective functions,and then selected another subset from the set of all solutions evaluated by the real objective functions to update the surrogate models.Additionally,this paper proposed an environmental selection method based on adaptive DPP.The method focused on improving the convergence of the population in the early stages of the evolutionary process and enhancing the diversity of the population in the later stages.Finally,this paper verified the effectiveness of the proposed algorithm on DTLZ,WFG,and MAF test problems.This paper compared the proposed algorithm with commonly used algorithms such as K-RVEA,KTA2,and CSEA using the IGD+metric.The experimental results show that the proposed algorithm can obtain a better solution set,thereby demonstrating its effectiveness in solving expensive multi-objective optimization problems.
作者
吴子聪
李金龙
Wu Zicong;Li Jinlong(School of Artificial Intelligence&Data Science,University of Science&Technology of China,Hefei 230026,China)
出处
《计算机应用研究》
北大核心
2025年第9期2607-2613,共7页
Application Research of Computers
基金
国家自然科学基金面上项目(61573328)。
关键词
代理辅助多目标优化
进化算法
模型管理
环境选择
行列式点过程
收敛性
多样性
surrogate-assisted multi-objective optimization
evolutionary algorithm
surrogate management
environmental selection
determinantal point processes
convergence
diversity
