基于双分类器辅助进化的多目标优化算法  

Multi-objective Optimization Evolutionary Algorithm Based on Double Classifier-Assisted Evolution

作　　者：焦大利 姚亦飞 王成章 程杜 李清亮 于繁华 JIAO Dali;YAO Yifei;WANG Chengzhang;CHENG Du;LI Qingliang;YU Fanhua(College of Computer Science and Technology,Changchun Normal University,Changchun 130032,China;College of Computer Science and Technology,Beihua University,Jilin 132021,China)

机构地区：[1]长春师范大学计算机科学与技术学院,吉林长春130032 [2]北华大学计算机科学与技术学院,吉林吉林132021

出　　处：《北华大学学报（自然科学版）》2023年第5期664-670,共7页Journal of Beihua University（Natural Science）

基　　金：吉林省科技发展计划项目(20230101370J);吉林省教育厅科技攻关项目(JJKH20220840KJ)。

摘　　要：在昂贵多目标优化问题中,常采用代理辅助进化算法以减少真实目标函数的评估次数,但传统的代理辅助进化算法因代理模型计算复杂而运行时间较长.为缩短运行时间,提出基于双分类器辅助进化的多目标优化算法(DC-MOEA),利用两个随机森林分类器,分别预测解的多样性优劣和解的收敛性优劣,选出同时具备优秀收敛性和多样性的解进行真实评估和环境选择.DC-MOEA对决策变量分类并用分类后的数据训练分类器,降低整体复杂度,减少运行时间.通过仿真试验,利用基准测试函数比较在不同问题上的性能,验证本算法在多样性探索和收敛性增强方面的能力.Surrogate-Assisted Evolutionary Algorithms(SAEAs)are often used to reduce the evaluation numbers of the original Function Evaluations(FE)in expensive Multi-objective Optimization Problems(MOPs).However,traditional SAEAs often suffer from long running times due to the computational demands of the surrogate models.To shorten the execution time,a novel approach called a multi-objective optimization evolutionary algorithm based on double classifier-assisted evolution(DC-MOEA)is proposed.It utilizes two random forest classifiers to predict the diversity and convergence qualities of solutions,respectively.Subsequently,the algorithm selects solutions that exhibit both excellent convergence and diversity,then original FE and environmental selection are performed.DC-MOEA classifies the decision variables and trains the classifiers by using the classified data,thereby reducing the overall complexity and running time.Through simulation experiments conducted on various benchmark test functions,we compare the performance of DC-MOEA with other algorithms,demonstrate its capabilities in exploring diversity and enhancing convergence.

关 键 词：多目标优化问题 代理辅助进化算法 多目标进化算法 随机森林分类器 

分 类 号：TP183[自动化与计算机技术—控制理论与控制工程]