基于自适应采样策略的模糊分类代理辅助进化算法  

Fuzzy Classification Surrogate-assisted Evolutionary Algorithm Based on Adaptive Sampling Strategy

作　　者：李二超[1] 吴煜 LI Erchao;WU Yu(College of Electrical and Information Engineering,Lanzhou University of Technology,Lanzhou 730050,China)

机构地区：[1]兰州理工大学电气工程与信息工程学院,甘肃兰州730050

出　　处：《郑州大学学报(工学版)》2025年第2期51-59,共9页Journal of Zhengzhou University(Engineering Science)

基　　金：国家自然科学基金资助项目(62063019);甘肃省自然科学基金资助项目(24JRRA173)。

摘　　要：针对基于分类代理辅助进化算法模型管理效率不高和如何有效降低真实函数评估次数的问题,提出了一种基于自适应采样策略的模糊分类代理辅助进化算法。首先,算法通过帕累托支配关系筛选样本来构造代理模型;其次,采用基于转移的密度估计策略提高选择压力,兼顾收敛性与多样性,同时利用十折交叉验证得到精度信息用来划分状态;最后,设计了一种自适应模型管理策略,其考虑当前种群的收敛性、多样性和不确定性,并根据不同精度状态采用有针对性的采样方式,该算法能够在保证整体性能的前提下,合理减少真实评估次数。为验证所提算法性能,将该算法与其他4种算法在MaF、WFG测试集和汽车侧面碰撞设计与驾驶室设计的实际工程问题上进行了分析对比实验,实验结果表明:所提算法在有限次评估条件下,在解决昂贵多目标优化问题时具有较好的竞争力。In order to solve the problem of low management efficiency of classification surrogate-assisted evolutionary algorithms based on fuzzy classification and how to effectively reduce the number of real function evaluations,in this study a fuzzy classification surrogate-assisted evolutionary algorithm was proposed based on an adaptive sampling strategy.Firstly,in the algorithm the agent model was constructed by screening samples through the Pareto dominance relationship.Then,the selection pressure was improved by using a transfer-based density estimation strategy,which balanced convergence and diversity.At the same time,ten-fold cross-validation was used to obtain accuracy information to divide the states.Finally,an adaptive model management strategy was designed.Considering the convergence,diversity,and uncertainty of the current population,targeted sampling methods was adopted according to different accuracy states.The algorithm could ensure overall performance while rationally reducing the number of real evaluations.To verify the performance of the proposed algorithm,it was compared with four other algorithms on the MaF and WFG test sets and real-world engineering problems of automotive side impact design and driving cabin design.The experimental results showed that the proposed algorithm,which could prove that the algorithm had good competitiveness in solving expensive multi-objective optimization problems with the limited number of real evaluations.

关 键 词：代理辅助进化算法 代理模型 昂贵多目标优化问题 模型管理 

分 类 号：TP18[自动化与计算机技术—控制理论与控制工程]