基于聚类的代理模型构建策略  

Clustering-based strategy for construction of surrogate models

作　　者：陈彧[1] 杨媛媛 CHEN Yu;YANG Yuan-yuan(School of Science,Wuhan University of Technology,Wuhan 430070,China)

机构地区：[1]武汉理工大学理学院,湖北武汉430070

出　　处：《计算机工程与设计》2021年第6期1623-1630,共8页Computer Engineering and Design

基　　金：国家自然科学基金项目(61303028)。

摘　　要：为提高代理模型效率,对数据进行聚类,根据模型构建的需求从聚类结果中选择少量数据训练模型参数,得到一种基于聚类的代理模型构建策略。以该策略构建的代理模型作为近似评价模型,设计一种基于聚类的代理模型辅助粒子群优化算法,提出一种以代理模型辅助迭代搜索作为加速策略的混合变量多目标进化算法。数值实验结果表明,基于聚类的代理模型构建策略具有时间复杂度低、模型构造效率高的优点,模型辅助的迭代搜索过程具有较好的全局探索和局部开发能力。To improve efficiencies of surrogate models,a clustering-based construction strategy of surrogate models(CCS-SM)was proposed,where model parameters were confirmed by a few data points purposefully selected from the clustering results.Taking the constructed surrogates as models of approximate evaluation,a particle swarm optimization assisted by surrogates based on clustering(PSO-ASC)was proposed,as well as a surrogate-assisted mixed-variable multiobjective evolutionary algorithm(SaMV-MOEA)accelerated by the surrogate-assisted iterations.Results of numerical experiments show that CCS-SM is an efficient construction strategy of low computation complexity,which contributes to surrogate-assisted searches with good abi-lities of global exploration and local exploitation.

关 键 词：代理模型 模型管理 粒子群优化算法 集成代理模型 聚类 

分 类 号：TP183[自动化与计算机技术—控制理论与控制工程]