双阶段填充采样辅助的昂贵多目标优化  

Expensive multi-objective optimization assisted by two-stage infill sampling

作　　者：秦淑芬 孙超利[1] QIN Shu-fen;SUN Chao-li(College of Computer Science and Technology,Taiyuan University of Science and Technology,Taiyuan 030024,China)

机构地区：[1]太原科技大学计算机科学与技术学院,山西太原030024

出　　处：《计算机工程与设计》2024年第8期2492-2502,共11页Computer Engineering and Design

基　　金：国家自然科学基金项目(62303344、62372319);山西重点研发基金项目(202102020101002);山西省青年科学研究基金项目(202203021222196);太原科技大学校博士科研启动基金项目(20222053);山西省高等学校科技创新计划基金项目(288)。

摘　　要：针对代理模型引导多目标优化算法,求解决策变量数量增多的昂贵多目标优化问题时,搜索效率较低的问题,提出一种双阶段填充采样辅助的昂贵多目标优化算法。第一阶段,利用一组方向向量引导产生靠近真实最优解集的样本,加快模型引导算法搜索;第二阶段,由代理模型估计获得估值误差,融合个体与样本之间相似性、个体估值收敛性,选择个体用于真实评价后填充样本集,实现模型性能的提升。在100维和200维的多目标基准测试问题上的实验结果表明,所提算法在同等有限资源内获得了比其它算法更为显著的优势。Aiming at the problem of the low efficiency of search,an expensive multi-objective optimization algorithm assisted by two-stage infill sampling was proposed when the surrogate model-guided multi-objective optimization algorithm was used to solve those multi-objective optimization problems with the increasing number of decision variables.In the first stage,the samples close to the true optimal solution set were generated based on a guiding direction set for speeding up the model-guided search.In the second stage,the errors obtained from the surrogate model estimation were integrated with the similarity between individuals and samples and the convergence of individuals to select individuals for expensive function evaluations to infill the sample set and improve the model’s performance.Experimental results on 100-and 200-dimensional multi-objective benchmark problems show the proposed method is more significantly superior to other compared algorithms within the same limited resources.

关 键 词：昂贵多目标优化 代理模型辅助的进化优化 双阶段采样 定向采样 填充采样 估值误差 个体收敛性 

分 类 号：TP181[自动化与计算机技术—控制理论与控制工程]