基于代理模型估值不确定度的昂贵多目标优化问题研究  

A research for expensive many-objective optimization problem based on uncertainty of surrogate

作　　者：张晶[1] 裴东兴[2,3] 马瑾[1] 沈大伟[2,3] ZHANG Jing;PEI Dongxing;MA Jin;SHEN Dawei(Department of Internet of Things Technology,Shanxi Vocational&Technical College of Finance&Trade,Taiyuan,Shanxi 030031,China;Science and Technology on Electric Test and Measurement Laboratory,North University of China,Taiyuan,Shanxi 030051,China;Key Laboratory of Instrumentation Science and Dynamic Measurement,Ministry of Education,North University of China,Taiyuan,Shanxi 030051,China)

机构地区：[1]山西财贸职业技术学院物联网技术系,山西太原030031 [2]中北大学电子测试技术国家重点实验室,山西太原030051 [3]中北大学仪器科学与动态测试教育部重点实验室,山西太原030051.

出　　处：《石河子大学学报（自然科学版）》2024年第1期110-116,共7页Journal of Shihezi University(Natural Science)

基　　金：国防科技重点实验室基金项目(61420010402001);山西省高等学校科技创新项目(2020L0762)。

摘　　要：针对代理模型辅助的多目标优化算法中个体不确定度之间相互冲突的问题,本文提出个体每个目标估值不确定的填充准则,同时,为了减少训练模型消耗的计算资源,提出基于非支配排序的样本选择算法。为了验证该算法的可行性,采用DTLZ和WFG测试函数进行测试,得出结果与近些年发表5种具有代表性的同类型算法进行对比,结果说明该算法可以有效的解决昂贵高维高目标优化问题。In surrogate assisted many objective optimization,conflicting uncertainties of surrogate between objective is a challenge.Hence,an many objective optimization algorithm with uncertainty of surrogate is proposed called,US-MOEA.The main work of this paper is as follows:first of all,infill criterion based on the uncertainty of predicted value is proposed to select promising solutions for re-evaluating by expensive optimization objective function.Then,in order to reduce the computational resources,a method based on non-dominated sorting is used to select some individual as train sample.In order to verify the effectiveness of proposed algorithm,the DTLZ and WFG test suits problem are applied and compare with five the-state-of-art algorithms proposed in recent years.The experimental results illustrate that the US-MOEA is an effectively method for solving expensive many objective optimization problems.

关 键 词：进化算法 昂贵多目标优化问题 代理模型 填充准则 不确定度 

分 类 号：TP391[自动化与计算机技术—计算机应用技术]