双模型驱动的多偏好策略自适应差分演化算法  

Dual Model⁃Driven Differential Evolution Algorithm with Multi⁃Preference Strategy Adaption

作　　者：龚懿昀 于海波 王韵 康丽 曾建潮 GONG Yiyun;YU Haibo;WANG Yun;KANG Li;ZENG Jianchao(School of Electrical and Control Engineering,North University of China,Taiyuan 030051,China;School of Computer Science and Technology,North University of China,Taiyuan 030051,China;School of Environment and Safety Engineering,North University of China,Taiyuan 030051,China)

机构地区：[1]中北大学电气与控制工程学院,山西太原030051 [2]中北大学计算机科学与技术学院,山西太原030051 [3]中北大学环境与安全工程学院,山西太原030051

出　　处：《中北大学学报（自然科学版）》2024年第5期638-646,共9页Journal of North University of China(Natural Science Edition)

基　　金：国家自然科学基金青年基金项目(62106237);国家自然科学基金联合基金项目(U21A20524);山西省自然科学基金项目(202203021222057)。

摘　　要：为增强代理模型辅助进化算法对高维昂贵优化问题的求解性能,提出了一种双模型驱动的多偏好策略自适应差分演化算法。该算法基于全局和局部两种代理建模方法,有机融合了3种具有不同寻优偏好的进化策略。每次迭代,通过利用优化过程中最优解在线更迭反馈信息,以序贯方式自适应调整不同进化策略调用频次,以高效平衡算法的全局勘探和局部开采。为促进种群内个体间优秀信息共享,设计了一种精英个体驱动的差分扰动策略,以增量潜在优解区域的最优样本先验。通过处理26个不同规模的高维基准测试问题,结果表明,所提算法的收敛性能和优化效率较4种先进的同类型算法在至少17个测试问题上绝对占优。In order to enhance the performance of surrogate-assisted evolutionary algorithms for solving high-dimensional expensive optimization problems,this paper proposed a dual model-driven differential evolution with multi-preference strategy adaption(SOEA-SS).SOEA-SS relied on three multipreference evolutionary strategies supported by global and local surrogates.At each iteration,SOEA-SS adaptively adjusted the evolutionary strategies in a sequential manner to strike the global exploration and local exploitation equilibrium,according to the online feedback concerning the update of optimal solution.In order to promote the optimal information sharing among the population,an elites-driven differential perturbation strategy was developed to enrich the prior knowledge of the optimal regions.Experimental results show that SOEA-SS has significant superiority over four advanced algorithms on at least 17 out of 26 high-dimensional benchmark problems.

关 键 词：代理模型 昂贵优化 差分演化 策略自适应 精英扰动 

分 类 号：TP273[自动化与计算机技术—检测技术与自动化装置]