摘要
针对目标函数评价昂贵的优化问题,在计算资源有限的情况下很难获得足够数据训练一个准确的全局代理模型,然而,不准确的全局代理模型其潜在优势是可以平滑局部极值点,从而可以引导算法加速找到最优解。另一方面,局部模型虽然不能辅助算法跳出局部最优,但是其相对于全局模型在局部区域具有较好的拟合效果。本文利用这两类模型的优点,针对计算昂贵问题提出了基于半监督学习代理模型的混合进化算法(SSL-SAHA)。在现有算法的基础上,对局部搜索部分进行了改进。利用在全局搜索过程中建立的集成模型选择一些未真实计算的个体,一起用于训练局部模型,从而提高局部RBF模型的估值准确度。实验结果表明,此算法可以有效求解计算昂贵问题。
It is difficult to get sufficient data to train a global surrogate model in a limited computational budget when the objective of an optimization problem is expensive to be evaluated.However,a potential advantage of the global surrogate model is that it can smooth out the local optima of the problem,which can speed up the method to find the optimal solution.On the other hand,a local surrogate model is not able to jump out from the local optimum,but it has a better performance to fit the original expensive problem in the local region.Thus,in this thesis,we propose a surrogate-assisted hybrid algorithm with semi-supervised learning(SSL-SAHA)for sovling expensive optimization problems.Based on the existing algorithms,the local search is modified.The surrogate ensemble built in the global search is used for choosing some solutions,which have not been evaluated using the exact expensive objective function,to join the local surrogate model training,being expected to improve the accuracy of the local RBF surrogate model.The results show that the proposed method can effectively solve the computationally expensive problem.
作者
任志海
李贞
Ren Zhihai;Li Zhen(Shanxi Electronic Science and Technology Institute,Linfen,China)
出处
《科学技术创新》
2024年第12期91-95,共5页
Scientific and Technological Innovation
基金
山西省高等学校科技创新项目(项目编号:2023L451)。
关键词
代理模型
元启发式算法
全局搜索
局部搜索
半监督学习
surrogate models
meta-heuristic algorithms
global search
local search
semi-supervised learning
