代理模型选样准则研究

致力于针对确定问题,利用其在已知空间中的样本信息指导该问题其他空间中的样本选择.基于Kriging模型和拉丁超立方设计选样方法,以某数学函数为例,首先研究了设计空间大小对样本选择的影响,然后具体分析了样本分布特性对代理模型预测误差的影响.引入了样本平均疏密度、样本稀疏度、样本紧凑度等概念衡量样本特性,依据不同空间中各参数与代理模型预测精度的关系提出了样本稀疏度准则,并以不同维数的数学函数和翼型气动力分析模型验证了准则的准确性和实用性.

Study of metamodel sampling criterion

For certain problem,a practical criterion which is to distill information from known space to guide sampling in another space is proposed in the paper.A mathematical function is taken as illustration to reflect influences of several factors based on Kriging and Latin Hypercude Design Sampling.The influence of design space to sampling is discussed firstly,and then the impact of samples distribution to metamodel prediction accuracy is analyzed.In order to describe the samples distribution characteristics,conc...