用于不确定性分析的高斯过程响应面模型设计点选择方法

Design point choice of Gaussian process response surface model applied to uncertainty analysis

为推动高斯过程响应面模型在复杂耗时数值模拟不确定性分析中的应用,提出一种可自动实现位置优化的高效设计点选择方法,即先在标准超立方体上生成拉丁超立方设计点,然后利用输入变量的已知概率分布将其映射回原始设计空间.将该方法与基于假设均匀分布的传统拉丁超立方设计方法进行比较,探讨它们对所建立的高斯过程响应面模型和不确定性分析结果的影响.算例表明新方法具有一定优势.

An efficient method of design point choice is presented,which can automatically optimize the location of the points,to advance the application of Gaussian process response surface model to uncertainty analysis of the complex time-consuming numerical simulation.The method produces the design points in a standard Latin hypercube,and then maps them back to the original design space according to the known probability distribution of input variables.The method and the traditional Latin Hypercube Design method based on an assumed uniform distribution are compared and their effect on the established Gaussian process response surface model and uncertainty analysis result is discussed.An example indicates the advantage of the method.