自适应细分二次多项式响应面算法

Adaptive Regressive quadratic polynomial Response Surfaces with Domain Decomposition

复杂模型(或黑箱函数,black-box function)的全局近似方法广泛应用于模型简化、灵敏度分析和参数(或控制)优化中。现有的单一响应面存在着采样估值与近似精度之间的矛盾,因此提出自适应细分二次多项式响应面算法:在构造初始响应面基础上,对设计域进行递归分割,每步选择最粗糙单元,沿着最粗糙维度方向进行自适应细分。通过实验测试,该方法以多响应面来近似黑箱函数,从而达到采样点少而响应面精度高的目标。

Global approximation for a complex model（or black-box function ）can be applied in many fields such as model simplification, sensibility analysis,parameter（or control） optimization.Using existing single response surface methods,there are some contradictions between Sampling valuations and approximation.An adaptive regressive quadratic polynomial response surfaces method is proposed： On the basis of construction of original response surface,recursive dividing the domain,the roughest cells are selected and split along the roughest dimension direction to decompose the domain.Through The test,the method can approximate black -box function to less sampling points and high accuracy.