Global approximation using adaptive regressive polynomial response surfaces with domain decomposition

Global approximation for a complex function or model with large domain can be applied in many fields such as sensibility analysis, parameter or control optimization and component simulation in a complex top-level system. The approximate substitute for the original model is also called response surface (RS). After analyzing the drawbacks of the existing response surface methods, propose an adaptive regressive polynomial response surfaces method using quadratic functions with domain decomposition. During the iteration for updating RS set through Latin hypercube testing, the roughest cells are selected and split along the roughest dimension direction to decompose the domain. After the RS set is accurate enough according to some criteria, the domain combination process is executed to unite the adjacent cells so that the number of the response surfaces in the RS set can be eliminated while keeping the given accuracy.