A GENE EXPRESSION PROGRAMMING BASED KRIGING METHOD FOR METAMODEL CONSTRUCTION

摘要

To reduce the computational time and cost of running computer-based simulation experiments, metamodels are becoming more and more popular for replacing the simulation codes for design and optimization. In this paper, a gene expression programming (GEP) based kriging method is proposed. In this method, the GEP algorithm is used to create regression functions in kriging metamodels. An asymmetric function is taken as an illustrative example to prove the validity and effectiveness of the proposed method. Compared to the GEP and original kriging methods, the proposed method can achieve more accurate approximation of a high-dimensional design space. The GEP based kriging method can not only improve the prediction performance of the original kriging method if strong trends exist in the original function relationships, but also deal with the approximation of the high-dimensional design space, where GEP shows poor performance.