Byeong-Soo Kim
The Journal of Mechanical Science and Technology, vol. 23, no. 4, pp.1175-1181, 2009

Abstract : In order to increase the efficiency of design optimization, many efforts have been made on studying the metamodel techniques for effectively representing expensive and complex models. In this study, a comparison is conducted on the accuracy of several widely used meta-model techniques - moving least squares (MLS), Kriging, support vector regression (SVR) and radial basis functions (RBF) - which are able to approximate non-convex functions well.. RMSE (root mean squared error) value is identified as a measure of the accuracy for this comparison. Each metamodel technique is used to approximate the six well-known mathematical functions and a resign of experiment (DOE) is generated by using the Latin hypercube design (LHD), which is also performed for each resulting metamodel. The results show that Kriging and MLS can create a more accurate metamodel than SVR and RBF with the mathematical functions tested.

Keyword : Metamodel; Moving least squares; Kriging; Radial basis function; Support vector regression