정확한 크리깅 모델을 생성하기 위한 상관계수 선정 방법

Abstract

Design analysis and computer experiments (DACE) model, specifically kriging surrogate model, is widely used to describe nonlinear responses and utilized to substitute highly expensive model in engineering design. Correlation coefficient considerably influences on the predictivity of kriging surrogate model. Even though the correlation coefficient is highly associated with the nonlinearity of response, in optimization of the correlation coefficient, same initial value without such consideration has been used on previous research. This decreases the accuracy of the model and increases the possibility of convergence to local optimum. Therefore, in this study, initial value of correlation coefficient is determined by considering nonlinearity of the response and the correlation coefficient is estimated through maximum likelihood estimation (MLE) using sensitivity based optimization. Previous and proposed kriging surrogate models are compared by maximum likelihood function value and root mean squared error value.