Abstract
Surrogate-assisted evolutionary algorithms (SAEAs) have received increasing attention in recent years. Kriging is one of the most popular surrogate models due to its ability to provide approximation uncertainty as well as the approximation value without additional computational cost. The kernel function used in the Kriging model plays a very important role in Kriging, as it has considerable influence on the approximation performance of Kriging models. However, little work has been dedicated to the efficiency of training the Kriging model and the influence of the kernel function on the performance of SAEAs. In this paper, we conduct extensive empirical experiments to examine the performance of Kriging-assisted SAEAs on a widely used test suite of benchmark problems of a dimension of 20 or 30 using different kernel functions. Detailed analyses of the training efficiency and the performance of Kriging-assisted SAEAs with different kernel functions are given.