A Kriging-based global optimization method using multi-points infill search criterion

Abstract

The efficient global optimization algorithm-based Kriging is adaptive to solve expensive simulation optimization problems. Expected improvement criterion with minimum predicted objective and maximum standard deviation in efficient global optimization is optimized to find the next optimum. However, the single-point infill sampling criteria and multimodality of expected improvement will result in a large number of simulation time consumption, low search efficiency, and poor convergence performance. To improve this situation, a Kriging-based global optimization method using multi-points infill search criterion is proposed. Unlike efficient global optimization, it uses multi-objective optimization methods to minimize the objective estimation and maximize standard deviation estimation for Kriging model. The criterion selects and screens multiple update sampling points based on Pareto optimal solutions from the Pareto front. The optimization method is tested by the nine numerical problems and an engineering simulation application. In contrast with efficient global optimization, the proposed method is able to deliver good optimization results in search efficiency and convergence accuracy.