A kriging-based sequential optimization algorithm with hybrid infill sampling strategy

Abstract

The appropriate infill sampling criteria to generate more promising updated points is crucial for kriging-based global optimization. For this purpose, a kriging-based sequential optimization algorithm with hybrid infill sampling strategy (KSO-HIS) is proposed. In each iteration, three efficient sampling criteria (i.e., predicted objective minimization criterion, improved expected improvement and new curvature maximization criterion) based on kriging are respectively optimized to produce three optimal solutions by TR (Trust Region) method. Then, a new screening strategy is adopted to determine final expensive evaluation points from the three optimal solutions. The proposed method is compared with three other optimization methods. The test results of eight benchmark functions and a simulation case verify that KSO-HIS can deliver better sampling and convergence performance.