A two-stage infill strategy and surrogate-ensemble assisted expensive many-objective optimization

Abstract

Many optimization problems are expensive in practical applications. The surrogate-assisted optimization methods have attracted extensive attention as they can get satisfyingly optimal solutions in a limited computing resource. In this paper, we propose a two-stage infill strategy and surrogate-ensemble assisted optimization algorithm for solving expensive many-objective optimization problems. In this method, the population is optimized by a surrogate ensemble. Then a two-stage infill strategy is proposed to select individuals for real evaluations. The infill strategy considers individuals with better convergence or greater uncertainty. To calculate the uncertainty, we consider two aspects. One is the approximate variance of the current surrogate ensemble and the other one is the approximate variance of the historical surrogate ensemble. Finally, the population is revised by the recently updated surrogate ensemble. In experiments, we testify our method on two sets of many-objective benchmark problems. The results demonstrate the superiority of our proposed algorithm compared with the state-of-the-art algorithms for solving computationally expensive many-objective optimization problems.