A fast surrogate-assisted particle swarm optimization algorithm for computationally expensive problems

Although many surrogate-assisted evolutionary algorithms (SAEAs) have been proposed to solve computationally expensive problems, they usually need to consume plenty of expensive evaluations to obtain an acceptable solution. In this paper, we proposed a fast surrogate-assisted particle swarm optimization (FSAPSO) algorithm to solve medium scaled computationally expensive problems through a small number of function evaluations (FEs). Two criteria are applied in tandem to select candidates for exact evaluations. The performance-based criterion is used to exploit the current global best and accelerate the convergence rate, while the uncertainty-based criterion is used to enhance the exploration of the algorithm. The distance-based uncertainty criterion in SAEAs does not consider the fitness landscape of different problems. Therefore, we developed a criterion to estimate uncertainty by considering the distance and fitness value information simultaneously. This criterion can make up for the disadvantage of the conventional distance-based uncertainty criterion by considering the fitness landscape of a problem. In addition, it can be applied in any surrogate-assisted evolutionary algorithm irrespective of the used surrogate model. Twenty-three benchmark functions widely adopted in the literature and a 10-dimension propeller design problem are used to test the proposed approach. Experimental results demonstrate the superiority of the proposed FSAPSO algorithm over seven state-of-the-art algorithms. (C) 2020 Elsevier B.V. All rights reserved.