Surrogate-Assisted Ensemble Social Learning Particle Swarm Optimization

Abstract

The surrogate-assisted optimization algorithm is a method to solve expensive optimization problems by constructing a predicted evaluation model to replace the real objective function. The evaluation of the objective function is generally time-consuming and the target is to use a small amount of exact function evaluation (EFE) to achieve better solutions in shorter time. The generation and selection of the candidate solutions to have the EFE are the most important. For the generation of candidate solutions, this paper proposes an ensemble method for two state-of-the-art models, i.e. the Gaussian process (GP) and the Radial basis function (RBF) model to have better prediction of the solutions. For the selection of candidate solutions, the traditional similarity-based multipoint infill criterion (SMIC) strategy is modified and the proposed method is termed the best SMIC (bSMIC). The social learning particle swarm optimization (SLPSO) algorithm is used as the basic optimization algorithm. The effectiveness of the proposed ensemble surrogate-assisted SLPSO (ESLPSO) has been fully analyzed in various problems and compared with other algorithms.