A hybrid surrogate co-assisted evolutionary algorithm with prediction fusion interpolation sampling strategy for expensive optimization problems

Abstract

Hybrid surrogate-assisted evolutionary algorithms, which achieve the purpose of assisting search by hybridizing local and global surrogate models, are a kind of competitive state-of-the-art techniques for solving expensive optimization problems (EOPs). The sampling points of the local models have been introduced but are failed to investigate directly in this field. In fact, sampling points are one of important determinants of model performance. Unlike the existing technologies including superior individuals sampling and neighboring individuals sampling, this paper develops a prediction fusion interpolation sampling strategy (PFs) and proposes a hybrid surrogate co-assisted evolutionary algorithm with it (HSCEAwP). The presented PFs applies all the best predictions of the local and global models of all historical populations as the sampling points of the next local surrogate model. The proposed HSCEAwP inherits the optimization framework of the generalized multifactorial evolutionary algorithm. The radial basis function model is chosen as the modeling basis of the local and global surrogate models. The performance of PFs under radial basis function model is analyzed theoretically and experimentally based on the interpolation principle. The performance of HSCEAwP is tested on eight common benchmark problems, ten CEC2017 composition problems and an electrostatic precipitator optimization problem. The experimental results demonstrate more reliable performance of HSCEAwP to well-established algorithms in terms of solving accuracy.