Comparative Studies on Typical Opposition-Based Learning Schemes in Differential Evolution with Gaussian Perturbation

Abstract

AbstractSome optimization problems have been well solved after integrating the concept of opposition-based learning (OBL), and opposition-based differential evolution (ODE) is a typical representative. ODE can be seen as a rigorous point-to-point algorithm, which may lead to neglect of some good solutions in the neighborhood of the opposite point. It provides a relatively narrow search area for candidate solutions and is not enough to maintain population diversity. Gaussian perturbation strategy is proposed to implement around the opposite point to expand the search neighborhood of the opposite point and to find more and better solutions. A stochastic centroid opposition (SCO) is firstly proposed on the basis of centroid opposition (CO) in this paper. Then four typical OBL schemes, namely, the original OBL, generalized OBL, centroid opposition and stochastic centroid opposition, are embedded in differential evolution (DE) with Gaussian perturbation strategy. Comparison experiments based on CEC 2014 benchmark suite are reported and the results indicate that the opposition algorithms with Gaussian perturbation strategy based on individuals’ dynamic boundaries perform better than the corresponding centroid algorithms.KeywordsOpposition-based learning (OBL)Differential evolutionGaussian distributionNeighborhood searchOpposition-based computation