Differential evolution with self-adaptive Gaussian perturbation

Abstract

Differential evolution is a population-based metaheuristic that is widely used in Black-Box Optimization. The mutation is the main search operator and there are different implementation schemes reported in state of art literature. Nonetheless, such schemes lack mechanisms for an intensification stage, which can enable better search and avoid local optima. This article proposes a way to adapt the Covariance Matrix parameter of a Gaussian distribution that is used to generate a disturbance that improves the performance of two well-known mutation schemes. This disturbance allows working with problems with correlated variables. The test was performed over the CEC 2013 instances and the results were compared through the Friedman nonparametric test.