Kernel-based gradient evolution optimization method

Abstract

A new hybrid metaheuristic method is proposed which is inspired from the recently proposed gradient-evolution method. This study introduces for the first time the concept of reproducing kernel in order to correctly estimate the numerical gradient which is used as an updating rule. This proposed method is called Kernel-based Gradient Evolution (GE) which inherits good exploration and exploitation abilities as well as non localization through several operators taken from the original gradient-evolution algorithm. In the proposed KGE algorithm the gradient vector is computed using a kernel reproducing function. Thus the local gradient estimation is numerically computed considering a local Taylor series expansion. The IEEE CEC 2019 test suite is considered in order to evaluate the numerical performance of the proposed KGE algorithm. A numerical comparison between some competitive methods and the one proposed in this work is reported. The obtained numerical results suggested that in terms of convergence, KGE algorithm performs in most cases significantly better than the original GE algorithm and gets respectable results against other considered methods.