Hybridization of Differential evolution with Least-Square Support Vector Machines
Résumé
In this article we give a hybrid method for choosing ”good” individuals for the next generation of Differential Evolution (DE). Keeping DE to work as is, we add the Least-Square Support Vector Machine (LS-SVM) approximation in the end of each generation cycle. Such approximation uses a subset of selected individuals of the population. As a SVM core we choose a second order polynomial kernel function. The next individual is the optimum of the SVM approximation function which can be computed analytically as a solution of a system of linear equations. This method leads us to the improvement of algorithm convergence.
Domaines
Recherche opérationnelle [math.OC]
Origine : Fichiers produits par l'(les) auteur(s)
Loading...