Incorporating more scaled differences to differential evolution

Abstract

Differential Evolution is an evolutionary algorithm composed of vectors and based on the application of scaled differences of two vectors over a third one, being all of them different. The variants of this algorithm propose different types of vectors for the scaled difference, and different number of scaled differences, to alter differently-selected vectors. The successful track of Differential Evolution has propitiated numerous variants. These variants use a limited number of vectors for forming the scaled differences and, in general, only one vector for receiving these differences. In this work, new variants with scaled differences using all the population vectors are proposed. These variants are confronted to a wide set of fitness functions and to a set of Differential Evolution variants.