Weibull-based scaled-differences schema for Differential Evolution

Differential Evolution is one of the most efficient real-parameter optimization algorithm. It is based on the application of the scaled difference of a pair of population members to another population member, all of them distinct. Diverse variants have been proposed within this schema. In this work, the statistical distribution of these differences of high-performance variants of Differential Evolution is modelled through a Weibull probability distribution. From the application of this model to diverse Differential Evolution variants and benchmark functions, a pattern for the most efficient variants can be drawn. As a consequence, a variant where the scaled differences are replaced by random numbers generated from a Weibull distribution is proposed and evaluated.