Diversity-Driven Selection Operator for Combinatorial Optimization

Abstract

A new selection operator for genetic algorithms dedicated to combinatorial optimization, the Diversity Driven selection operator, is proposed. The proposed operator treats the population diversity as a second objective, in a multiobjectivization framework. The Diversity Driven operator is parameterless, and features low computational complexity. Numerical experiments were performed considering four different algorithms in 24 instances of seven combinatorial optimization problems, showing that it outperforms five classical selection schemes with regard to solution quality and convergence speed. Besides, the Diversity Driven selection operator delivers good and considerably different solutions in the final population, which can be useful as design alternatives.