Global and local selection in differential evolution for constrained numerical optimization

Abstract

The performance of two selection mechanisms used in the most popular variant of differential evolution, known as DE/rand/1/bin, are compared in the solution of constrained numerical optimization problems. Four performance measures proposed in the specialized literature are used to analyze the capabilities of each selection mechanism to reach the feasible region of the search space, to find the vicinity of the feasible global optimum and the computational cost (measured by the number of evaluations) required. Two parameters of the differential evolution algorithm are varied to determine the most convenient values. A set of problems with different features is chosen to test both selection mechanisms and some findings are extracted from the results obtained.