Improving the JADE algorithm by clustering successful parameters

Abstract

Differential evolution (DE) is one of the most powerful and popular evolutionary algorithms for real parameter global optimisation problems. However, the performance of DE greatly depends on the selection of control parameters, e.g., the population size, scaling factor and crossover rate. How to set these parameters is a challenging task because they are problem dependent. In order to tackle this problem, a JADE variant, denoted CJADE, is proposed in this paper. In the proposed algorithm, the successful parameters are clustered with the k-means clustering algorithm to reduce the impact of poor parameters. Simulation results show that CJADE is better than, or at least comparable with, several state-of-the-art DE algorithms.