Self-adaptive Clustering-Based Differential Evolution with New Composite Trial Vector Generation Strategies

Abstract

Differential evolution (DE) algorithms is a population-based algorithm like the genetic algorithms. But there are some problems in DE,such as slow and/or premature convergence. In this paper, a self-adaptive clustering-based differential evolution with new composite trial vector generation strategies (SaCoCDE) is proposed for the unconstrained global optimization problems. In SaCoCDE, the population is partitioned into k subsets by a clustering algorithm. And these cluster centers and the best vector in the current population are used to design the new differential evolution mutation operators. And these different mutation strategies with self-adaptive parameter settings can be appropriate during different stages of the evolution. This method utilizes the concept of the cluster neighborhood of each population member. The CEC2005 benchmark functions are employed for experimental verification. Experimental results indicate that CCDE is highly competitive compared to the state-of-the-art DE algorithms.