Differential Evolution (DE) is a simple but efficient Evolutionary Algorithm (EA) for stochastic real parameter optimization. With various types of mutation and crossover applicable to DE, there exist many variants of DE. The empirical comparisons between the performances of these variants on chosen benchmarking problems are well reported in literature. However, attempts to analyze the reason for such identified behavior of the variants are scarce. As an attempt in this direction, this paper empirically analyzes the performance as well as the reason for such performance of 14 classical DE variants on 4 benchmarking functions with different modality and decomposability. The empirical analysis is carried out by measuring the mean objective function values (MOV), success rate (Sr), probability of convergence (Pc), quality measure (Qm) and empirical evolution of the variance of the population (Evar). The study also includes reporting evidences for the variants suffering with stagnation and/or premature convergence. © 2014 S. Thangavelu, G. Jeyakumar and C. Shunmuga Velyautham.
cited By 0
Dr. Thangavelu S., Dr. Jeyakumar G., and Dr. Shunmuga Velayutham C., “Population variance based empirical analysis of the behavior of differential evolution variants”, Applied Mathematical Sciences, vol. 9, pp. 3249-3263, 2015.