An Empirical Study of Control Parameters for the Third Version of Generalized Differential Evolution (GDE3)


Abstract

In this paper the influence of control parameters to the search process of previously introduced Generalized Differential Evolution 3 (GDE3) is empirically studied. GDE3 is an extension of Differential Evolution (DE) for constrained multi-objective optimization. Besides the number of generations and the size of the population, GDE3 has two control parameters, CR and F, which have to be set before applying the method for solving a problem. The effect of these two control parameters is studied with a set of bi-objective test problems and performance metrics. The objectives of this study are to investigate the sensitivity of the control parameters according to different performance metrics, to understand better the effect and tuning of the control parameters on the search process, and to compare results with the corresponding results of the first version of GDE. A similar non- linear relationship between the control parameters and performance metrics was observed as in earlier studies with the first version of GDE. According to the relationship a larger F value can be used with a small CR value than with a large CR value.