Differential evolution Parallel models Algorithm performance Experimental comparison

The influence of parallelism on the performance of competitive adaptive differential evolution is studied. Two serial competitive differential evolution variants described in literature and six adequate parallel variants (newly proposed) were experimentally compared. All the parallel differential evolution variants in this study are based on a migration model with the star topology. The algorithms were compared on six benchmark functions with two levels of dimension (D = 10 and D = 30). The number of the function evaluations and the reliability rate of the search were used as basic characteristics of algorithm?s performance. The experimental results show that the parallelism applied to competitive differential evolution together with a proper setting of the parameters controlling the parallel model can improve the performance of the algorithm, at least in some problems where the computational costs are ubstantially reduced.