global optimization, differential evolution, population diversity, parameter adaptation

A new method of the population-size adaptation for differential evolution (DE) is proposed. The adaptation is based on decreasing or increasing the population size in dependence on the current measure of the population diversity. The proposed method was implemented to six DE variants using a fixed population size in their original versions. The DE variants for experimental comparison were selected from the set of the state-of-the-art algorithms as well as from the well-performing variants that have appeared in literature recently. Experimental comparison was carried out on the CEC 2014 benchmark suite at three levels of problem dimension. The results of the experimental comparison demonstrate the benefit provided by the application of the proposed adaptive mechanism for the performance of the algorithms.