Adaptive population tuning scheme for differential evolution

Recently, various offspring generation strategies and parameter adaptation mechanisms have been developed to enhance the reliability and robustness of differential evolution (DE). However, the population size is generally fixed throughout the evolutionary search in most existing DE-variants, which leads to unsatisfactory performance. Based on the solution-searching status, in this paper, an adaptive population tuning scheme (APTS) for DE is proposed to dynamically adjust the population size. More specifically, on the basis of a ranking technique, a dynamic population strategy is adopted to remove redundant individuals from the population according to its ranking order. It is also applied to perturb the population and generate “fine” individuals. The proposed APTS is controlled by a status monitor, which is used to keep track of the progress of individuals and improve the performance of dynamic population strategy. In addition, this APTS framework is incorporated into several recently reported DE variants. The experimental results over 25 commonly used CEC2005 test functions demonstrate the effectiveness and usefulness of the proposed method.