differential evolution, adaptation, population size, population diversity, experiments

A new mechanism for the adaptation of the population size in differential evolution (DE) is described. The adaptive mechanism is based on linear reduction of the population diversity and enables to decrease and also increase the population size during the search. The efficiency of DE variants with and without this adaptive mechanism is experimentally compared on the CEC 2014 test suite. Moreover, this new mechanism is also compared with the linear reduction of the population size proposed in L-SHADE and used in its successful successors. The results of the comparison demonstrate the benefit from the new adaptive mechanism to the algorithm efficiency in more than a half of test problems while the deterioration of its efficiency occurs rarely. As it was expected, the use of the proposed mechanism is considerably beneficial, especially in more complex problems like multimodal, hybrid, or composition objective functions.