Transfer Knowledge Based Evolution of an External Population for Differential Evolution

Population size plays an important role in the optimization performance of Differential Evolution. Researches in earlier literature usually employed constant population size, and these recommended settings of different population sizes usually varied from one DE variant to another. As we know, smaller population size settings perform better on some objective functions while bigger settings perform better on the other within the same number of function evaluations. Therefore, adaptive schemes for population size became much more popular recently and performed very well on a large number of benchmark functions. These schemes dynamically changed the population size either in increasing or decreasing approaches during the evolution. Moreover, most of these adaptive schemes mainly focused on decreasing population size. Nevertheless, this paper reveals an approach to diversify the individuals (increase the population size) by employing an external population without increasing number of function calls. This approach employs transfer knowledge learned from the target population in the evolution of an external population for Differential Evolution. CEC2013 test suite for real-parameter single objective optimization is employed in the verification of our approach and experiment results show that the proposed approach is very useful in maintaining a better diversity of individuals without increasing function calls.