Multi-population-based Algorithms with Different Migration Topologies and Their Improvement by Population Re-initialization.

In this paper, the possibilities of improving the performance of multi-population-based algorithms were tested. In the proposed approach, it was decided to test various methods and parameters of the population re-initialization mechanism, aimed at improving the diversity of individuals and preventing premature convergence, which is associated with a possible improvement in obtained results. In addition to the standard approach with random re-initialization of a new population, it was decided to test an approach in which selected populations are initialized with the use of modified individuals from better-performing populations. This approach has not been thoroughly tested so far, in particular for many different migration topologies and different population-based algorithms. The presented approach was specifically tested for the MNIA algorithm, eliminating the need to select one specific algorithm for the optimization. The simulations were performed for typical benchmark functions. The results of the simulations allow us to conclude that the proposed approach, depending on the parameters, improved the optimization process.