Mechanisms to Avoid the Premature Convergence of Genetic Algorithms

The optimization by genetic algorithms often comes along with premature convergence bias, especially in the multimodal problems. In the paper, we propose and test two mechanisms to avoid the premature convergence of genetic algorithms by preserving the population diversity in two different manners. These are the dynamic application of many genetic operators, based on the average progress, and the population partial reinitialization. The mechanisms were tested by implementing them in the NSGA_II algorithm, applied to one of the most difficult job shop scheduling test problems, ft10. The comparative analysis between the new algorithm and the NSGA_II in the absence of the submitted mechanisms, alongside with an elitist and the canonic genetic algorithm, proves the usability of both proposed mechanisms.