Universal strategy of dynamic subpopulation number management in practical network optimization problems

Abstract

Multi-population approach, also known as Island Model (IM), is a well-known technique improving diversity preservation. It is employed by evolutionary and Swarm Intelligence methods. The diversity increased through the maintenance of many subpopulations, leads to finding higher quality solutions. Since IM is computationally expensive, it is important to keep the balance between the diversity maintenance and the convergence. Therefore, this paper concentrates on proposing the new strategy of Dynamic Subpopulation Number Control (SDSNC). When SDSNC is used, the number of subpopulations may change during the method run, adjusting the subpopulation number to the current method state, making such method highly-effective. In this paper, we show that the SDSNC strategies that were proposed for two up-to-date, NP-hard, practical problems are not universal. Therefore, we propose Store and Use Later (SUL) strategy that applied to problem-dedicated methods outperforms or at least equals two other strategies for both considered problems.