Diversity-Based Multi-Population Differential Evolution for Large-Scale Optimization

There are increasing large-scale optimization problems in science and engineering nowadays. This paper proposes a diversity-based multi-population differential evolution (DB-MPDE) to maintain the population diversity, which is crucial for the large-scale optimizations. The performance of multi-population algorithms is sensitive to the exchanged information involved in the migration process. In our proposed DB-MPDE algorithm, information of diversity between sub-populations is utilized to determine the exchanged information. Both diverse and similar exchanged information is involved. Diverse exchanged information helps a lot in maintaining population diversity and similar exchanged information could accelerate convergence speed. In this way, the balance between global search and local search ability of the proposed algorithm can be achieved. A set of 20 benchmark functions is used to test the proposed DB-MPDE algorithm. Results show that the proposed DB-MPDE outperforms some well-known multi-population DE approaches.