Island-cellular model differential evolution for large-scale global optimization

Abstract

The Island-Cellular Model (ICM) is an important population distribution approach for Evolutionary Algorithms (EAs). This hybrid approach combines the Island Model (IM) and Cellular Model (CM) in a two-layer hierarchical model. Although the ICM has been shown to be an efficient way to implement EAs, there is still a lack of knowledge about its parameters and its performance for Large-Scale Global Optimization (LSGO) problems. However, the ICM approach is able to enrich the evolutionary search by keeping the population diversity in EAs. Thus, this paper proposes to implement the ICM approach for LSGO problems using the Differential Evolutionary (DE) algorithm. It also proposes an experimental study of the ICM parameters by investigating their impact on the EA. Experimental results on Large-Scale Global Optimization Benchmark Functions show that the ICM approach for the DE algorithm improves its performance. Furthermore, the results collected from different ICM approaches indicate that there is a trade-off between solution quality and convergence speed.