Adaptive multi-context cooperatively coevolving in differential evolution

Abstract

This paper presents an adaptive multi-context cooperatively coevolving differential evolution (AMCC-DE) algorithm, in order to address the issue of scaling up differential evolution algorithms on large-scale global optimization (LSGO) problems. The proposed AMCC-DE builds on the success of an early AMCCPSO in which the adaptive multi-context cooperatively coevolving (AMCC) framework is employed. In the proposed AMCC-DE, several superior individuals are employed as the multiple context vectors (CV) to provide robust and effective coevolution, and these CVs are selected by each individual based on their adaptive probabilities. To keep the diversity of these CVs, the mutation operation of CV is defined and conducted in each generation. Moreover, a new mutation operator is also proposed and employed in the AMCC-DE to generate promising individuals. On a comprehensive set of 1000-dimensional LSGO benchmarks, the performance of AMCC-DE compared favorably against some state-of-the-art evolutionary algorithms. Experimental results indicate that the proposed AMCC-DE is effective on LSGO problems, and the proposed mechanisms in AMCC-DE can also be generally extended to other EAs.