A Diversity-guided Distributed Evolutionary Algorithm

Evolutionary algorithms are consistently troubled by early convergence to suboptimal solutions, and this problem is just as prevalent with distributed evolutionary algorithms (dEAs). One of several approaches used to address this issue (but still not much explored, especially in dEAs) is the use of diversity-guided control, where population diversity is monitored and steps are taken to boost diversity when necessary. Here, we explore such a mechanism in a dEA context. We monitor the diversity of subpopulations, and migrate suitable chromosomes into them when their diversity is low. This builds on previous work where we have explored a number of topology and migration mechanisms; in the current work, we investigate whether diversity-guidance can provide additional benefits. Experimental results are presented on twelve difficult benchmark function optimization problems. The results suggest that diversity guidance provides additional benefits, especially in the higher dimensional cases, although the method does not outperform a recent alternative strategy, in which the migration of new chromosomes was based on monitoring progress, rather than diversity.