On Dropping Niches in Parallel Niching Genetic Algorithms

s to solve for an "island" niching model are the selection of individuals for migration and the insertion of new individuals. A viable solution to select the migrants exploits the statistics computed on each island to collect the set of the best individuals for each cluster. To insert new individuals the simplest solution deletes a set of randomly chosen individuals. A more sophisticated strategy temporally expands the subpopulation and discards the exceeding individuals after the first generation. Detection of redundancy: after collecting the statistics, the information for each cluster can be summed up into two values: the centroid G(C c ) and its size \Pi c . When comparing clusters of distinct islands, two clusters are considered as the same niche if the distance between their centroids is smaller than the clustering parameter dmin . Choice of the niches to drop: when an