A Clustering Based GA for Multimodal Optimization in Uneven Search Space

A new DBSCAN clustering based niching GA (DCNGA) algorithm that used to locate all global optima in uneven multimodal domain is presented in the paper. In order to remove some a priori knowledge requirements of standard sharing, such as peaks number and sharing distance, DCNGA using DBSCAN clustering algorithm to identify the niche automatically, and dynamic clustering niching is also used to form the steady proportionate population by adjusting fitness based upon the dynamic niche population size. Moreover, the mating restriction scheme is also used to build up the exploitation of sub-population. With these strategies we are able to preserve diversity within the population and to identify global or local optima without further post-processing. Experiment results demonstrate that DCNGA not only can detected all global optimal of three benchmark problems successfully, but also some best local optima can be found in the final populations.