An Analysis of Sub-swarms in Multi-swarm Systems

Particle swarm optimization cannot guarantee convergence to the global optimum on multi-modal functions, so multiple swarms can be useful. One means to coordinate these swarms is to use a separate search mechanism to identify different regions of the solution space for each swarm to explore. The expectation is that these independent subswarms can each perform an effective search around the region where it is initialized. This regional focus means that sub-swarms will have different goals and features when compared to standard (single) swarms. A comprehensive study of these differences leads to a new set of general guidelines for the configuration of sub-swarms in multi-swarm systems.