Colonial multi-swarm: A modular approach to administration of particle swarm optimization in large scale problems

In this paper we have introduced the colonial multi-swarm, an algorithm with modular characteristics that can be augmented on several existing variants of Particle Swarm Optimization to alleviate the premature convergence problem. Colonial multi-swarm ensures a decent degree of exploration by administrating a number of parallel swarms. It uses a meta-level decision system that allows the particles to switch their swarms when they are not successful in their own. This migration strategy helps preserve the local exploitation of thriving swarms as well as providing diversity and ground for emerge of other swarms. The module has been implemented on conventional PSO. The method is tested on 28 benchmark problems and has exhibited significant improvement over the original algorithm and also notable merits compared to the latest version of Standard PSO (SPSO 2011), presented by the particle swarm central, in large scale problems. It is shown that application of colonial multi swarm exhibits bolder improvement over the mentioned algorithms as the problem dimensionality rises.