PARTICLE SWARM OPTIMIZATION WITH ADAPTIVE SOCIAL AND COGNITIVE COMPONENTS
Efficiency of solution finding by particle swarm optimization depends significantly on specific values of social and cognitive components used by a researcher. There is no known way currently to determine whether specific values of the components would provide maximal search efficiency in a particular case, or not. In order to eliminate this flaw, this article provides a modification of particle swarm optimization with adaptive social and cognitive components, which allows to fit particles movement to a particular problem during optimization process, thus removing the need of adjusting components manually. This adaption is based on genetic algorithms principles: it starts with a selection of the best performing particles, then crossover of their social and cognitive components with other particles, then mutation to provide some fluctuations of components. To evaluate algorithm’s performance a series of experiments on minimizing few test functions has been made. Minimums found by adaptive and canonical algorithms were averaged out and compared. Based on results, a statistical hypothesis that adaptive algorithm has better performance than canonical algorithm was confirmed. Provided research proves efficiency of adaptive particle swarm.
Keywords: mathematical optimization, particle swarm optimization, adaptation, genetic algorithms.