Multi swarm optimization algorithm with adaptive connectivity degree

Abstract

Particle swarm optimization algorithms are very sensitive to their population topologies. In all PSO variants, each particle adjusts it flying velocity according to a set of attractor particles. The cardinality of this set is a feature of neighborhood topology. In order to investigate this property exclusively, this paper defines the concept of connectivity degree for the particles and presents an approach for its adaptive adjustment. The presented approach is based on cellular learning automata (CLA). The entire population of the particles is divided into several swarms, and each swarm is resided in one cell of a CLA. Each cell of the CLA also contains a group of learning automata. These learning automata are responsible for adjusting the connectivity degrees of their related particles. This task is achieved through periods of learning. In each period, the learning automata realize suitable connectivity degrees for the particles based on their experienced knowledge. The empirical studies on a divers set of problems with different characteristics show that the proposed multi swarm optimization method is quite effective in solving optimization problems.