Scale-free fully informed particle swarm optimization algorithm

This paper proposes a novel PSO algorithm, referred to as SFIPSO (Scale-free fully informed particle swarm optimization). In the proposed algorithm a modified Barabási–Albert (BA) model [4] is used as a self-organizing construction mechanism, in order to adaptively generate the population topology exhibiting scale-free property. The swarm population is divided into two subpopulations: the active particles and the inactive particles. The former fly around the solution space to find the global optima; whereas the latter are iteratively activated by the active particles via attaching to them, according to their own degrees, fitness values, and spatial positions. Therefore, the topology will be gradually generated as the construction process and the optimization process progress synchronously. Moreover, the cognitive effect and the social effect on the variance of a particle’s velocity vector are distributed by its “contextual fitness” value, and the social effect is further distributed via a time-varying weighted fully informed mechanism that originated from [27]. It is proved by the results of comparative experiments carried out on eight benchmark test functions that the scale-free population topology construction mechanism and the weighted fully informed learning strategy can provide the swarm population with stronger diversity during the convergent process. As a result, SFIPSO obtained success rate of 100% on all of the eight test functions. Furthermore, SFIPSO also yielded good-quality solutions, especially on multimodal test functions. We further test the network properties of the generated population topology. The results prove that (1) the degree distribution of the topology follows power-law, therefore exhibits scale-free property, and (2) the topology exhibits “disassortative mixing” property, which can be interpreted as an important condition for the reinforcement of population diversity.