Decentralized Asynchronous Particle Swarm Optimization

In this article we discuss a decentralized totally asynchronous realization of the particle swarm optimization (PSO) algorithm, which is suitable for parallel implementation. The proposed method has important differences from the PSO implementations considered in the literature. In the proposed method the particles are allowed to exchange information and to update their estimates at totally independent time instants. Moreover, time delays during information exchange between particles (leading to use of outdated information) are also allowed. Furthermore, particle neighborhoods are allowed to dynamically change with time. We also provide a mathematical model of the proposed method based on results in the parallel and distributed computation literature. The performance of the proposed algorithm is tested using numerical simulations with benchmark functions.