The Particle Swarm Optimisation (PSO) algorithm has undergone countless modifications and adaptations since its original formulation in 1995. Some of these have become mainstream whereas many others have not been adopted and faded away. Thus, a myriad of alternative formulations have been proposed to the extent that the question arises as to what the basic features of an algorithm must be to belong in the PSO family. The aim of this paper is to establish what defines a PSO algorithm and to attempt to formulate it in such a way that it encompasses many existing variants. Therefore, different versions of the method may be posed as settings within the proposed unified framework. In addition, the proposed formulation generalises, decouples and incorporates features to the method providing more flexibility to the behaviour of each particle. The closed forms of the trajectory difference equation are obtained, different types of behaviour are identified, stochasticity is decoupled, and traditionally global features such as sociometries and constraint-handling are re-defined as particle's attributes.
- Pub Date:
- April 2021
- Computer Science - Neural and Evolutionary Computing
- Preprint The final authenticated article will be published by Springer-Nature in the Lecture Notes in Computer Science series. This research will be presented at the Twelfth International Conference on Swarm Intelligence (ICSI 2021)