Random Asynchronous PSO

In this work we propose the Random Asynchronous PSO (RAPSO) algorithm, a rather simple but intuitive variant of the Asynchronous PSO (APSO) that introduces a randomized order in which particles share their information. Our algorithm, while conceived as serial in terms of execution, is able to model the behavior of the Parallel Asynchronous PSO (PAPSO) but in a controlled environment where results from independent runs are reproducible and the implementation does not need additional computational requirements (e.g. computational grids or GPUs). We support our proposal by comparing it favorably with respect to the original APSO in ten well-known benchmark functions. Statistical tests show that RAPSO generally yields better results (or at least not worse) and generally faster than APSO, making it a more attractive choice to tackle optimization problems where asynchronous updates is a desirable property.