Footprinting the Behaviour of Particle Swarm Optimization with Increasing Dimensionality

It is well documented that the performance of Particle Swarm optimization changes (deteriorates) with increasing dimensionality of the search space. It is less well documented that the operational behaviour of Particle Swarm optimization (PSO) can also change with increasing dimensionality. The current study documents these changes by using Self-Organizing Maps to “footprint” the operation of PSO. Increasing dimensionality produces key changes to the footprints in a multi-modal search space, but these changes do not occur in a unimodal search space. A deeper analysis is then conducted to connect the observed changes in footprints in multi-modal search spaces to changes in the operational behaviour of PSO caused by the effects of increasing dimensionality. The collected data indicate a correlation between the performance degradation of PSO and the decreased rates of success of exploratory moves, and this trend can be isolated from the effects of the exponentially increasing search space volumes that are produced in higher dimensions for continuous domain search spaces.