Particle Swarm Optimization - Parameter Selection

Parameter Selection

The choice of PSO parameters can have a large impact on optimization performance. Selecting PSO parameters that yield good performance has therefore been the subject of much research.

Basically, it can be imagined that the function which is to be minimized forms a hyper-surface of dimensionality same as that of the parameters to be optimized (search variables). It is then obvious that the 'ruggedness' of this hyper-surface depends on the particular problem. Now, how good the search is depends on how extensive it is, which is decided by the parameters. Whereas a 'lesser rugged' solution hyper-surface would need fewer particles and lesser iterations, a 'more rugged' one would require a more thorough search- using more individuals and iterations. This is analogous to another realistic situation of flocks searching for a good 'food' traversing a very difficult terrain containing gardens all over, some better than others where a hugely populated flock would be inevitable in order to reach the best (read global optimum) 'food' source, compared to another terrain where there are very few gardens on an otherwise non-vegetated land, where it becomes easy to search for 'food' and lesser number of individuals and iterations will suffice.

The PSO parameters can also be tuned by using another overlaying optimizer, a concept known as meta-optimization. Parameters have also been tuned for various optimization scenarios.