Using neighbourhoods with the guaranteed convergence PSO

摘要

The standard particle swarm optimiser (PSO) may prematurely converge on suboptimal solutions that are not even guaranteed to be local extrema. The guaranteed convergence modifications to the PSO algorithm ensure that the PSO at least converges on a local extremum at the expense of even faster convergence. This faster convergence means that less of the search space is explored reducing the opportunity of the swarm to find better local extrema. Various neighbourhood topologies inhibit premature convergence by preserving swarm diversity during the search. This paper investigates the performance of the guaranteed convergence PSO (GCPSO) using different neighbourhood topologies and compares the results with their standard PSO counterparts.