The latest vs. averaged recent experience: Which better guides a PSO algorithm?

Abstract

A particle swarm optimization strategy based on the use of learned experiences averaged over a number of iterations is presented. The personal and the global best solutions over a number of latest iterations are stored and averages of the stored solutions are used in the velocity computations. Experiments on real-parameter optimization problems published in CEC 2005 test suite demonstrate that the proposed strategy exhibits better performance than conventional PSO for most of the benchmarks, whereas the conventional PSO performed better for only the two non-continuous test cases.