Abstract:

A two-stage implementation strategy based on canonical particle swarm optimization was proposed. With the cost of acceptably additional evaluations, this strategy achieved higher success rate which were demonstrated by a suite of benchmark functions. The simulation showed that two-stage implementation strategy could bring forward relatively higher success rate under different upper limitation of iterations. At the same time the proposed strategy reduced the sensitivity of learning rate and presented a stable performance. It was revealed experimentally that the number of sub-populations should be set at a moderate value to consider both the reliability and the computation cost.

Key words: particle swarm optimization, optimization algorithm, global optimization, computational intelligence, swarm intelligence