Dynamics analysis on a self-organized particle swarm optimization

This paper proposes a new self-organized particle swarm optimization (SOPSO). In the algorithm, particles can adjust dynamically its moving mode based on the swarm states, so that the algorithm has high efficiency to solve the test functions. The dynamics of the algorithm exhibits a `heavy tail´ distribution. The distribution of a parameter which reflects the search range in the solution space of the algorithm has a tail that resembles the Levy-flight. This power law distribution demonstrates that the SOPSO has the properties of the self-organized criticality, so that the search pattern is characterized by many small range scans connected by larger range reorientation jumps.In this way, a good balance between small range search (local exploit) and large scale explore (global explore) can be achieved. The paper also investigates the dynamics properties of two other standard PSOs (GBest and LBest) which trapped into the local optimum when searching the solution. The tails of these two PSOs´ descend faster than that of the SOPSO, which means the ability of these two PSOs to global explore is limited so that they are easy to be trapped in the local optimum.