Particle Swarm Optimization With Dynamic Neighborhood Structure

With the rapid development of the computer science and engineering technology, high-performance optimization technology and intelligence optimization is in urgent need. Particle swarm optimization is a new swarm intelligence stochastic optimization algorithm originating from artificial life and evolutionary computation. Because of its simple principles, few parameters, higher optimization efficiency and implemented easily, particle swarm optimization obtained rapidly the approval in the international evolutionary computation research area and has been successfully applied in many domains such as image manipulation, data mining,structural design and so on. But as a new random searching algorithm, PSO still suffers from premature convergence, especially for higher-dimension complex optimization problems.The research of complex network is also mature gradually and starts to seep to numerous different domains. In recent years, the small-world effect and scale-free characteristics of realistic network has aroused research upsurge to complicated network in academic circle. After reviews of the particle swarm optimization research and the related theories on complex network, the influence of neighborhood structure to PSO's convergence rate and the performance is discussed. Meanwhile, the small-world network model and free-scale network model are combined with particle swarm optimization, and a novel particle swarm optimization with dynamic neighborhood structure is proposed. The major works in the paper are given as follows: (1) Taking advantage of the major features of small-world network of "high clustering coefficient, small average shortest path length", the fixed small-world network model is introduced into particle swarm optimization, and a new particle swarm optimization based on small-world neighborhood structure (SWN-PSO) is presented, and the influence of small-world neighborhood structure to particle swarm optimization is analyzed. (2) The process of small-world neighborhood model generated is introduced into particle swarm optimization, and a novel particle swarm optimization with dynamic evolutionary neighborhood of the small-world model (DSWN-PSO) is given. In evolutionary process, population structure evolves from regular ring lattice to small-world network that search pattern is adjusted by controlling information communication between particles, and based on the diversity of the population, evolutionary opportunity of neighborhood structure is selected. (3) Referred to the scale-free network of"dynamic growing","preferential attachment", and property of small-world network, a particle swarm optimization with highly-clustered scale-free neighborhood (HCSN-PSO) is presented. The algorithm combines preferential attachment for degree and close nodes, a scale-free network model with high clustering is produced for subtle search. The benchmark functions are used in experiments, and the experimental results and theoretical analyses show that the new methods have improved convergent performance and efficiency of PSO obviously.