具有自组织种群结构的微粒群算法  

Particle Swarm Optimization Based on Self-organization Topology

作　　者：莫思敏[1,2] 曾建潮[2] 徐卫滨[3] 

机构地区：[1]兰州理工大学电信工程学院,兰州730050 [2]太原科技大学复杂系统和计算智能实验室,太原030024 [3]太原科技大学经济与管理学院,太原030024

出　　处：《系统仿真学报》2013年第3期445-450,共6页Journal of System Simulation

基　　金：山西省青年基金(2011021019-3)

摘　　要：为了提高微粒群算法(PSO)的性能,模仿人类解决问题时的交互方式,提出一种基于适应值驱动、以朋友机制局部择优自组织种群结构的微粒群算法。以环形结构作为算法的初始结构,每代算法执行后,微粒根据其适应值,采用不同的阈值来决策是否需要建立新连接,并通过选择邻居的邻居中适应值最优且优于其本身的微粒,以概率P进行连接。通过这样的演化机制,逐步增加结构的平均聚集系数和降低平均路径长度,以达到近似小世界特性的网络结构。实验结果表明无论采用何种阈值,概率P对结构演化过程及算法性能的影响都非常大。在适当的概率P值下,基于该演化机制的微粒群算法获得了比基于其他结构的微粒群算法更好的性能。To improve the performance of Particle Swarm Optimization （PSO）, PSO based on self- organization topology （SOT-PSO） was proposed by imitating the interaction ways that people solved problems. A ring lattice was used as initial structure of algorithm. After the implementation of PSO, the particle applied different threshold values to decide if new connections need to be established according to its fitness. Other particle that was the best one and meanwhile better than itself among neighbor＇s neighbors of the particle was linked by the probability P if needed. Thus, the average clustering coefficients was increased gradually and the average path length was decreased gradually. The approximately small-world network structure was formed in the end. Simulation results demonstrate that the probability P has significant influences on both the population structure evolution and the performance of PSO regardless of different threshold values used and SOT-PSO is almost superior to all other variants of PSO when the probability P is selected properly.

关 键 词：微粒群算法 小世界网络 朋友机制 自组织种群结构 

分 类 号：TP301.6[自动化与计算机技术—计算机系统结构]