Journal of Frontiers of Computer Science and Technology ›› 2020, Vol. 14 ›› Issue (11): 1908-1919.DOI: 10.3778/j.issn.1673-9418.2006101

Previous Articles     Next Articles

Particle Swarm Optimization with Social Influence

SONG Wei, HUA Ziyu   

  1. 1. School of Artificial Intelligence and Computer Science, Jiangnan University, Wuxi, Jiangsu 214122, China
    2. Jiangsu Provincial Engineering Laboratory of Pattern Recognition and Computational Intelligence, Jiangnan University, Wuxi, Jiangsu 214122, China
  • Online:2020-11-01 Published:2020-11-09

融入社会影响力的粒子群优化算法

宋威华子彧   

  1. 1. 江南大学 人工智能与计算机学院,江苏 无锡 214122
    2. 江南大学 江苏省模式识别与计算智能工程实验室,江苏 无锡 214122

Abstract:

At present, particle swarm optimization (PSO) and its variants have been proven to be useful methods to solve complicated optimization problems. However, PSO and most of its variants only consider the impact of global best position and personal historical best position, which results in an issue of insufficient convergence and diversity. In this paper, PSO with social influence (PSOSI) is proposed to deal with this issue. Specifically, three strategies are employed in PSOSI. Firstly, each particle will choose two exemplars as its “social learning” part including the global best particle and its best companion particle, which brings more useful knowledge for each particle. Besides, gravity coefficient is introduced to describe the influence caused by the exemplars, enhancing the diversity of population while ensuring the best experience being shared. In addition, each particle further learns from its best companion particle on each dimension, realizing global to local variable-scale search, and enhancing the overall convergence ability. Moreover, 10 state-of-the-art PSO variants and 3 other typical optimization algorithms are com-pared with it on 28 benchmark functions of CEC2013 test suite. The experimental results demonstrate the superiority of PSOSI.

Key words: particle swarm optimization (PSO), social influence, social learning, gravity coefficient, variable-scale search

摘要:

目前粒子群优化(PSO)算法及其变体已被证明是有用的方法来求解复杂优化问题。然而,PSO及其大多数变体仅考虑全局最优位置和个体历史最优位置对个体的影响,导致算法的多样性不足,易于陷入局部最优。针对此问题,提出了一种融入社会影响力的粒子群优化(PSOSI)算法。具体而言,该算法设计了三种策略:首先,每个粒子将选择全局最优粒子和最优伙伴粒子作为其“社会学习”的对象,为每个粒子带来更丰富的有用知识;其次,引入引力系数分别描述各榜样带来的影响,在保证最优经验得到分享的同时,增强了种群学习的多样性;此外,每个粒子还在各维度上进一步向其最优伙伴粒子学习,实现从全局到局部的变尺度搜索,增强算法的整体收敛能力。实验对CEC2013测试集的28个基准函数进行测试,并与主流的10种PSO变体和3种非PSO优化算法进行比较,实验结果验证了PSOSI的优越性。

关键词: 粒子群优化(PSO), 社会影响力, 社会学习, 引力系数, 变尺度搜索