A novel elite guidance-based social learning particle swarm optimization algorithm

基于精英引导的社会学习粒子群优化算法

¹ Air and Missile Defense College, Airforce Engineering University, Xi'an 710051, China
² School of Information Engineering, Binzhou Vocational College, Binzhou 256600, China

Received: 10 September 2023

Abstract

To improve the premature convergence and poor global search capability of the classical particle swarm algorithm(PSO), this paper proposes a novel elite guidance-based social learning particle swarm optimization (ESLPSO) algorithm. In the ESLPSO algorithm, a hierarchical topological search method is proposed. The method divides particles into optimal elite particles and other civilian particles according to their fitness performance, revolutionizing the update sample of the traditional population iterative search and enhancing the guidance of the whole population evolution information. In addition, an elite particle-guided social learning strategy is designed to better utilize the multidimensional information on population evolution by increasing the uncertainty of state superposition. On this basis, the extremum perturbation migration mechanism motivates the particles to experience new search paths and regions, increase population diversity and balance the population's exploration and exploitation in the search process. Moreover, the Cubic chaos initialization is employed to endow the initial particle population with a wide coverage in the search space. Finally, 12 benchmark test function sets covering unimodal, multimodal and rotated-multimodal functions are used to validate the performance of the proposed algorithm. The results on comparing the ESLPSO algorithm with other eight improved PSO algorithms show that the ESLPSO algorithm has excellent search performances in solving different types of functions, having efficient robustness and excellent solutions.

摘要

为了改进经典粒子群算法(PSO)过早收敛和全局搜索能力不足的缺点, 提出了一种基于精英引导的社会学习粒子群优化算法(ESLPSO)。在ESLPSO中, 提出了一种分层拓扑结构的搜索方法。这一策略根据粒子的适应度表现将粒子分化为最优的精英粒子和其余的平民粒子, 革新了传统种群迭代搜索的更新样本, 由此加强了整个种群演化信息的引导作用。采用Cubic混沌初始化赋予了初始粒子群体在搜索空间内的广域覆盖能力。设计了精英粒子引导的社会学习策略, 通过增加态叠加的不确定性更好地利用了种群演化的多维信息。在此基础上, 结合极值扰动迁移机制激励粒子经历新的搜索路径和区域, 增加种群的多样性, 平衡种群在搜索过程中的探索和开发能力。基于12个涵盖单峰、多峰以及旋转多峰的基准测试函数集对所提算法的性能进行了验证。此外, ESLPSO与其他8种PSO改进算法的比较结果表明, ESLPSO在解决不同类型函数方面表现出了优秀的搜索性能, 具有高效的求解稳定性和优异的求解结果。