计算机科学与探索 ›› 2014, Vol. 8 ›› Issue (2): 218-225.DOI: 10.3778/j.issn.1673-9418.1306011

• 人工智能与模式识别 • 上一篇    下一篇

差分和扰动混合的多策略粒子群优化算法

赵新超+,刘子阳   

  1. 北京邮电大学 理学院,北京 100876
  • 出版日期:2014-02-01 发布日期:2014-01-26

Hybrid Particle Swarm Optimization with Differential and Perturbation

ZHAO Xinchao+, LIU Ziyang   

  1. School of Science, Beijing University of Posts and Telecommunications, Beijing 100876, China
  • Online:2014-02-01 Published:2014-01-26

摘要: 通常的粒子群优化算法采取单一的学习策略,不利于搜索信息的有效保留,因此将改进的差分变异策略引入到粒子的速度更新中以增强算法的群体多样性;综合利用差分变异与扰动策略两种不同的产生新解的方式,提出了一种多策略交叉学习机制算法DPPSO(hybrid particle swarm optimization with differential and perturbation)。每一个粒子通过引进的差分变异操作和扰动操作分别产生一个中间粒子,再选择较好的粒子作为当前粒子的新位置,从而实现所有粒子动态地选择更好的生成策略来更新自己的位置和速度,因此该交叉策略能够有效提高PSO算法的群体多样性和搜索路径的多样性,粒子可以获取更好的启发式信息,沿着不同的路径被引向更有潜力的搜索区域。实验结果表明了两种策略的有效性和互补性,DPPSO算法比其他三种算法有更好的综合表现,具有有效的全局收敛能力和准确定位能力。

关键词: 粒子群优化(PSO), 多策略, 差分变异, 扰动策略, 数值优化

Abstract: Most particle swarm optimization (PSO) algorithms use a single learning pattern for all particles, which does not benefit to save the heuristic information. This paper introduces a modified differential mutation strategy to update the velocity of particle in order to diversify the PSO population. Then this paper proposes a multiple strategies interactive learning mechanism based PSO (DPPSO), which combines differential mutation with the perturbed particle updating strategy as the updating pattern. Each particle generates two intermediate particles from two strategies and selects a better one as its new position. Therefore, each particle can dynamically select its biased generation strategy to diversify the particle population and the searching trajectories. The particles can obtain more beneficial heuristic information, which guides particles to the promising search area. The experimental results indicate the effectiveness and the reciprocal reinforcement of two strategies. Compared with other PSO variants, DPPSO has the best comprehensive performance, with powerful global exploration and fine locating abilities.

Key words: particle swarm optimization (PSO), multiple strategies, differential mutation, perturbation strategy, numerical optimization