计算机工程与应用 ›› 2017, Vol. 53 ›› Issue (16): 116-120.DOI: 10.3778/j.issn.1002-8331.1603-0122

• 模式识别与人工智能 • 上一篇    下一篇

自适应K值的粒子群聚类算法

白树仁1,2,陈  龙2   

  1. 1.湖南大学 湖南省超级计算长沙中心,长沙 410006
    2.湖南大学 信息科学与工程学院,长沙 410006
  • 出版日期:2017-08-15 发布日期:2017-08-31

Particle clustering algorithm with adaptive K values

BAI Shuren1,2, CHEN Long2   

  1. 1.The National Supercomputing Center in Changsha, Hunan University, Changsha 410006, China
    2.College of Information Science and Engineering, Hunan University, Changsha 410006, China
  • Online:2017-08-15 Published:2017-08-31

摘要: 传统K-means算法除了对初始聚类中心的选择非常敏感,易收敛到局部最优解外,还存在着K值难以确定的问题,不合适的K值往往会得到较差的聚类结果。而K值问题也是聚类分析中的一个重要的研究方向,在粒子群聚类算法的基础上,结合K-means算法,提出了自适应K值的粒子群聚类算法。当算法收敛时,可通过比较不同K值时全局最优适应度值之间的关系来决定K值的增大与减小。实验表明改进的算法可以有效指导K值的选取,并且具有较好的聚类效果。

关键词: 粒子群聚类算法, K-means算法, 自适应K值, 收敛

Abstract: Traditional K-means algorithm is not only sensitive to the choice of initial clustering center and is easy to converge to the local optimal solution, but also has the problem of determining the value of K:inappropriate K values often lead to poor clustering results. The K value problem is an important research direction of clustering analysis, on the basis of particle clustering algorithm, combining the K-means algorithm, this paper proposes the particle swarm optimization clustering algorithm with adaptive K values. When the algorithm convergence, by comparing the relationship between different values of global optimal fitness under different K values, the increase or decrease of K values can be detemined. Experiments show that the improved algorithm can guide the selection of K values, and has a better clustering effect.

Key words: particle swarm optimization algorithm, K-means algorithm, self-adaptive K values, convergence