计算机工程与应用 ›› 2019, Vol. 55 ›› Issue (8): 102-109.DOI: 10.3778/j.issn.1002-8331.1809-0330

• 模式识别与人工智能 • 上一篇    下一篇

具有动态特性的聚类弹性网络算法研究

雍巧玲,衣俊艳   

  1. 北京建筑大学 电气与信息工程学院,北京 100044
  • 出版日期:2019-04-15 发布日期:2019-04-15

Elastic Net Algorithm with Dynamic Characteristics for Clustering

YONG Qiaoling, YI Junyan   

  1. College of Electrical and Information Engineering, Beijing University of Civil Engineering and Architecture, Beijing 100044, China
  • Online:2019-04-15 Published:2019-04-15

摘要: 聚类分析是一种非常重要的聚类工具,被广泛应用在各科学领域的聚类问题中。其中,弹性网络是一种较好的聚类分析算法,尤其在高维空间有很大优势。提出了一种新的聚类弹性网络算法CENA(Clustering Elastic Net Algorithm)。该算法将一个面向聚类的描述数据点与弹性节点关系的能量函数用于ENA(Elastic Net Algorithm)求解模式中,结合极大熵原理,计算得到自由能函数。当自由能函数达到全局极小时,即可获得弹性网络的聚类解。通过大量实验证明,提出的CENA算法运行结果稳定,可以有效提升算法空间搜索能力,节省运行时间开销,规避参数调节问题。该算法相较于经典划分聚类算法,大大提高了聚类质量。

关键词: 聚类分析, 弹性网络, 极大熵, 自由能

Abstract: Cluster analysis is a very important clustering tool and is widely used in various scientific projects. And elastic net is a better clustering analysis method, especially in high-dimensional space. And a new Clustering Elastic Net Algorithm(CENA) is proposed in this paper. The algorithm has a new free energy function, and applies the maximum entropy principle. When the free energy reaches the global minimum, the clustering solution is obtained by the net. In this paper, a large number of random data points are tested. Experimental results show that the clustering results of CENA are stable. Compared with classical partitioning algorithms, CENA greatly improves the clustering quality. In general, the SED values of CENA are 20% lower than other classical partitioning algorithms.

Key words: cluster analysis, elastic net, maximum entropy, free energy