一种基于小波概要的数据流量子聚类算法

摘要

针对传统的聚类分析技术面对长度无限且随时变化的海量级数据流无法直接使用或使用缺陷突出等问题,从数据流自身特性出发,结合小波变换与量子理论,提出一种新的数据流量子聚类算法.该算法首先采用离散小波变换,从每个数据流中动态分层地提取出其概要结构作为其相应的特征属性,同时计算出每个数据流到聚类中心的近似距离,结合量子理论估算出较优的核宽度调节参数进行类调整,最终获得一个较为理想的聚类效果.实验表明,该算法较好地解决了传统聚类方法无法良好解决的多数据流并行聚类问题,并表现出较好的聚类性能.%In view of the traditional clustering analysis technology, faced with the problem unlimited length and always changing mass data stream can't be directly used or the use of defects, a new data stream quantum clustering algorithm is proposed based on the characteristics of data stream, combined with the wavelet transform and quantum theory.Firstly, the discrete wavelet transform is used to extract the profile structure from each data stream as its corresponding characteristic property, and calculate the approximate distance of each data stream to the clustering center.The optimal kernel width adjustment parameters are estimated by quantum theory, and an ideal clustering effect is obtained finally.Experiments show that the proposed algorithm can solve the problem of multiple data stream parallel clustering which can't be solved by traditional clustering method, and show good clustering performance.