面向数据流的多粒度时变分形维数计算

Multi-Granularity and Time-Varying Fractal Dimension on Data Stream

在大数据时代,数据流是一种常见的数据模型,具有有序、海量、时变等特点.分形是许多复杂系统的重要特征,分形维数是度量系统分形特征的重要指标量.数据流作为动态的复杂系统,其上的分形维数应具有动态、时变、多粒度等特性.提出了多粒度时变分形维数的概念,并设计了基于小波变换技术的数据流多粒度时变分形维数算法.该算法通过对数据流进行离散小波变换,并利用多粒度小波变换树结构在内存中保存数据流的概要信息,可以同时在不同的时间粒度上实时地计算数据流时变分形维数.该方法具有较低的计算复杂度,实验结果表明:该方法可以有效地监控数据流分形维数在不同粒度上的时变特征,深刻地揭示数据流的演化规律.

In the era of big data, data stream is a common data model with characteristics such as orderly, massive and time-varying. Fractal is an important feature of many complex systems, and is mainly represented by fractal dimension. Data stream can be viewed as a dynamic and complex system, and its fractal dimension should also have characteristics of dynamic, time-varying and multi-granularity. This paper presents a method of measuring multi-granularity and time-varying fractal dimension on a data stream based on discrete wavelet transform. The method can simultaneously measure the time-varying fractal dimension on a data stream by using the summary information from wavelet transforming of the data stream saved in a multi-granularity wavelet transforming tree in memory. This method has low computational complexity, and effectively reveals the evolution of a data stream. Experimental results show that it can effectively monitor the time-varying characteristic of fractal dimension on a data stream at different granularity.