基于符号化的时间序列复杂网络构造及其拓扑结构研究

Research on building complex network based on symbolization of time series and its topological properties

复杂网络理论是时间序列分析中一种有力的工具,但在面对高频数据时,现有建网方法是低效的。因此,提出利用时间序列符号化技术压缩原始序列,并构造网络的方法。该方法使用最小二乘估计时序分段斜率,提取序列的局部特征,并构造字典判断节点是否邻接。模拟实验表明,所建网络的标度指数、集群系数与过程的Hurst指数高度相关,可以精确地捕捉原过程的复杂性特征。

Complex network theories may be a powerful tool in time series analysis. But when facing high frequency data,current method of building network is highly inefficient. Thus,this paper proposed a symbolic method for time series compression and built network. This method used least square to estimate the slope of each segments and then extracted local features from time series. Then it determined node＇s adjacency through a dictionary. Simulation studies show the scaling exponent,cluster coefficient of the network are highly correlated with Hurst exponent and can exactly capture complex properties of original series.