面向非平稳时间序列的因果关系发现算法  

Causal Discovery Algorithm for Non-stationary Time Series

作　　者：周嘉颖 周跃进 ZHOU Jiaying;ZHOU Yuejin(School of Mathematics and Big Data,Anhui University of Science and Technology,Huainan 232001,China)

机构地区：[1]安徽理工大学数学与大数据学院,安徽淮南232001

出　　处：《长春师范大学学报》2024年第2期50-58,共9页Journal of Changchun Normal University

基　　金：深部煤矿采动响应与灾害防控国家重点实验室基金资助项目(SKLMRDPC22KF03)。

摘　　要：针对传统因果关系算法不能分析非平稳时间序列和可变时滞时间序列数据因果关系的问题,本文提出一种基于分段聚合近似可变时滞转移熵(PAAVL-TE)的因果关系算法。利用分段聚合近似法对时间序列进行转换,提取时间序列的特征信息,运用动态时间弯曲距离寻找相似程度最高的时间序列计算可变时滞时间序列的转移熵,实现了非平稳时间序列的因果分析。通过计算机仿真模拟实验将提出的算法与存在的算法相比较,证实算法有效性。将该算法用于北京市昌平区PM 2.5浓度和气象数据分析,表明本文算法具有广泛的应用性。To infer the causality of non-stationary time series accurately,a method of causal discovery algorithm based on piecewise aggregate variable-lag transfer entropy was proposed to overcome shortcomings of traditional methods of analyzing the causality of non-stationary and variable-lag time series.Firstly,we used piecewise aggregate approximation to transform the time series and extract features.Then,dynamic time warping was used to infer variable-lag transfer entropy,which implemented causal relations between non-stationary time series.The effectiveness of the proposed method is verified by the results of simulations and comparison with existing methods.We successfully applied this method to the concentration of PM 2.5 and meteorological time series in Changping,Beijing,which shows that the algorithm is widely applicable.

关 键 词：非平稳时间序列 分段聚合近似 转移熵 可变时滞 因果关系 

分 类 号：TP301[自动化与计算机技术—计算机系统结构]