具有不充分信息的高维时间序列因果关系网络研究  

Research on Causal Network of High-dimensional Time Series with Insufficient Information

作　　者：王双成 郑飞 赵大平 WANG Shuang-cheng;ZHENG Fei;ZHAO Da-ping(School of Information Management,Shanghai Lixin University of Accounting and Finance,Shanghai 201620,China;Institute of Data Science and Interdisciplinary Studies,Shanghai Lixin University of Accounting and Finance,Shanghai 201209,China;School of International Economics and Trade,Shanghai Lixin University of Accounting and Finance,Shanghai 201620,China)

机构地区：[1]上海立信会计金融学院信息管理学院,上海201620 [2]上海立信会计金融学院数据科学交叉研究院,上海201209 [3]上海立信会计金融学院国际经贸学院,上海201620

出　　处：《小型微型计算机系统》2023年第5期981-990,共10页Journal of Chinese Computer Systems

基　　金：国家社会科学基金项目(18BTJ020)资助。

摘　　要：具有不充分信息的高维时间序列因果关系网络学习重要且困难,信息不充分会导致许多因果关系丢失,从而造成传递信息的不完整.本文首先提出了汇聚递减变量排序方法,并基于局部贪婪搜索-打分进行因果关系网络学习,来降低对数据量的需求和提高学习效率与可靠性;再通过建立信息提取变量来获取变量组的压缩信息,以弥补由弱因果关系的缺失所引起的传递信息丢失和实现高维数据的降维;最后基于递归汇聚结构和后验分布抽样识别准确率分别建立时间序列变量之间的影响程度计算、影响的敏感性计算和汇聚与扩散影响计算方法,并使用宏观经济时间序列数据进行相应的实验验证与分析.It is important and difficult to learn causality network of high-dimensional time series with insufficient information.Insufficient information will lead to the loss of many causal relationships,resulting in the incomplete transmission of information.In this paper,firstly,we propose the aggregation decreasing variable ranking method,and conduct causal network learning based on local greedy search-scoring to reduce the demand for data number and improve the learning efficiency and reliability.Then,the compressed information of variable group is obtained by establishing information extraction variables to Make up for the loss of transmission information caused by the lack of weak causality and realize the dimensionality reduction of high-dimensional data.Finally,based on the recursive aggregation structure and the sampling identification accuracy of a posteriori distribution,the influence degree calculation,influence sensitivity calculation and aggregation and diffusion influence calculation methods between time series variables are established respectively,and the corresponding experimental verification and analysis are carried out with macroeconomic time series data.

关 键 词：时间序列 贝叶斯网络 影响程度 敏感性 汇聚与扩散 

分 类 号：TP181[自动化与计算机技术—控制理论与控制工程]