首页|一种基于互信息度量的时序数据因果发现方法

一种基于互信息度量的时序数据因果发现方法

扫码查看
在时序数据因果关系发现研究中,传统算法针对时间窗口内时序数据之间的因果关系进行分析,存在因果关系识别准确率受限、算法复杂度较高等问题.为解决该问题,首先对概要因果图、因果概要互信息和条件因果概要互信息进行定义,在此基础上推导出基于因果互信息的时序变量定向规则,而后区分是否存在混杂因子,结合 PC(Peter and Clark)和 FCI(fast causal inference)算法分别提出改进的 PCSMI(Peter and Clark summary mutual information)和 FCISMI(fast causal inference summary mutual information)算法.实验结果表明改进后算法能够在低复杂度条件下有效提升时序数据因果发现的准确率.
A causal discovery method for time series data based on mutual information measurement
In the research on causal discovery of time series data,the traditional algorithm analyse the causal relationship between time series data in the time window,which has problems such as limited causality recognition accuracy and high algorithm complexity.In order to solve this problem,this paper first defines the summary causal diagram,causal summary mutual information and conditional causal summary mutual information,derives the temporal series variable orientation rule based on causal mutual information,and then distinguishes whether there are confounding factors,and proposes improved Peter and Clark summary mutual information(PCSMI)and fast causal inference summary mutual information(FCISMI)algorithms combined with Peter and Clark(PC)and fast causal inference(FCI)algorithms,respectively.Experimental results show that the improved algorithm can effectively improve the accuracy of causal discovery of time series data under low complexity conditions.

causal discoverytime series datasummary causal diagramcausal summary mutual informationconditional causal summary mutual informationPCFCI

李德志、鲁云军、吴健平、李强

展开 >

国防科技大学信息通信学院,武汉 430019

因果发现 时序数据 概要因果图 因果概要互信息 条件因果概要互信息 PC FCI

"十四五"装备预研项目

315057206

2024

控制与决策
东北大学

控制与决策

CSTPCD北大核心
影响因子:1.227
ISSN:1001-0920
年,卷(期):2024.39(9)
  • 1
  • 3