文档介绍：: .
计算机工程 标函数，保证因果结构稀疏性，进而通过基于 EM 算法与爬山法的迭代优化算法，引入因
果先验，进一步提高了模型的可靠性。实验表明，该算法在由不同参数生成的模拟数据上均表现突出，且在两个通讯网络的
真实数据集中，F1 评分相比对比算法提升了 ％和 %，有更高的准确率。而通过引入根因标注和因果依赖性先验，
该算法的 F1 评分能进一步提升 %和 %，体现了引入先验的有效性。
关键词： 事件序列；格兰杰因果；霍克斯过程；贝叶斯信息准则；EM 算法；爬山法
开放科学(资源服务) 标志码(OSID):
A study of Hawkes process for Granger causality discovery among
faults
CAI Ruichu, WU Siyu, QIAO Jie
(School of Computer, Guangdong University of Technology, Guangzhou Guangdong 510006, China)
【Abstract】The causal discovery among faults has important value in the fields of automated operation and maintenance, medical
care, etc. Existing causality modeling methods still have the following two challenges on fault event sequences: 1) it is difficult to
introduce causal priors, and there are problems such as too dense algorithm results; 2) there are problems of poor causality reliability
on sparse and low time precision data. To this end, we model the causality of different fault types as Granger causality based on the
Hawkes process and propose a Hawkes process for Granger causality discovery of fault sequences. The algorithm extends the Hawkes
process to the discrete time domain to model event sequences with low time precision and ensures the sparsity of the causal structure
by constructing an objective function based on the Bayesian information criterion. Then an iterative optimization algorithm based on
EM algorithm and hill-climbing method i