Article Details

Abstract

To address the issues of high alarm volume, complex correlations, and low efficiency in manual root cause localization within urban rail transit communication systems, this paper proposes a method for alarm correlation and root cause localization based on knowledge graphs and multi-source knowledge fusion. This method constructs an equipment-connection knowledge graph using a graph database, clusters the alarms to be analyzed according to temporal density, and maps them to graph nodes to form a set of preliminary connected subgraphs. It then integrates knowledge from historical case libraries, merges subgraphs with implicit relationships, and establishes a complete alarm correlation graph. Based on this correlation graph, the method further combines multi-source information such as case libraries, alarm categories (physical operation/status), equipment hierarchy, and temporal rules to determine the root cause alarms. By fully considering multi-dimensional correlations in space, time, and business, this approach improves the accuracy of alarm correlation and root cause localization, effectively enhancing operational efficiency and the level of intelligence in O&M, and thus shows considerable value for engineering applications.

Keywords

Knowledge Graph; multi-source knowledge fusion; alarm correlation; root cause localization; urban rail transit O&M

References

1. [1] Wu Yangyang, Chen Huainan A correlation analysis model for communication network alarms based on association rules [J]. Communication and Computer (Chinese and English versions), 2004, 1 (1): 57-63
2. [2] Xiao Hailin, Li Xingming Application of Analytic Hierarchy Process in Correlation Analysis of Communication Network Alarm [J]. Telecommunication Science, 2006 (11): 36-39
3. [3] Liu Wenya, Xu Yongneng Mining of subway fault association rules based on improved Apriori algorithm [J]. Journal of Weapon Equipment Engineering, 2021, 42 (12): 210-215
4. [4] Zhao Jing, Li Jun, Long Chun, etc Long term abnormal behavior detection based on frequent itemset mining [J]. Computer Application Research, 2020, 37 (S2): 221-223
5. [5] Li Tongyan, Li Xingming Research on Alarm Preprocessing Method Based on Double Constrained Sliding Time Window [J]. Computer Application Research, 2013, 30 (2): 582-584
6. [6] Lin Bin, Zeng Ying, Li Xingnan, etc Analysis Method of Power Communication Alarm Based on Neighbor Propagation Clustering [J]. Electronic Design Engineering, 2016, 24 (16): 142-145
7. [7] Yang Jian, Lan Mingchao Alarm clustering and correlation method based on business time window selection [J]. Optical Communication Research, 2019 (1): 33-36
8. [8] Yin Shan, Kuang Liwei Alarm clustering correlation analysis method based on business topology [J]. Computer and Digital Engineering, 2022, 50 (5): 1069-1072
9. [9] Qin Huayun, Wu Kan, Mao Heng Research on Alarm Root Cause Localization Method Based on Knowledge Graph and Causal Algorithm [J]. Network Security Technology and Application, 2023 (3): 41-43
10. [10] Chen Xiaoyong, Cai Yongjian, Gu Danpeng, etc Application of Graph Database in Engineering Data Center [J]. Computer Age, 2021 (9): 42-45
11. [11] Li Hang Statistical Learning Methods [M] Version 2 Beijing: Tsinghua University Press, 2019:15-17

About the authors

Huo Yanyu (1991-), female, Han ethnicity, from Shijiazhuang, Hebei, engineer, master's student, research interests include intelligent operation and maintenance of urban rail communications.