«上一篇/Previous Article|本期目录/Table of Contents|下一篇/Next Article»

《计算机技术与发展》[ISSN:1006-6977/CN:61-1281/TN]

卷:

期数:

2025年07期

页码:

207-213

栏目:

新型计算应用系统

出版日期:

2025-07-10

文章信息/Info

Title:

Design and Implementation of Device Log Analysis System Based on Feature Extraction

文章编号:

1673-629X(2025)07-0207-07

作者:

杜佳*; 樊琳卓

中国电信股份有限公司陕西分公司 智能云网调度运营中心,陕西 西安 710075

Author(s):

DU Jia*; FAN Lin-zhuo

Intelligent Cloud Network Operation Center,China Telecom Corporation Limited Shaanxi Branch,Xi’an 710075,China

关键词:

云网设备; 特征提取; 日志分析; 先验知识; 无监督学习; 设备操作合规性检查

Keywords:

cloud network device;  feature extraction;  log analysis;  priori knowledge;  unsupervised learning;  compliance checking of device operation

分类号:

TP311;TP391

DOI:

10.20165/j.cnki.ISSN1673-629X.2025.0079

摘要:

为了及时发现云网设备运维过程中的不合规操作,保障云网设备安全稳定,设备日志分析技术备受关注,特征提取是该技术的基本方法,特征质量直接影响着最终结果。 针对电信场景,提出了一种将先验知识和无监督学习相融合的特征提取方法。 该方法首先以华为、中兴通讯、烽火通信、新华三等市场主流通信设备制造商的产品手册,中国电信云网设备变更操作管理办法,运维案例库为基础,人工筛选出符合目标要求的特征;然后为日志中的每个操作条目提取操作摘要,即主题词,以 TF-IDF 思想为基础,计算出每个操作摘要的权重值,从而实现无监督学习的特征提取;最后将先验知识和无监督学习两种方法提取的特征相融合,组成变更操作特征库,增强了特征在电信企业管理要求和未知场景预测两方面的代表性。 测试结果表明,该方法能对设备操作日志进行高效分析,实现合规性检查,用户自定义特征的日志信息提取,子日志划分等功能,且对机器硬件要求低,部署便捷、成本低廉,为电信运营商的设备运维安全检查提供支持。

Abstract:

To promptly detect non - compliant operations during cloud - network device operation and maintenance and ensure cloud -network security and stability,device log analysis technology has garnered significant attention,where feature extraction serves as a foun-dational methodology,with feature quality directly impacting final outcomes. A feature extraction method integrating prior knowledge and unsupervised learning is proposed for telecommunication scenarios. Firstly, by referencing product manuals from mainstream communication device manufacturers ( e. g. , Huawei, ZTE, FiberHome, H3C), China Telecom Cloud - Network Device Operation Management Regulations, and operation and maintenance case repositories, features aligning with target requirements are manually filtered. Subsequently,for each operational command in the logs,an operation abstract (keywords) is extracted. Based on the TF-IDF (Term Frequency - Inverse Document Frequency ) framework, weight values for each operation abstract are calculated, enabling unsupervised feature extraction. Finally,manually filtered features and unsupervised- learned features are combined to construct a change operation feature library,enhancing the representativeness of features in both telecom enterprise management requirements and unknown scenario prediction. The test results show that the proposed method can conduct efficient analysis of equipment operation logs,achieve functions such as compliance check,log information extraction of user-defined features, sub-log division,etc. Moreover, it has low re-quirements for machine hardware,is easy to deploy and has low cost,providing support for the security check of equipment operation and maintenance of telecommunications operators.

常用功能

更新日期/Last Update: 2025-07-10