湖南电力 ›› 2025, Vol. 45 ›› Issue (4): 126-132.doi: 10.3969/j.issn.1008-0198.2025.04.018

• 电力人工智能与数字化 • 上一篇    下一篇

基于规则-统计-Transformer三阶段融合的电力监控系统网络安全告警信息降噪方法

朱宏宇1,2, 陈乾3, 李明光4, 罗伟强3   

  1. 1.国网湖南省电力有限公司信息通信分公司,湖南 长沙 410004;
    2.泛在电力物联网湖南省重点实验室,湖南 长沙 410004;
    3.国网湖南省电力有限公司,湖南 长沙 410004;
    4.湖南大学信息科学与工程学院,湖南 长沙 410082
  • 收稿日期:2025-05-21 修回日期:2025-07-07 出版日期:2025-08-25 发布日期:2025-09-05
  • 通信作者: 朱宏宇(1991),女,硕士,研究方向为电力监控系统网络安全。
  • 基金资助:
    国网湖南省电力有限公司科技项目(5216A8220005)

An Alarm Denoising Method for Network Security in Power Monitoring System Based on Rule-Statistical-Transformer Three-Stage Fusion

ZHU Hongyu1,2, CHEN Qian3, LI Mingguang4, LUO Weiqiang3   

  1. 1. State Grid Hunan Electric Power Company Limited Information and Communication Company, Changsha 410004, China;
    2. Hunan Key Laboratory for Internet of Things in Electricity, Changsha 410004, China;
    3. State Grid Hunan Electric Power Company Limited, Changsha 410004, China;
    4. College of Computer Science and Electronic Engineering, Hunan University, Changsha 410082, China
  • Received:2025-05-21 Revised:2025-07-07 Online:2025-08-25 Published:2025-09-05

摘要: 为降低电力监控系统中的冗余告警,提高告警分析效率与准确性,提出一种基于规则-统计-Transformer三阶段融合的告警信息降噪方法,通过分层过滤机制实现高效精确的误告警识别。首先,采用规则匹配引擎快速筛除已知模式的误告警,完成数据初步清洗。其次,基于统计分析方法检测重复告警与周期性噪声,实现二次降噪。然后,利用Transformer模型对剩余告警进行深度语义分析,通过注意力机制捕捉复杂误报模式。最后,依托某网省公司的网安平台采集数据进行效果验证。实验结果表明,所提方法在确保检测精度的同时,可以显著提高检测效率,数据量较大情况下执行时间降低15%以上。

关键词: 告警降噪, Transformer, 三阶段融合, 规则匹配, 统计分析, 智能电网

Abstract: To reduce redundant alarms in the power monitoring system and improve the efficiency and accuracy of alarm analysis, an alarm denoising method based on rule-statistical-transformer three-stage fusion is developed in this study. Efficient and accurate false alarm identification is achieved through a layered filtering mechanism. First, false alarms with known patterns are rapidly screened using the rule matching engine, completing initial data cleaning. Second, duplicate alarms and periodic noise are detected through statistical analysis, accomplishing secondary noise reduction. Subsequently, in-depth semantic analysis of remaining alarms is performed by the transformer model, with complex false alarm patterns being captured via the attention mechanism. Finally, data are collected from a certain provincial company's network security platform for effect verification. The experimental results demonstrates that the proposed method significantly improves detection efficiency while maintaining accuracy. The execution time is reduced by over 15% for large-scale datasets.

Key words: alert denoising, Transformer, three-stage fusion, rule-based matching, statistical analysis, smart grid

中图分类号: