基于联邦学习与时空知识迁移的配电通信网络分布式云边协同异常
DOI:
CSTR:
作者:
作者单位:

国网福建省电力有限公司经济技术研究院

作者简介:

通讯作者:

中图分类号:

TN929.5

基金项目:

2025年国网福建省电力有限公司科技项目资助(SGFJJY00XGJS2500095)


Distributed Cloud-Edge Collaborative Anomaly Detection Method for Power Distribution Communication Networks Based on Federated Learning and Spatiotemporal Knowledge Transfer
Author:
Affiliation:

Economic and Technological Research Institute,State Grid Fujian Electric Power Co Ltd,Fuzhou

Fund Project:

  • 摘要
  • |
  • 图/表
  • |
  • 访问统计
  • |
  • 参考文献
  • |
  • 相似文献
  • |
  • 引证文献
  • |
  • 资源附件
  • |
  • 文章评论
    摘要:

    配电通信网络作为智能电网的核心支撑,持续产生高频、多源的时间序列数据,传统集中式异常检测模式面临隐私安全风险、系统效能不足与跨域数据协同困难的挑战。为此,本文提出一种基于联邦学习与时空知识迁移的配电通信网络分布式云边协同异常检测方法:首先,提出一种集成自适应频谱分析、图注意力网络变量关系挖掘与多尺度时序卷积局部特征提取的协同检测模型,有效增强对时序数据频域特征、变量间动态依赖与局部异常模式的感知能力;其次,设计基于差分隐私扩散模型的时序数据合成方法,在保护原始数据隐私的前提下生成高质量共享数据,为联邦训练提供数据基础;最后,构建双维度知识迁移机制,结合空间知识对齐与时间知识传递,缓解数据异质性带来的模型偏差与“灾难性遗忘”问题,并依托云边协同架构实现跨区域分布式异常检测。实验表明,该方法在保障数据隐私的同时,有效提升了异常检测的时效性与泛化能力,为智能电网配电通信网络的安全高效运维提供技术支撑。

    Abstract:

    As the core support of smart grids, power distribution communication networks continuously generate high-frequency and multi-source time-series data. Traditional centralized anomaly detection modes face challenges such as privacy and security risks, insufficient system efficiency, and difficulties in cross-domain data collaboration. To address these issues, this paper proposes a distributed cloud-edge collaborative anomaly detection method for power distribution communication networks based on federated learning and spatiotemporal knowledge transfer: Firstly, a collaborative detection model integrating adaptive spectrum analysis, graph attention network-based variable relationship mining, and multi-scale temporal convolution for local feature extraction is proposed. This model effectively enhances the ability to perceive frequency-domain features of time-series data, dynamic dependencies between variables, and local anomaly patterns. Secondly, a time-series data synthesis method based on the differential privacy diffusion model is designed. It generates high-quality shared data while protecting the privacy of original data, providing a data foundation for federated training. Finally, a two-dimensional knowledge transfer mechanism is constructed, combining spatial knowledge alignment and temporal knowledge transfer. This mechanism alleviates model bias and the "catastrophic forgetting" problem caused by data heterogeneity, and realizes cross-regional distributed anomaly detection relying on the cloud-edge collaborative architecture. Experiments show that the proposed method not only ensures data privacy but also effectively improves the timeliness and generalization ability of anomaly detection, providing technical support for the safe and efficient operation and maintenance of power distribution communication networks in smart grids.

    参考文献
    相似文献
    引证文献
引用本文
分享
相关视频

文章指标
  • 点击次数:
  • 下载次数:
  • HTML阅读次数:
  • 引用次数:
历史
  • 收稿日期:2026-01-26
  • 最后修改日期:2026-01-26
  • 录用日期:2026-01-28
  • 在线发布日期:
  • 出版日期:
文章二维码