一种基于自适应多通道注意力和知识蒸馏的电力通信业务识别方法
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:

基金项目:


A Power Communication Service Recognition Method Based on Adaptive Multi-Channel Attention and Knowledge Distillation
Author:
Affiliation:

Econormic 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:

    随着电力通信网络的迅速发展,高效准确的业务识别对于保障电力系统的稳定运行至关重要。因此,本文提出了一种基于自适应多通道注意力和知识蒸馏的电力通信业务识别方法,旨在提高识别精度并优化计算资源的使用。首先,提出了一种自适应多通道注意力网络(Adaptive Multi-Channel Attention Network,AMCAN),该网络融合了自适应多通道机制、注意力机制与交互注意力机制,通过多个并行特征通道有效捕捉输入数据中的多样化特征信息,并引入自适应权重机制,实现对不同通道贡献的动态加权调整。其次,设计了一种自适应知识传递蒸馏方法,该方法通过两个阶段逐步传递教师模型的知识,并随着迭代次数动态调整传递过程,使轻量级学生模型在显著降低计算资源消耗的同时,依然能够保持较高的识别精度。最后,在Moore数据集上的实验表明,所提出的方法在精确率、准确率、召回率和F1分数方面均优于对比方法,具有广泛的实际应用前景。

    Abstract:

    With the rapid development of power communication networks, efficient and accurate service recognition is crucial for ensuring the stable operation of power systems. Therefore, this paper proposes a power communication service recognition method based on adaptive multi-channel attention and knowledge distillation, aimed at improving recognition accuracy and optimizing the use of computational resources. First, an Adaptive Multi-Channel Attention Network (AMCAN) is introduced, which integrates adaptive multi-channel mechanisms, attention mechanisms, and interactive attention mechanisms. This network effectively captures diverse feature information in the input data through multiple parallel feature channels and introduces an adaptive weight mechanism to dynamically adjust the contributions of different channels. Secondly, an adaptive knowledge transfer distillation method is designed, which gradually transfers knowledge from the teacher model in two stages, dynamically adjusting the transfer process with each iteration. This allows the lightweight student model to maintain high recognition accuracy while significantly reducing computational resource consumption. Finally, experiments on the Moore dataset demonstrate that the proposed method outperforms comparison methods in precision, accuracy, recall, and F1 score, indicating its broad practical application prospects.

    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:2025-02-26
  • 最后修改日期:2025-02-26
  • 录用日期:2026-07-08
  • 在线发布日期:
  • 出版日期:
文章二维码