半月刊

ISSN 1000-1026

CN 32-1180/TP

+高级检索 English
基于特征提取的面向边缘数据中心的窃电监测
作者:
作者单位:

1.电力传输与功率变换控制教育部重点实验室(上海交通大学),上海市 200240;2.北京交通大学电气工程学院,北京市 100044;3.国网湖北省电力有限公司电力科学研究院,湖北省武汉市 430077

作者简介:

通讯作者:

基金项目:

国家自然科学基金联合基金资助项目(U1866206)。


Electricity Theft Detection for Edge Data Center based on Feature Extraction
Author:
Affiliation:

1.Key Laboratory of Control of Power Transmission and Conversion (Shanghai Jiao Tong University), Ministry of Education, Shanghai 200240, China;2.School of Electrical Engineering, Beijing Jiaotong University, Beijing 100044, China;3.Electric Power Research Institute of State Grid Hubei Electric Corporation, Wuhan 430077, China

Fund Project:

This work is supported by Joint Funds of National Natural Science Foundation of China (No. U1866206).

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

    随着电网信息物理系统的发展,一部分数据处理功能逐渐下沉到靠近终端用户的边缘层。为了给后续分析提供可靠的数据源,及时发现异常用电行为,窃电监测是边缘数据中心重要功能之一。文中提出一种针对边缘数据中心的窃电监测方法,该方法利用深度卷积生成对抗网络(DCGAN)鉴别器提取得到的特征,在边缘数据中心对二范数线性支持向量机(L2SVM)进行训练。实验结果证实,DCGAN具有较好的收敛性能,鉴别器提取得到的正常与窃电行为用电特征具有明显划分,且比基于主成分分析(PCA)特征提取方法更加有效,此外,与基于径向基核函数的支持向量机(SVM)反窃电方法相比,所提方法准确度更好、且计算复杂度低,适合边缘数据中心部署。

    Abstract:

    With the development of grid cyber-physical system, part of data processing task gradually sinks to the edge side near end users. To provide dependable data source and detect anomaly consumption behavior, electricity theft detection is one of important functions of edge data center. In our electricity theft detection framework, the features are extracted by the discriminator of deep convolutional generative adversarial networks (DCGAN) and used to train L2-regularized linear support vector machine (L2SVM) in edge data center. The results show that DCGAN has good convergence performance. The features of normal and abnormal data can be easily separate and the proposed feature extraction method is more efficient than the method based on principal component analysis (PCA). Compared with support vector machine (SVM), the proposed method can achieve higher accuracy and have lower computation complexity and therefore is suitable for an edge data center.

    表 2 Table 2
    图1 基于边缘数据中心的窃电监测架构Fig.1 Framework of electricity theft detection base on edge data center
    图2 DCGAN架构Fig.2 Architecture of DCGAN
    图 DCGAN动态训练过程Fig. Dynamic training processes of DCGAN
    图 生成数据和真实数据经验累积概率分布比较Fig. Comparison of empirical cumulative probability distribution between generated data and real data
    图 基于T-SNE算法的提取特征可视化Fig. Visualization of feature extraction based on T-SNE algorithm
    图 不同监测方法混淆矩阵热图比较Fig. Comparison of different monitoring methods for confusion matrix heat map
    表 1 Table 1
    参考文献
    相似文献
    引证文献
引用本文

张宇帆,艾芊,李昭昱,等.基于特征提取的面向边缘数据中心的窃电监测[J/OL].电力系统自动化,http://doi.org/10.7500/AEPS20190624003.
ZHANG Yufan,AI Qian,LI Zhaoyu,et al.Electricity Theft Detection for Edge Data Center based on Feature Extraction[J/OL].Automation of Electric Power Systems,http://doi.org/10.7500/AEPS20190624003.

复制
分享
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:2019-06-24
  • 最后修改日期:2020-03-17
  • 录用日期:2019-10-18
  • 在线发布日期:
  • 出版日期: