电信科学 ›› 2024, Vol. 40 ›› Issue (2): 72-82.doi: 10.11959/j.issn.1000-0801.2024027

• 研究与开发 • 上一篇    

基于深度学习的智能电网窃电检测混合模型研究

廖银玲1, 李金灿1, 王冰2, 张君2, 梁耀元2   

  1. 1 广西电网有限责任公司，广西 南宁 530023
    2 广西电网有限责任公司梧州供电局，广西 梧州 543002
  • 修回日期:2024-02-06 出版日期:2024-02-01 发布日期:2024-02-01
  • 作者简介:廖银玲（1986- ），女，现就职于广西电网有限责任公司，主要研究方向为安全管理、数字化应用等
    李金灿（1976- ），女，现就职于广西电网有限责任公司，主要研究方向为电力用户用电安全、用电设施隐患排查及风险管控等
    王冰（1972- ），女，广西电网有限责任公司梧州供电局高级工程师，主要研究方向为电力用户用电安全、用电设施隐患排查及风险管控等
    张君（1988- ），女，现就职于广西电网有限责任公司梧州供电局，主要研究方向为用电检查、抄核收、市场交易等
    梁耀元（1990- ），女，现就职于广西电网有限责任公司梧州供电局长洲供电分局，主要从事电力用户用电管理、电力用户安全隐患排查及整改、电力客户需求响应、电量抄核及电费回收等工作

A hybrid model for smart grid theft detection based on deep learning

Yinling LIAO1, Jincan LI1, Bing WANG2, Jun ZHANG2, Yaoyuan LIANG2   

  1. 1 Guangxi Power Grid Co., Ltd., Nanning 530023, China
    2 Wuzhou Power Supply Bureau of Guangxi Power Grid Co., Ltd., Wuzhou 543002, China
  • Revised:2024-02-06 Online:2024-02-01 Published:2024-02-01

在线阅读

15

PDF下载

42

摘要：

针对传统窃电检测模型受维度诅咒、类不平衡等问题，提出一种能有效检测智能电网窃电行为的混合深度学习模型，利用深度学习卷积神经网络（AlexNet）处理维度诅咒问题，显著提升数据处理的准确性；通过自适应增强（AdaBoost）对正常和异常用电行为分类，进一步提高分类精度；使用欠采样技术解决类不平衡问题，确保模型在各类数据的均衡性能；利用人工蜂群算法对AdaBoost和AlexNet的超参数进行优化，有效提高整体模型性能。使用真实智能电表数据集评估混合模型的有效性，与同类模型相比，提出的混合深度学习模型在准确率、精确度、召回率、F1 分数、马修斯相关系数（MCC）和曲线下面积-接收者操作特征曲线（AUC-ROC）分数上分别达到了 88%、86%、84%、85%、78%和 91%，不仅提高了用电行为监测的准确性，也为电力系统的智能分析提供了新视角。

关键词: 深度学习卷积神经网络, 自适应增强, 深度驱动模型, 窃电检测, 特征提取

Abstract:

A hybrid deep learning model was proposed to effectively detect electricity theft in smart grids.The hybrid model employed a deep learning convolutional neural network (AlexNet) to tackle the curse of dimensionality, significantly enhancing data processing accuracy and efficiency.It further improved classification accuracy by differentiating between normal and abnormal electricity usage using adaptive boosting (AdaBoost).To resolve the issue of class imbalance, undersampling techniques were utilized, ensuring balanced performance across various data classes.Additionally, the artificial bee colony algorithm was used to optimize hyperparameters for both AdaBoost and AlexNet, effectively boosting overall model performance.The effectiveness of this hybrid model was evaluated using real smart meter datasets from an electricity company.Compared to similar models, this hybrid model achieves accuracy, precision, recall, F1-score, Matthews correlation coefficient (MCC), and area under the curve-receiver operating characteristic curve (AUC-ROC) scores of 88%, 86%, 84%, 85%, 78%, and 91%, respectively.The proposed model not only increases the accuracy of electricity usage monitoring, but also offers a new perspective for intelligent analysis in power systems.

Key words: AlexNet, adaptive boosting, deep driven model, theft detection, feature extraction

中图分类号: 

No Suggested Reading articles found!