基于DCNN和SVC的窃电检测  

Electricity Theft Detection Based on Deep Convolutional Neural Network and Support Vector Classification

作　　者：张梦楠 李红娇 ZHANG Meng-nan;Li Hong-jiao(College of Computer Science and Technology,Shanghai University of Electric Power,Shanghai 200090,China)

机构地区：[1]上海电力大学计算机科学与技术学院,上海200090

出　　处：《计算机仿真》2022年第6期92-97,429,共7页Computer Simulation

基　　金：国家自然科学基金(61403247,No.61702321);上海市信息安全综合管理技术研究重点实验室开放课题(.AGK2015005)。

摘　　要：针对目前窃电检测方法存在对大规模特征分类准确率较低的问题,提出基于DCNN和SVC的窃电检测方法。从电力数据的二维角度出发,对用户的电力数据按照周数进行矩阵化,利用改进的DCNN算法对二维矩阵进行自主学习,提取特征数据并降低分类器的输入特征维数,将DCNN提取的特征数据输入到SVC分类器中,识别窃电用户。采用国家电网公开数据集建立实验模型,进一步验证方法可行性,结果表明所提方法不仅能降低输入特征维度,而且提高了窃电检测的准确率。To solve the problem of low accuracy of large-scale feature classification in current electricity theft detection methods, a electricity theft detection method based on DCNN and SVC is proposed. From the two-dimensional perspective of power data, the user’s power data is matrixed according to the number of weeks. The improved DCNN algorithm is used to independently learn the two-dimensional matrix, extract the feature data and reduce the input feature dimension of the classifier. The feature data extracted by DCNN is input into the SVC classifier to identify the power stealing user. An experimental model was established using the public data set of the State Grid to further verify the feasibility of the method. The results show that the method not only reduces the input feature dimension, but also improves the accuracy of electricity theft detection.

关 键 词：窃电检测 深度卷积神经网络算法 支持向量分类机 

分 类 号：TP3[自动化与计算机技术—计算机科学与技术]