基于混合神经网络的配电网用户窃电检测方法  

A detection method for electricity theft by distribution network users based on a hy⁃ brid neural network

作　　者：成跃宇 成国锋 CHENG Yueyu;CHENG Guofeng(State Grid Yangzhou Power Supply Company,Yangzhou,Jiangsu 225009,China)

机构地区：[1]国网江苏省电力有限公司扬州供电分公司,江苏扬州225009

出　　处：《浙江电力》2023年第11期96-103,共8页Zhejiang Electric Power

基　　金：国网江苏省电力有限公司扬州供电分公司科技项目(63106022005)。

摘　　要：针对传统的基于一维用电量数据挖掘分析的用户窃电检测方法检测精度低的问题,提出了一种基于混合神经网络的配电网用户窃电检测方法。首先,为了增强正常用户与窃电用户用电量的特征差异性,采用MTF(马尔可夫变迁场)对一维用电量数据进行图变换,实现用电数据的二维化;同时,为提高模型的准确性及泛化性,引入了用户用电量档案数据。然后,采用混合神经网络分别对预处理后的二维用电图像、档案数据进行特征量提取及融合,以实现配电网用户窃电检测。最后,通过两组对比实验,验证所提方法的有效性和精确性。实验结果表明:与其他模型相比,基于混合神经网络在窃电识别的准确率、查全率及ROC(接受者操作特征)曲线下面积均有较大的提升,具有较好的识别性能。Given the low accuracy of the traditional electricity theft detection method based on one-dimensional elec⁃tricity consumption data mining and analysis,a detection method for electricity theft by distribution network users based on a hybrid neural network is proposed.Firstly,to enhance the characteristic difference between the electric⁃ity consumption of normal users and that of power theft users,the Markov transition field(MTF)is used to trans⁃form one-dimensional electricity consumption data into two-dimensional graphs.Moreover,to improve the accuracy and generalization of the model,profile data of users'electricity consumption is introduced.Then,the hybrid neural network is used to extract and fuse the feature quantities of the preprocessed two-dimensional electricity consump⁃tion graphs and profile data respectively to detect electricity theft by distribution network users.Finally,the effec⁃tiveness and accuracy of the proposed method are verified through two sets of comparison experiments.The experi⁃mental results show that the method based on a hybrid neural network is superior to other models in detection accu⁃racy of electricity theft,recall rate,and AUROC(area under the receiver operating characteristics),and has higher detection performance.

关 键 词：配电网 用户窃电检测 马尔可夫变迁场 混合神经网络 

分 类 号：TM73[电气工程—电力系统及自动化] TP183[自动化与计算机技术—控制理论与控制工程]