基于TensorFlow深度学习框架的窃电分析研究  

Research on Electric Theft Analysis Based on Tensorflow Deep Learning Framework

作　　者：李莎[1,2] LI Sha(Changsha University of Technology.,Changsha 410000,China;Yiyang Power Supply Branch of State Grid Hunan Electric Power Co.,Ltd.,Yiyang 413000,China)

机构地区：[1]长沙理工大学,湖南长沙410000 [2]国网湖南省电力有限公司益阳供电分公司,湖南益阳413000

出　　处：《电气传动自动化》2021年第1期4-6,共3页Electric Drive Automation

摘　　要：目前,对于电力用户异常用电的问题为研究,提出基于深度学习用户异常用电模式检测相关模型,同时借助Tensorflow框架,进一步构建多层特征匹配网络以及特征提取网络。长短期记忆特征提取网络,能够从大量时间序列中获取不同序列特征,对于全连接网络多层特征匹配网络,研究学者提出,利用所提取特征数据,以完成异常用电数据的相关检测。通过实际研究发现,相比非深度学习检测模型来说,研究基于深度Tensorflow学习框架模型,能够有效完成异常用电模式的检测工作,除此之外,相比多层长短期记忆特征提取,目前来说,所提取的模型鲁棒性和准确性较好。In order to solve the problem of abnormal power consumption of clients,a relevant model of abnormal power consumption pattern detection based on deep learning is proposed.Meanwhile,the multi-layer feature matching network and feature extraction network are further built with the help of TensorFlow framework.Long and short-term memory feature extraction network can obtain different sequence features from a large number of time series.For full-connection network multi-layer feature matching network,researchers propose to complete correlation detection of abnormal power consumption data by using the extracted feature data.Through field research,it is found that compared with the non-deep learning detection model,the deep Tensorflow learning framework model can effectively implement the detection of abnormal power consumption patterns.In addition,compared with the feature extraction of multi-layer long and short-term memory,the present extracted model has better robustness and accuracy.

关 键 词：TensorFlow深度学习 框架 窃电分析 

分 类 号：TP181[自动化与计算机技术—控制理论与控制工程]