智能电网异常用电检测框架  

Smart Grid Abnormal Power Detection Framework

作　　者：刘晓华 高俊成 王少林 张海荣 LIU Xiao-hua;GAO Jun-cheng;WANG Shao-lin;ZHANG Hai-rong(Marketing Department of Xinjiang Electric Power Co.,Ltd.,Urumqi 830000 China;Audit Department(Audit Center),Xinjiang Electric Power Co.,Ltd.,Urumqi 830000 China;Xinjiang Information Industry Co.,Ltd.,Urumqi 830000 China)

机构地区：[1]新疆电力有限公司营销部,新疆乌鲁木齐83OOOO [2]新疆电力有限公司审计部(审计中心),新疆乌鲁木齐830000 [3]新疆信息产业责任有限公司,新疆乌鲁木齐830000

出　　处：《自动化技术与应用》2020年第12期124-128,160,共6页Techniques of Automation and Applications

摘　　要：智能电表的大量使用为电网中的异常行为检测提供了新的手段。本文基于智能电表采集的细颗粒的用电数据提出了面向智能电网的异常用电检测框架。该框架引入了用于客户分割的有限混合模型聚类和用于识别适合于预测的新特征的遗传编程算法的组合,并采用了梯度提升算法作为分类算法。利用来自4000多户家庭的用电数据集的试验表明,该检测框架具有明显优于其他的机器学习算法的性能。The large use of smart meters provides a new means of detecting abnormal behavior in the power grid.Based on the excellent granularity of electricity collected by smart meters,this paper proposes an abnormal power detection framework for smart grids.The framework introduces a combination of finite hybrid model clustering for customer segmentation and a genetic programming algorithm for identifying new features suitable for prediction,and uses a gradient lifting algorithm as the classification algorithm.Experiments using power data sets from more than 4000 households show that the detection framework has significantly better performance than other machine learning algorithms.

关 键 词：智能电网分析 大数据 异常检测 

分 类 号：TP399[自动化与计算机技术—计算机应用技术]