万伟　刘红旗　孙洪昌　张峰　王洋　孙伟卿

摘要：针对窃电、滥用电等用户异常用电行为给电力公司造成了巨额的经济损失的问题，通过数据驱动方法，利用区域内居民用户日负荷数据分别从横向与纵向两个层面，对用户用能行为进行定量的综合评分，进而识别用户异常用电行为。首先，建立K-Means和SVM分类模型，將单个居民日负荷数据与周边具有相似用电行为的居民进行比较，用于生成用户用电行为评价的横向评分。其次，利用LSTM模型建立用户负荷预测模型，实现与自身历史用电行为的对比，生成用户用电行为评价的纵向评分。最终，通过设定权重进行综合评分。当评分低于一定阈值时进行预警。算例部分利用30个用户4年数据对提出方法进行验证，横向评分结果准确率达到99.9%以上，纵向评分的拟合优度达到95%以上，验证了方法的准确性。

关键词：用电异常检测; 长短期记忆网络; K-Means; 支持向量机; 用电评分

DOI：10.15938/j.jhust.2022.04.008

中图分类号： TM71

文献标志码： A

文章编号： 1007-2683（2022）04-0053-10

Early Warning Method of Abnormal Electricity

Consumption Behavior Based on Data Driven

WAN Wei LIU Hongqi SUN Hong-chang ZHANG Feng WANG Yang SUN Wei-qing

（1.Heze Power Supply Company of State Grid Shandong Electric Power Company， Heze 274000， China;

2.Department of Electrical Engineering， University of Shanghai for Science and Technology， Shanghai 200093， China）

Abstract：In view of the huge economic losses caused to power companies by abnormal power consumption behaviors of users such as stealing and abusing electricity， based on data driven method， the daily load data of residents in the region is used to score the users′ energy consumption behavior quantitatively from the horizontal and vertical levels. Firstly， based on K-means and SVM （Support Vector Machine） classification model， the daily load data of individual residents data are compared with those of residents with similar electricity consumption behaviors to generate the user′s horizontal score. Secondly， the user load forecasting model is established by using LSTM （Long Short-Term Memory） model to realize the comparison with their own historical electricity consumption behavior and generate the users vertical score. Finally， according to the set weight， a comprehensive score is made. When the score is lower than a certain threshold， early warning is given. The proposed method is verified by the data of 30 users for 4 years， the accuracy of horizontal score is more than 99.9%， and the goodness of fit of vertical score is more than 95%， which proves the feasibility of the method.

Keywords：detection of abnormal power consumption; long short-term memory; K-Means; support vector machine; power rating

0引言

我国电力公司电量自动化管理技术相对匮乏，造成了窃电等异常用电行为多发，给电力公司带来了巨大的经济损失，同时也扰乱了市场经济秩序[1]。近年来，随着电力体制改革，电力企业逐渐认识到提高电量管理水平、加大反窃电打击力度、强化电网自动化运营能力，才能保证电网安全生产，获取更大的经济效益[2]。减少非技术损失对电力公司而言至关重要，非技术损失是指配电网侧电力用户窃电等一系列欺骗性用电行为所导致的电能损失[3]。……