基于样本优化选取的支持向量机窃电辨识方法

目前窃电行为普遍存在，如何提高用户用电系统的窃电辨识能力是电力公司一直关注的热点问题。随着智能电表在各地区的普及，数据挖掘等大数据分析技术在用电数据处理上的应用越来越受到重视。针对电力公司亟待解决的反窃电问题，在研究支持向量机原理和分析用电数据特性的基础上，将One-class SVM算法引入到疑似窃电判断当中，提出了一种将电量波动特征和One-class SVM结合的窃电辨识模型。首先提出改进的电量数据波动系数来表征电量波动，然后设计了基于One-class SVM窃电辨识方案。提出一种以电量波动系数作为指标选取训练样本的方法，训练得到相应分类模型，通过该模型分析用电数据从而辨别出是否存在窃电行为。算法验证结果表明该方法能提高窃电辨识的准确率和效率，具备一定的实用性。

electricity theft identification using support vector machine based on sample optimization selection

Nowadays, Energy theft is widespread, how to improve the electricity theft identification of the user''s power system has been concerned about the hot issues by the power company. With the popularity of smart meters in all regions, data mining and other large data analysis technology in the application of electricity data processing has been receiving increasing attention. Focused on the problem of anti-stealing electricity which power companies are concerned, and based on the study of the principle of support vector machine and the analysis of the characteristics of electricity data, the One-class SVM algorithm is introduced into the judgment of suspected energy theft, and an electricity theft identification model combining power fluctuation feature and One-class SVM is proposed. This paper first proposes an improved power data fluctuation coefficient to characterize the fluctuation of electricity, and then designs an electricity theft identification scheme based on One-class SVM. This method combines the fluctuation characteristics of electricity to select the load data samples, and constructs the detection model of energy theft based on the electricity data, then identifies whether there is electricity theft behavior. Experiments show that this method can improve the accuracy and efficiency of electricity theft identification and it has certain practicability.