一种基于SMOTE和XGBoost的窃电检测方案

用户侧窃电行为造成的非技术损失不仅增加了电网的运营成本，还会干扰电力系统稳态。现有的检测方案忽略了用电数据的时序性及正负样本 分布不均、维度高的问题，这将极大地影响检测的准确率。因此，提出了一种基于SMOTE和XGBoost的窃电检测方案。针对电力数据的时序性和类不平 衡的特点，利用SMOTE算法进行过采样解决了数据不平衡的问题，并构造时序特征挖掘用户用电模式；再使用XGBoost执行用电数据的特征提取和分类 过程。实验表明，通过SMOTE算法可以提高不平衡数据分类的有效性，对比于传统的检测模型，XGBoost算法在窃电检测场景的多项评价指标下均取得 了更好的效果，其中准确率提升至92.45%。

Scheme for Electricity Theft Detection Based on SMOTE and XGBoost

The non-technical losses of user side electric power caused by electricity theft not only increase the operating costs of power grid， but also interfere with the steady state of the power system. Existing detection schemes greatly affect detection accuracy of by ignoring the problems including the time series of electricity data， the uneven distribution of positive and negative samples and high dimensions. Therefore， a scheme for electricity theft detection based on SMOTE and XGBoost is proposed. Targeting the characteristics of the time series and the class imbalance of power data， SMOTE algorithm oversampling is used to solve the problem of data imbalance， and time series features are constructed to mine power consumption patterns. In addition， XGBoost is used to perform the feature extraction and classification of the power data. The experiments show that the effectiveness of the imbalanced data classification can be improved by the SMOTE algorithm； Compared with the traditional detection model， the XGBoost algorithm is of better results in the detection with multiple evaluation indicators of the electricity theft detection scenarios， of which the accuracy rate has been increased to 92.45%.