基于神经网络算法的用户窃电行为检测模型研究

Research on Electricity Theft Behavior Detection Model Based on Neural Network Algorithm

为了检测存在窃电行为的用电用户,减少电力企业经济效益损失,文章采用湖州市真实用户用电数据,基于机器学习BP神经网络算法,构建低压用户与专变用户用电行为特征,建立窃电风险等级模型。模型采用6折交叉验证方法,平均AUC值达到0.85,验证集中窃电用户命中率达到0.75,比以往人为排查的方法提高了效率和准确性,一定程度上能够辅助反窃电技术的智能化和信息化,为反窃电管理的完善提供有效的技术支持。

In order to detect the electricity theft behavior and reduce the economic loss of power enterprises, this paper presents a risk rating model of electricity theft based on BP neural network algorithm. The applied real data is collected from power users in Huzhou, Zhejiang province, the model average AUC reaches 0.85 with 6-fold cross validation, the hit ratio of validation set is 0.75. The result shows the improvement of efficiency and accuracy compared with the artificial investigation. The proposed method can provide a support for the intelligent electricity theft detection and enhance anti electricity theft management.