基于改进自编码器和随机森林的窃电检测方法

Detection method for electricity theft based on improved autoencoder and random forest

作为智能电网的关键技术之一,高级计量架构凭借实时双向通信、按需应答等优点为电网提供重要的数据来源。面对当前日趋严重的窃电问题,有必要利用高级计量架构的数据发现非法窃电行为。因此,该文提出一种基于改进自编码器和随机森林的窃电嫌疑用户检测方法。通过改进自编码器提取隐含在电力用户用电量信息中的特征,应用批标准化算法优化训练过程,并采用这些特征来构建随机森林模型判断窃电嫌疑用户。运用真实数据集,通过仿真实验并对比现有的BP神经网络、极限学习机等模型验证所提出方法的有效性和准确性。

As one of the key technologies of smart grid,advanced metering infrastructure provides an important data source for the grid by the advantage of real-time two-way communication and on-demand response.As the increasingly serious of power theft problems,it is necessary to utilize the data of advanced metering infrastructure to find the illegal consumers.Therefore,this paper proposes a method for detecting suspected power theft users based on an improved autoencoder and random forest.By improving the selfencoder to extract the characteristics implicit in the power consumption information of power users,a batch of standardized algorithms is used to optimize the training process,and these characteristics are used to construct a random forest model to determine suspected power theft users.The real data set is used to verify the effectiveness and accuracy of the proposed method through simulation experiments and comparison with existing BP neural network,extreme learning machine and other models.