基于小时尺度周期特征自编码器的用户窃电识别方法

User Electric Theft Detection Method Based on Hour Scale Periodic Feature LSTM-transformer

近年来,基于线损归因分析的方法在高损线路窃电检测中得到大量成功应用。非高损线路中同样可能存在窃电用户,但窃电电量较小容易被正常的线损波动所掩盖,会削弱线损归因分析类方法的适用性,亟需研究基于用户自身计量数据而不依赖于接入线路线损数据的窃电检测方法。利用电力专变用户用电具有小时尺度周期性的特点,提出一种面向非高损线路的自编码器的用户窃电检测方法。首先利用正常与窃电用户在小时尺度用电的周期性差异,使用自编码器对用户用电行为进行深度学习,拟合出基于用户实际用电数据的重构值。然后,采用改进的高斯分布对用户用电数据与重构值的均方误差进行自适应阈值设置,以消除专变用户日际用电波动性的影响,即可根据均方误差是否超过自适应阈值来进行窃电识别。最后,根据某地实际电网用户数据进行算例仿真,验证了所提方法的有效性。

In recent years,the method based on the line loss attribution analysis has been successfully applied in the detection of high-loss line stealing.The stealing users may also exist in the non-high-loss lines,but the amount of electricity stolen is too small to be discovered by the normal line loss fluctuations,which will weaken the applicability of the line loss attribution analysis.Therefore,it is urgent to study the new method for stealing detection based on the user's own measurement data rather than on the line loss data.A autoencoder stealing detection for the non-high-loss lines is proposed by using the characteristics of hour-scale periodicity of power consumption of the power transformer users.Firstly,using the periodic difference of the electricity consumptions between the normal and the stealing users at the hourly scale,the autoencoder is used to conduct deep learning on the users'electricity consumption behavior,and the reconstruction value based on the users'actual electricity consumption data is fitted.Then,the improved Gaussian distribution is used to set the adaptive threshold for the mean square error between the user's power consumption data and the reconstructed value so as to eliminate the influence of the daily power consumption fluctuation of the special transformer users,i.e.the power stealing recognition can be carried out according to whether the mean square error exceeds the adaptive threshold.Finally,the effectiveness of the proposed method is verified by a numerical example based on the actual power grid user data in a certain place.