Detection Technology for Electricity Theft Behavior of Power Users Based on Integrated LSTM-AE

In order to effectively prevent and control users' electricity theft behavior, this paper uses the long short-term memory (LSTM) network in deep learning to construct an autoencoder (AE) to identify users with abnormal electricity consumption. Firstly, it is necessary to replace the fully connected layer of the traditional autoencoder with a long short-term memory (LSTM) neural network in order to endow the autoencoder (AE) with a temporal memory function. The LSTM-AE model is to be trained using a substantial dataset comprising normal electricity consumption data. Thereafter, the model is to be employed to analyse the data. to be detected as the input of the model to obtain a reconstructed sequence of user electricity consumption. The error between the input sequence and the reconstructed sequence is calculated, users who exceed the threshold will be identified as electricity theft users. Finally, the kernel density estimation (KDE) method is utilised in order to ascertain the threshold, with a view to enhancing recognition efficiency.. Simulation experiments have shown that this method effectively improves recognition accuracy compared to traditional methods.