Electricity theft detection method based on multi‐domain feature fusion

To solve the problem of low accuracy of the previous electricity theft detection methods, the authors propose a multi‐domain feature (MDF) fusion electricity theft detection method based on improved tensor fusion (ITF). Firstly, the original electricity consumption series is transformed by gram angle field (GAF) to obtain the time‐domain matrix. The original electricity consumption series is converted into frequency‐domain by Maximal Overlap Discrete Wavelet Transform (MODWT) to obtain the frequency‐domain matrix. Then, the convolutional neural networks (CNN) are used to extract features of the time‐domain matrix and frequency‐domain matrix, respectively. Next, in order to fuse single‐domain feature information and MDF interaction information while reducing redundant information, the authors propose an ITF method to obtain a multi‐domain fusion tensor. Finally, the multi‐domain fusion tensor is input into the electricity theft inference module to judge whether the user implements electricity theft behaviour. The authors simulate six electricity theft types and evaluate the method's performance separately for each electricity theft type. The results show that the proposed method outperforms other methods. We propose a novel deep learning method for electricity theft detection based on multi‐domain feature fusion. This method can extract both time‐domain features and frequency‐domain features of electricity consumption series. We also propose a novel feature fusion method for multi‐domain feature fusion by introducing an attention mechanism to traditional tensor fusion.