An Improved CNN Approach for Electricity Theft Detection

To address the issue of incomplete feature extraction in the context of electricity theft detection, an improved convolutional neural network (CNN) model is proposed, which incorporates both the periodicity of user electricity consumption and the influence of traditional holidays. Additionally, a method is introduced to handle missing values in electricity consumption data by combining the electricity consumption periodicity and local temporal characteristics for imputation. Experimental results demonstrate that compared to existing electricity theft detection methods, the proposed approach achieves better reconstruction of actual electricity consumption data and significantly improves the accuracy of electricity theft detection.