CNN-DBSCAN Intermittent Electricity Theft Detection Method for High Loss Station Areas Based on Correlation Analysis

The randomness of electricity theft behavior among intermittent electricity theft users results in data characteristics that differ from those of long-term electricity theft users. Consequently, directly applying the electricity theft detection method proposed for long-term electricity theft users to identify intermittent electricity theft users will result in a certain degree of error. The majority of existing methodologies for the detection of electricity theft under the assumption that the users under analysis and subsequent detection have engaged in electricity theft throughout the detection cycle. This approach, however, fails to account for the existence of intermittent electricity theft users within the actual system. In this paper, we put forward a correlation analysis-based CNN-DBSCAN method for the detection of intermittent electricity theft in high-loss stations with abnormal line loss rates. The method employs a multi-scale CNN model to identify the most probable time periods for all users under investigation for electricity theft. These time periods are then compared with the actual time periods when the line loss rate data exhibits abnormal behaviour. The Pearson maximum correlation analysis results are obtained under the time-distance weighting, and all the correlation analysis results for each user are clustered through the DBSCAN clustering algorithm. This process allows for the identification of outlier points, which are then classified as instances of electricity theft. The proposed method is evaluated using artificial electricity theft samples constructed based on real electricity consumption data, with a view to verifying its effectiveness in detecting intermittent electricity theft users. The superiority of the proposed method is demonstrated through comparison experiments with other methods.