Enhancement of electricity theft detection model for electricity metering system using machine learning and Pearson’s correlation coefficient

Abstract Power metering system electricity theft may jeopardize the security and stability of the power system in addition to reducing the financial gains of the power grid. This research suggests a power theft detection model based on the Pearson correlation coefficient and support vector machine (SVM) to increase the detection accuracy. The relationship between the user’s daily electricity consumption and the line loss of electricity of the day in the station region is first examined using the Pearson correlation coefficient to determine whether a user is committing power theft. The SMOTE technique is used to oversample a few classes of samples to address the issue of the limited amount of data on electricity theft users. This prevents sample imbalance from having an impact on the detection model. The SVM algorithm is then used to build the electricity theft detection model, and the Gaussian kernel function is used to translate the linearly indivisible power consumption data to the high-dimensional space to guarantee the dataset’s linear divisibility. With a classification accuracy of 93.8 % and a positive sample classification accuracy of 95.4 % for the detection of electricity theft users, the experimental portion uses the State Grid of China’s actual electricity consumption dataset. Following model training and cross-validation, the results demonstrate that the model can successfully identify electricity theft, demonstrating the model’s superiority and applicability in this area.