Traditional methods for abnormal power usage detection of power users cannot effectively deal with complex power usage patterns and sudden abnormalities, resulting in low detection efficiency and poor accuracy. In this paper, a support vector machine (SVM)-based abnormal power usage detection method for grid users is proposed. First, a series of indicators for characterizing abnormal power consumption are constructed by extracting features such as user power consumption changes, power consumption differences, and line losses. Then, the improved K-medoids clustering algorithm is used to preprocess and cluster analyze the power consumption data to filter out the abnormal power consumption data. Finally, SVM was utilized for the classification and detection of abnormal electricity consumption. The experimental results show that after processing the data of 3305 electricity users, the proposed method achieves 99.6% in detection accuracy, which is significantly better than the traditional DT-SVM method and PSO-SVM method (87.6% and 92.5%, respectively). In addition, the proposed method also shows a large advantage in training time, which is only 18.21 seconds, compared with 53.62 seconds for DT-SVM and 45.26 seconds for PSO-SVM, which is a significant efficiency improvement. The experiment verifies the effectiveness and superiority of the method in abnormal power usage detection of grid users.
