Research on Electricity Anomaly Detection Based on Clustering and Multi-Classification Modeling

To enhance abnormal power consumption detection and enable fast, accurate identification of irregularities, this paper proposes a power theft detection method based on clustering and multi-classification models. First, customer power usage patterns are analyzed using K-means clustering, grouping customers with similar behaviors. Within each group, four classifiers—RF, BP, SVM, and KNN—are trained and evaluated to select the best model for each group. Testing on a residential data set from a Shandong power company, results indicate that this clustering-based, multi-classification approach improves detection flexibility and adaptability over single-method models.