DBSCAN-Based Electricity Consumption Anomaly Detection Method Integrated With VAE

Abstract

With the large-scale deployment of smart grid technologies in China and rapid progress in power system informatization, power utilities have accumulated vast amounts of operational data through automated management systems, which serves as a critical foundation for driving the digital transformation and intelligent modernization of the national grid infrastructure. During the operational process of electromechanical equipment, anomalies may arise due to various potential non-standard electricity consumption behaviors, equipment malfunctions, and other factors. Failing to preprocess the contaminated raw data prior to analysis can significantly compromise the accuracy of data analysis. Anomaly detection technology enables the timely detection and localization of abnormal data, while also revealing electricity consumption trends. This aids staff in proactively responding to special or unexpected situations, thereby maintaining the safe operation of equipment. This paper introduces a DBSCAN (Density Based Spatial Clustering of Applications with Noise) algorithm integrated with VAE (Variational Autoencoder) (VAE-DBSCAN), utilizing local electricity consumption data provided by Shandong DeYou Electric Company. By incorporating a channel attention model, the method can detect anomalies in the electricity consumption data, and the anomalies can then be cleaned up. The ADF testing method is used to quantify the disparities between VAE-DBSCAN and other algorithms, verifying the superiority of VAE-DBSCAN in anomaly detection.