An Efficient Anomaly Detection Model of Power Consumption Behavior for Industrial Customers

The timely detection of anomalies in power consumption behavior is crucial for enhancing the efficiency of supply and management for the State Grid, ensuring the normal operation of enterprise production equipment, and enabling cost-saving measures for users. However, current methods face significant challenges in real-time processing and analysis of large-scale data, particularly in industrial contexts, and in accurately identifying genuine power consumption anomalies. This paper presents Sleuth, an efficient anomaly detection model tailored for industrial customers. Sleuth integrates the Matrix Profile technique with Long Short-Term Memory (LSTM) networks to address the complexities of temporal pattern recognition and future consumption value prediction. By leveraging the Matrix Profile for feature extraction and LSTM networks for precise forecasting, Sleuth enhances the detection of key features in industrial power consumption data. Our model demonstrates significant improvements in both the efficiency and accuracy of real-time anomaly detection. Empirical results on the ECG dataset show that Sleuth outperforms existing models in anomaly detection performance and achieves up to a 48% reduction in training time. Validation experiments on real-world power consumption datasets further confirm Sleuth's capability to swiftly and accurately identify anomalies.