A Novel Hybrid Deep Learning-Based Framework for Intelligent Anomaly Detection in Smart Meters

Smart meters deployment in residential buildings generates a large volume of time-series data that offers valuable insights about electricity consumption. The proposed work exploits smart meter data to identify abnormal behavior in electricity consumption and sensor malfunctioning. In this paper, a two-stage anomaly detection approach that integrates the capabilities of deep learning and ensemble learning is designed. This approach combines Bidirectional Long Short-Term Memory (BiLSTM) networks for feature extraction and a random forest model to detect anomalies. BiLSTM is applied to determine complex temporal features, and these extracted high-level features are taken by a random forest model for anomaly detection. The performance of the proposed work is verified using two real-world smart meter electricity datasets of a diverse collection of buildings. The performance metrics determine that the proposed BiLSTM-random forest approach provides superior results across key evaluation metrics, making it a practical solution for intelligent energy monitoring systems.