A Robust Deep Learning Framework for Anomaly Detection in Smart Grid

Identification of irregularities is a major and common problem that costs utility companies a lot of money and energy worldwide. The attack surface has definitely increased with the deployment of smart meters and other modern metering infrastructures. To combat such attacks, this study presents a robust deep learning framework designed for the identification of both cyber and physical anomalies within the power grid. Utilizing a Long Short-Term Memory (LSTM) model, the frame-work aims to discern underlying patterns in energy consumption. To assess its effectiveness, a comparative analysis was conducted, evaluating the performance of the proposed LSTM model against established machine learning models such as Random Forest (RF), K-nearest neighbor (Knn), XGBoost, Decision Tree (DT), and Naive Bayes (NB). To address data imbalance, six distinct synthetic attacks were generated. Additionally, 5-fold cross-validation was employed to enhance the reliability and generalization of the model evaluation process.