Abstract

The increasing demands for supervisory power management systems transform traditional power management systems into smart grids. To achieve this management, information and communication technologies (ICT) are integrated into power grids via the use of smart grid technology. This integration empowers consumers and providers of electrical utilities, enhances the efficiency and reliability of the power system, guarantees management continually, and controls client demands. However, smart grids are distributed in vast and complex networks, including millions of interconnected computers, devices, and other components. This distribution exposes the extensive power networks to several security flaws and breaches. This work presents a deep learning algorithm-based anomaly detection framework for smart grids. Extreme gradient boosting (XGBoost) is coupled with long-short term memory (LSTM). Based on high accuracy values, the proposed design is shown to be able to detect and identify abnormal behaviors. Furthermore, extra evaluation for the suggested design using other performance criteria, including mean square error, F1-score, precision, recall, and root mean square error.

Key words: Smart Grids; Security; Anomaly detection; Machine Learning; LSTM; XGBoost; DNN