A deep learning framework for intrusion detection system in smart grids using graph convolutional network

Abstract Smart grids are an emergent concept that has evolved from combining renewable energy source integration, digital communication advances, and increasing demands for dependable electricity supply. However, although these modernized grids make the process very efficient and reliable, at the same time, the increased vulnerability to cyberattacks makes the need for Intrusion Detection Systems robust, as these infrastructures are critical ones. Traditional intrusion detection systems frequently tend to be inadequate in managing the intricacy and continually evolving nature of threats associated with smart grids. This paper presents a streamlined deep-learning approach for developing an IDS. The process involves the following: a) Data Collection: DNP3 is widely used in most CIs across the United States for industrial communication protocol. It permits remote access between ICS and SCADA systems. b) Preprocessing using Encoding and Normalisation techniques to avoid anomalies and noises from the raw dataset, c) Feature Extraction for extracting features of the smart grid using Autoencoder, and d) Detection using Graphical Convolutional Network (GCN). GCN-based IDS demonstrates superior potential in identifying sophisticated cyber threats, ensuring real-time detection, and maintaining the integrity and reliability of smart grids. The experimental analysis’s findings show that the recommended approach outperforms current cutting-edge models, displaying exceptional sensitivity (0.98), specificity (0.98), accuracy (0.95), and F1 score (91).