Analysis of machine-learning algorithms for detection of cyber attacks in smart grid

Abstract

Smart grids are playing an essential role to digitalize power networks by enabling automation and data-driven methods. However, growing reliance on automation and data-driven decision-making makes these systems encounter various cyber threats, especially false data injection attacks (FDIA). These attacks can manipulate grid measurement data, leading to erroneous generation or distribution decisions and potentially causing large-scale outages. This study researches the application of machine learning algorithms in power networks, with a particular focus on FDIA. It evaluates various detection technologies, which include supervised and unsupervised machine learning models, which include their accuracy, speed, and scalability. This study also examines existing mitigation strategies and their integration with machine learning models to enable real-time response and recovery. The results highlight the strengths and limitations of each detection approach, underscoring the importance of combining machine learning with strong cybersecurity measures to protect smart grids. Finally, this study discusses research opportunities, which are improving detection systems in the future and improving the smart grid's ability to respond to evolving cyber threats.