A review on machine learning techniques for secured cyber-physical systems in smart grid networks

The smart grid (SG) is an advanced cyber-physical system (CPS) that integrates power grid infrastructure with information and communication technologies (ICT). This integration enables real-time monitoring, control, and optimization of electricity demand and supply. However, the increasing reliance on ICT infrastructures has made the SG-CPS more vulnerable to cyberattacks. Hence, securing the SG-CPS from these threats is crucial for its reliable operation. In recent literature, machine learning (ML) techniques and, more recently, deep learning (DL) techniques have been used by several studies to implement cybersecurity countermeasures against cyberattacks in SG-CPS. Nevertheless, the achieving high performance of these state-of-the-art techniques is constrained by certain challenges, including hyperparameter optimization, feature extraction and selection, lack of models' transparency, data privacy, and lack of real-time attack data. This paper reviews the advancement in using ML and DL techniques for cybersecurity countermeasures in SG-CPS. It analyzes the constraints that need to be addressed to improve performance and achieve real-time implementation. The various types of cyberattacks, cybersecurity requirements, and security standards and protocols are also discussed to establish a comprehensive understanding of the cybersecurity context in SG-CPS. This paper will serve as a comprehensive guide for new and experienced researchers.