Res2-UNeXt Combined with Federated Learning for Cyber-Attack Detection and Classification in Multi Area Smart Grid Power System
J. Jasper, Beekanahalli Mokshanatha Praveen, S. Berlin Shaheema · 2024
A smart grid (SG) combines an information network, a communication network, and an electrical grid. With the fast improvement of SG technology, cyber-physical systems have become more complex, making SGs more susceptible to cyber-physical attacks. Protecting energy networks and critical components of communication from external attacks is crucial for maintaining reliable and efficient power distribution. Detecting Intrusions is vital to delivering safe services and notifying system administrators. This research suggests an intrusion classification scheme to detect cyberattacks on contemporary smart power grids that integrate multi-area power systems. It utilizes Hybrid Res2-UNeXt combined with a federated learning-based optimization algorithm to learn complex electrical grid properties. Deep learning with federated learning provides a robust system for detecting and classifying intrusions, enhancing the security of smart grids. The proposed method achieved 96.6 % accuracy when analyzing the original set of features and delivered a maximum accuracy of 99% with the selected data set from the publicly available dataset from Mississippi State University. Therefore, the suggested intrusion categorization method might successfully defend smart power grid systems from online threats.