Federated Learning for Enhanced Malware Threat Detection to Secure Smart Power Grids

Cyber-Physical Systems (CPS) and the Internet of Everything (IoE) are vital in improving the efficiency of contemporary structures like Smart Grids by supporting the interconnection of people, processes, data, and things. Due to the convergence of devices and systems required by the IOE, smart grid has to be equipped with highly effective and reliable methods of attack detection. This research introduces a novel federated learning approach that utilizes the “CIC-MalMem-2022” dataset to enhance malware detection through a two-layered framework. The CPS layer focuses on local model training, while the Fog layer integrates these models into a global framework. The results demonstrate that the global model achieves superior performance with an accuracy of 99% and a precision of 99% on the training dataset, compared to the local model’s 85% accuracy and 75% precision on the test dataset. Additionally, the global model maintains 88% accuracy and 78% precision on the test set, underscoring the efficiency of the federated learning approach.