Enhancing Intrusion Detection Systems through Federated Learning and Gated Recurrent Units

Supervisory Control and Data Acquisition (SCADA) systems are essential for monitoring Critical Infrastructure (CI), such as Refineries, Pipelines, Power Grids and Mass Transport Systems. The interconnected nature of these systems and their associated networks are facing a wide range of cyber threats that pose a serious security threat. Recent research work has demonstrated Intrusion Detection Systems (IDS) are effective in mitigating cyber-attacks. Therefore, this article proposes a novel approach to address cyber-attack detection in a critical infrastructure based on Federated-Gated Recurrent Units (FGRU). The proposed algorithm locally trains the GRU model on local data. Federated learning is a distributed machine learning technique, where multiple IoT devices prepare an on-device model training on device local data. Then these IoT devices send the model weights to central server. The federated structure allows data aggregation through multiple communication rounds over each local ICS channel, allowing for data privacy and confidentiality. The server aggregates these updates to improve the global model. The model performance in the paper is validated on real-world gas pipeline ICS data, making it suitable to apply on industrial scenarios.