Abstract
As the realm of smart grids continues to evolve, embracing new technologies, researchers are exploring the potential ofupcoming 6G technology to address the challenges in management of smart grids. With the adoption of 6G technology,wireless energy meters, which play a key role in smart grid advancement, promise higher data rates, ultra-low latency, improvedconnectivity, and enhanced security. However, the integration of advanced technologies into smart grids, raises concernregarding cyberattacks such as distributed-denial-of-service (DDoS) attacks, which pose grave threat to the functionality andstability of the grid. To address these security challenges, smart grids traditionally implement intrusion detection systems (IDS)that analyse traffic logs from smart meters, but traditional IDS may face difficulties in detecting novel attacks such as subtle,multi-domain DDoS attacks. Towards securing smart grids, anomaly detection emerges as a crucial technique, integratedwith deep learning (DL), this technique can potentially identify deviations from normal, non-malicious network traffic, to detectcyberattacks, thereby enhancing grid security. However, it is seen that using user data for training DL models at the serverviolates data privacy regulations, which necessitates a balance between advanced anomaly detection and strict adherence todata privacy norms. Federated Learning (FL) has emerged as a suitable solution in this scenario, offering a privacy-focusedsolution allowing smart meters to train DL models with locally generated datasets and make predictions at the edge. In thiswork, we propose a hierarchical FL approach in smart meters for the 6G era, focusing on privacy-preserving anomaly detectionagainst DDoS attacks. Our work integrates a cloud-based service framework within an FL setup for smart grids, leveraging thescalability of cloud platforms and edge computing for efficient, secure, and cost-effective anomaly detection in line with 6Gtechnology requirements. Evaluation of our approach in local local simulation environment, using a workstation as the serverand Raspberry Pi devices as client nodes and cloud infrastructure provided by Amazon Web Services (AWS). Our goal is toinvestigate the feasibility of using cloud solutions to support federated learning-based anomaly detection in smart grids.Theperformance metrics between local and cloud simulations for our custom neural network showed that the variations betweentwo sets of simulations are not significant, and the proposed approach is suitable for deployment in real-world scenarios,especially for upcoming 6G-enabled smart grids where consistent performance is essential.