FedGNN: Federated Graph Neural Networks for Privacy-Preserving Cyber-Resilient Energy Optimization in IoT-Based Smart Grids

The rapid integration of Internet of Things (IoT) devices and distributed energy resources into smart grids has improved monitoring, control, and energy efficiency. However, it also exposes the grid to cyberattacks and privacy risks, as increased connectivity and data exchange can significantly disrupt energy management and system stability. Studies focused on centralized cybersecurity mechanisms that lacked scalability and did not emphasize the inherent graph structure of power networks. This study proposes a privacy-preserving and cyber-resilient energy-optimization framework, FedGNN, for IoT-enabled smart grids that jointly integrates federated learning, graph neural network-based trust inference, and trust-aware energy dispatch. The framework dynamically learns node-level trust scores from multi-feature measurements, including load, voltage, frequency, renewable generation, and battery storage, and incorporates them into real-time energy optimization. Results demonstrate that the proposed approach improves system resilience up to 12%, mitigates the impact of compromised nodes, and maintains operational reliability, while preserving the privacy of distributed data. A comparative analysis with baseline methods shows the proposed framework’s superior performance in energy deviation, resilience, and trust-aware decision-making. The results highlight the potential of integrating AI-driven trust mechanisms with federated learning for secure and efficient energy management in future IoT-enabled smart grids.