Blockchain-Enhanced Edge IIoT Framework with Graph Neural Networks for Real-Time Smart Grid Anomaly Detection

Abstract

In this paper, the concept of blockchain is combined with the edge IIoT framework and Graph Neural Networks (GNNs) to provide real-time anomaly detection and mitigation in smart grid infrastructures. This proposed architecture is based on three key components: decentralized blockchain security, edge Artificial Intelligence (AI) processing, and spatial-temporal anomaly learning through the Graph Convolutional Network (GCN) and the Temporal Graph Network (TGN). The framework was tested in a simulated smart grid environment with over 10,000 nodes that were interconnected and with various anomaly types such as cyberattacks, line faults, sensor failures, and voltage spikes. The experimental results demonstrate that the TGN model has an anomaly detection accuracy of 94.5% and a recall of 92.75%, which is about 5% higher than the GCN model, and the false positive rate has been reduced by 18%. The framework also reduced computation cost by 22% compared to the traditional centralized approach. The proposed framework required approximately 50 seconds for edge AI processing of all experimental samples, while blockchain validation introduced an additional 25 seconds of overhead, resulting in a total end-to-end processing latency of approximately 75 seconds. In addition, blockchain-based smart contracts facilitated an automated way of anomaly mitigation via node isolation and adaptive load redistribution. The proposed framework shows the potential of how blockchain, edge AI, and GNN models can enhance the smart grid system's resilience, security, and operational intelligence.