GNN-Based Predictive Consensus Node Selection for Blockchain-Enabled UAV Networks

Abstract

Uncrewed aerial vehicle (UAV) networks integrated with blockchain technology have been increasingly adopted to enable secure and decentralized coordination in distributed aerial systems. Within blockchain systems, the consensus mechanism plays a critical role in guaranteeing the consistency of shared data. However, the highly dynamic 3-D topology of UAV networks with unreliable wireless links can lead to consensus failure due to more frequent connectivity variations. In this article, the consensus node selection in a blockchain system becomes particularly critical. To tackle this problem, we develop a graph neural network (GNN)-based consensus node selection algorithm for blockchain-enabled UAV networks. GNN is particularly suitable for consensus stability prediction in highly dynamic UAV networks, as the UAV topology naturally forms a graph structure. To achieve a lightweight design, we only adopt relative angle, which is the most dominant factor in 3-D consensus stability proved by our theoretical analysis, as the edge feature in GNN. Simulation results demonstrate that GNN-based consensus node selection improves the consensus success rate by approximately 17.34%. Moreover, only incorporating the relative angle as an edge of GNN enables the proposed algorithm to achieve a near-optimal consensus success rate at low cost of computational overhead, with only 8.05-ms inference latency per consensus round.