Accelerating Distributed Average Consensus in Wireless Sensor Networks via GNN-Based Broadcast Probability Optimization

Achieving efficient distributed average consensus is crucial for collaborative applications in wireless sensor networks (WSNs). Traditional gossip-based methods encounter difficulties in balancing communication efficiency and consensus rate, especially in dynamic and resource-constrained wireless environments. To overcome these challenges, a graph neural network (GNN), specifically the message passing neural network (MPNN) framework, is proposed to optimize node broadcast probabilities for the probabilistic broadcast gossip scheme. By employing MPNN with attention mechanisms, the proposed method dynamically allocates broadcast probabilities based on both local node characteristics and global network topologies. Extensive simulations reveal that the proposed method significantly surpasses heuristic and optimization-based baselines, achieving a substantial reduction in communication costs. These results highlight the potential of GNNs in advancing distributed consensus protocols for WSNs.