VAST-GCN: An Attention-Driven Graph Convolutional Network (GCN) for Robust Cluster Head Selection in Vehicular Ad-Hoc Networks

Vehicular Ad-Hoc Networks (VANETs) need smart and flexible communication protocols to deal with fast-moving vehicles and ever-changing network structures. Picking the right cluster head (CH) plays a key role to keep connections stable and reduce routing overhead. This paper presents VAST-GCN (Vehicular Attention-based Spatial-Temporal Graph Convolutional Network), a new model that uses attention to make vehicle grouping and CH selection better across different network sizes. VAST-GCN mixes Graph Convolutional Networks (GCNs) with Spatial, Temporal, and Channel Attention systems. Approach in vehicle settings with 100, 500, and 1000 vehicles has been tested using real-time info like speed and place. The design has transformer blocks to model time-based features and attention modules to improve space and feature relationships leading to better vehicle data. Data have been grouped using the K-Means method and checked with modularity score, silhouette score, and group density. At the time of comparison, it has been observed that VAST-GCN does better than regular GCN and MIXHOP GCN models in cutting down loss making better community structures, and keeping CHs stable when there are few vehicles or they’re moving fast. The proposed VAST-GCN framework exhibits clear advantages over existing spatio-temporal GNNs by delivering superior modularity, silhouette scores, and cluster head stability across diverse vehicular scenarios. Its attention-driven architecture not only improves clustering accuracy but also reduces packet delay and enhances throughput, highlighting its excellence as a robust and scalable solution for dynamic VANET environments.