AI-Driven Real-Time Relay Optimisation in Smart Urban NR-V2X Networks via Learning-to-Optimise Graph Neural Networks

A Learning-to-Optimise framework using Graph Isomorphism Networks with Edge Features (GINE) enables real-time multi-hop relay selection in NR-V2X networks. By training on offline Mixed-Integer Linear Programming (MILP) solutions, the GNN-based model maps vehicle and RSU radio-link characteristics to near-optimal connectivity decisions in complex urban environments.