This paper investigates an algorithm that combines Graph Neural Networks (GNN) and Deep Reinforcement Learning (DRL) to optimize bandwidth allocation in UAV Ad Hoc Networks. Traditional routing protocols struggle with poor adaptability and low resource utilization in dynamic environments. By leveraging GNN to extract dynamic topological features and integrating DRL for decision optimization, the proposed GNN-DRL model significantly improves bandwidth utilization and stability. Experimental results show that GNN-DRL improves bandwidth utilization by approximately 6% across different topologies and exhibits stronger robustness under high utilization conditions, offering new insights into efficient routing optimization for UAV swarms.