This study proposes an intelligent routing protocol (iQSCR-DE-MADDPG) for Flying Ad Hoc Networks (FANETs). The protocol combines several state-of-the-art techniques likes Federated Reinforcement Learning (FRL), Graph Neural Networks (GNNs), Reconfigurable Intelligent Surfaces (RIS), and Hierarchical Reinforcement Learning (HRL) to improve routing efficiency in dynamic and resource-limited environments. First, the HRL-based clustering structures UAVs into mobility-conscious groups. DE-MADDPG is employed for power control within clusters, whereas GNNs inform topology-conscious routing decisions. Federated learning facilitates decentralized policy updates over UAVs while maintaining scalability and privacy. Simulation results concluded that the proposed iQSCR-DE-MADDPG solution significantly enhances important performance metrics which confirms in decreasing energy consumption by 30–35%, decreasing end-to-end delay by 27–32%, and improving packet delivery ratio (PDR) by 18–22% compared with existing routing protocols. These enhancements indicate the ability of proposed protocol to enable energy-efficient, scalable, and adaptive routing in future FANET environments.