Federated Reinforcement Learning with Blockchain-Integrated Trust for Secure Routing in 5G-Enabled Vehicular Networks

Secure and low-latency routing in 5G-enabled Vehicular Ad Hoc Networks (VANETs) is fundamentally challenged by high node mobility, dynamic topology changes, and the threat of malicious routing behaviour. Existing protocols including AODV and standalone reinforcement learning (RL) approaches fail to jointly address privacy-preserving distributed learning, trust-based route verification, and real-time adaptability. This paper proposes the Federated Reinforcement Learning-based Secure Routing (FRL-SR) framework, which integrates decentralized Q-learning at vehicular agents, weighted federated aggregation at roadside unit (RSU) edge servers, and a blockchain-based trust verification layer operating on PBFT consensus. Vehicles learn local routing policies from neighborhood observations without sharing raw data; RSUs periodically aggregate global models via data-weighted FedAvg; and every routing update is validated through blockchain transactions recording node ID, model hash, trust score, and digital signature. Extensive simulations on NS-3.39 with SUMO mobility and 5G NR-V2X communication demonstrate that FRL-SR achieves a mean Packet Delivery Ratio (PDR) of 94.0%, an average End-to-End Delay (E2E) of 42 ms, a Trust Violation Rate (TVR) below 1.5%, and model convergence in 14 federated rounds outperforming AODV, RL-Only, and FedRL baselines by 18–22%, 28–35%, and 23%, respectively. The results confirm the viability of FRL-SR as a scalable, secure, and latency-optimal routing solution for next-generation intelligent transportation systems.