A deep learning and blockchain enabled adaptive routing framework for secure Vehicular Ad Hoc Networks

Abstract

Vehicular Ad hoc Networks (VANETs) are one of the most important enablers of Intelligent Transportation Systems (ITSs), however, their dynamic topologies and susceptibility to malicious activities provide a serious challenge to the routing efficiency and security. To address these challenges, this paper introduces a new Deep Learning-based Adaptive Routing with Blockchain-enabled Trust Management (DL-ARBTM) mechanism that incorporates two mutually complementary stages. Deep Reinforcement Learning (DRL) is used during the adaptive routing phase, where the routing process is modeled as a Markov Decision Process (MDP), and the vehicles at the routing node are autonomous agents, which monitor the quality of links, mobility of nodes and the density of items to decide the best next-hop node dynamically. An Actor-Critic agent guarantees efficient policy learning and stabilizing in highly dynamic environments. During the trust management stage, blockchain technology with Practical Byzantine Fault Tolerance (PBFT) consensus has an immutable distributed registry of node trust scores that allow decentralized checks on forwarding behavior, key exchange, and efficient mitigation of Sybil, packet dropping, and misinformation attacks. The two-layered integration guarantees that DRL chooses routes based on nodes that are trusted and therefore improves reliability and resiliency. Unlike existing VANET routing solutions that treat adaptability and security independently, the proposed DL-ARBTM uniquely integrates actor-critic deep reinforcement learning with blockchain-based decentralized trust management. This joint design enables trust-aware adaptive routing, achieving higher packet delivery ratio, throughput, and trust accuracy with lower end-to-end delay compared to AODV, GPSR, Q-learning, and blockchain-only approaches, thereby advancing secure and scalable ITS communications.