A Deep Reinforcement Learning–Based Trust Management Model for Secure Routing in Vehicular Ad-Hoc Networks under Real-Time Traffic Variations

Résumé fourni par la source

Vehicular Ad-Hoc Networks (VANETs) are a fundamental component of intelligent transportation systems, enabling distributed communication among vehicles and roadside infrastructure.However, their highly dynamic topology, rapid node mobility, and vulnerability to routing and trust-based attacks make secure and adaptive route selection difficult under real-time traffic variations.To address these challenges, this paper proposes a secure routing framework that integrates trust modeling, trust quality assessment, and policy optimization through Deep Reinforcement Learning (DRL).The proposed routing strategy employs a Dueling Deep Q-Network (D-DQN) in a conflictdriven dynamic environment and is supported by a trust evaluation module that maintains behaviorbased trust scores and tracks the malicious tendency of participating nodes.The model is trained and evaluated through a realistic hybrid co-simulation framework based on Simulation of Urban Mobility (SUMO) and Network Simulator (NS3).Its performance is compared with two baseline routing methods, namely Asynchronous Advantage Actor-Critic (A3C) and Dynamic Source Routing (DSR).Experimental results show that the proposed D-DQN-based routing strategy achieves an average delivery ratio of 92.6%, an average latency of 84.3 ms, and a trust convergence value of 0.89.Compared with DSR under the same mobility and threat conditions, the proposed method delivers 24.7-fold faster policy-learning convergence, 19.3-fold higher trust stability, and 21.4-fold lower power consumption.Although A3C performs more stably than DSR, it remains less responsive to latency and shows weaker trust propagation.Overall, the results indicate that integrating D-DQN into trust-based routing can substantially improve routing reliability, policy stability, and real-time decision-making in VANETs.

Ce résumé expose les affirmations des auteurs. BNTIC ne l’interprète pas comme une validation indépendante des résultats.