Adaptive and Resilient T-PBFT Consensus for Internet of Vehicles Using Multi-Objective Optimization and Reinforcement Learning

Abstract The Internet of Vehicles (IoV) enables continuous communication among vehi-cles and roadside infrastructures, but its highly dynamic conditions, changing network density, resource constraints and exposure to attacks make efficient blockchain consensus challenging. This paper proposes an adaptive T-PBFT con-sensus framework that combines a trust evaluation mechanism inspired by the principles of EigenTrust, the NSGA-III multi-objective optimization algorithm and Deep Q-Network (DQN) reinforcement learning. The proposed approach jointly considers validator selection and T-PBFT configuration while optimiz-ing consensus latency, communication overhead, computational cost, security risk, validator trust and availability. NSGA-III generates candidate configura-tions representing different trade-offs among these objectives, while the DQN agent selects a suitable configuration according to the current network state. The framework is evaluated through a Python-based simulation under nor-mal, high-density, Sybil-attack, denial-of-service (DoS) and combined stress scenarios. Five independent random seeds are used for each scenario and the results are compared with a fixed T-PBFT configuration and an NSGA-III-only approach. The results show that adaptive approaches substantially improve consensus performance under non-DoS conditions, while the hybrid approach provides scenario-dependent benefits, particularly under Sybil attacks. How-ever, under severe DoS conditions, all evaluated methods experience complete consensus failure. These findings demonstrate the potential of combining multi-objective optimization and reinforcement learning for adaptive consensus, while highlighting the need for stronger attack-resilience mechanisms.