Citation

G, Elavel Visuvanathan and Sayeed, Md Shohel and Yogarayan, Sumendra and Kuresan, Harisudha (2026) AMAFQL-BC: A trust-driven energy-efficient federated Q-learning architecture for secure and adaptive vehicular intelligence. Internet of Things, 39. p. 102067. ISSN 25426605

Text AMAFQL-BC_ A trust-driven energy-efficient federated Q-learning architecture for secure and adaptive vehicular intelligence.pdf - Published Version Restricted to Repository staff only Download (8MB)

Abstract

Intelligent Internet of Vehicles (IoV) systems require collaborative learning architectures that remain dependable under heterogeneous traffic conditions, highly non-independent and identically distributed (Non-IID) data, privacy exposure, and the risk of poisoned or manipulated model updates. To address these challenges, this study introduces the Adaptive Multi-Agent Federated Q-Learning with Blockchain Security (AMAFQL-BC) framework, which strengthens cooperative decision learning through Multi-Agent Federated Q-Learning (MAFQL) driven by a Multi-Agent Deep Deterministic Policy Gradient (MADDPG) based actor–critic strategy for coordinated policy optimization across distributed vehicular nodes. The framework incorporates a Delegated Proof-of-Trust (DPoT) blockchain consensus mechanism to assess contribution reliability and filter adversarial or inconsistent model updates, while Advanced Encryption Standard in Galois/ Counter Mode (AES-GCM) authenticated communication ensures confidentiality and integrity during Vehicle-to-Everything (V2X) model exchange. System adaptability is further enhanced through Adaptive Resource Allocation (ARA) for bandwidth-aware scheduling, the Sculptor Optimization Algorithm (SOA) for hyperparameter refinement, and Energy-Aware Federated Optimization (EAFO) to balance client participation, computation effort, and communication cost under energy constraints. Experimental evaluation demonstrates that AMAFQL-BC attains 98.47% accuracy on the UNSW-NB15 dataset and 99.31% accuracy on the CIC-IDS-2017 dataset, with F1-Scores of 97.99% and 98.97%, respectively, while retaining strong performance under Non-IID learning conditions and maintaining robustness against poisoning attacks, preserving 97.18% accuracy even at 40% malicious nodes. The framework also achieves low latency (18–52 ms across scaling levels), reduced consensus time (0.07–2.55 s), and higher blockchain throughput (up to 406 transactions per second) compared with baseline approaches, indicating improved convergence stability, energy-efficient participation, and secure federated coordination. Overall, the AMAFQL-BC framework provides a scalable, resilient, and trustworthy learning architecture for real-time vehicular intelligence, enabling robust decision-making and efficient collaboration across large-scale IoV networks.

Item Type:	Article
Uncontrolled Keywords:	Internet of vehicles
Subjects:	Q Science > QA Mathematics > QA71-90 Instruments and machines > QA75.5-76.95 Electronic computers. Computer science
Divisions:	Faculty of Information Science and Technology (FIST)
Depositing User:	Ms Rosnani Abd Wahab
Date Deposited:	02 Sep 2026 08:56
Last Modified:	02 Sep 2026 08:56
URII:	http://shdl.mmu.edu.my/id/eprint/16553

Downloads

Edit (login required)