Research Article
Hybrid Federated and Multi-agent DRL-Based Resource Allocation in Digital Twin-IoV Networks
@INPROCEEDINGS{10.1007/978-3-031-67357-3_8, author={Bishmita Hazarika and Anal Paul and Keshav Singh}, title={Hybrid Federated and Multi-agent DRL-Based Resource Allocation in Digital Twin-IoV Networks}, proceedings={Industrial Networks and Intelligent Systems. 10th EAI International Conference, INISCOM 2024, Da Nang, Vietnam, February 20--21, 2024, Proceedings}, proceedings_a={INISCOM}, year={2024}, month={7}, keywords={Asynchronous federated learning Deep reinforcement learning Internet of Vehicles}, doi={10.1007/978-3-031-67357-3_8} }- Bishmita Hazarika
Anal Paul
Keshav Singh
Year: 2024
Hybrid Federated and Multi-agent DRL-Based Resource Allocation in Digital Twin-IoV Networks
INISCOM
Springer
DOI: 10.1007/978-3-031-67357-3_8
Abstract
This study introduces a combined machine learning strategy for optimizing resource distribution within a digital twin (DT) setup, aimed at offloading tasks in UAV-supported Internet-of-Vehicles networks. By merging asynchronous federated learning with multi-agent deep reinfo- rcement learning, we aim to boost the rate of task completion while simultaneously reducing consumed energy and delay, and consequently improving the system’s overall performance. We introduce a DT architecture within an IoV network supported by UAV for V2V and V2I task transitions accommodating three modes of processing tasks and a trio of task categories. We then formulate a problem focusing on elevating the overall effectiveness of the system, aiming to reduce both delay and energy usage. Addressing this complex challenge, our solution introduces a multi-agent DRL algorithm designated as MARS, dedicated to the effective distribution of resources within the IoV network supported by digital twins. This algorithm is refined through a blended AFL method, which we refer to as HAFL. MARS enhances the distribution of resources across various computational settings to augment the total system efficacy. Finally, comprehensive simulations affirm our approach’s superiority, bench- marked against a variety of established methodologies.
