Optimizing traffic safety message dissemination and resource allocation using adaptive deep reinforcement learning in fog-enabled internet of vehicles network

Abstract

The rapid transformation of the Internet of Vehicles (IoV) necessitates robust solutions to ensure traffic safety, optimal resource utilization, and reliable data dissemination. However, traffic safety message dissemination is hindered by fluctuating traffic densities, network congestion, and message priorities, while existing systems exhibit minimal adaptation to real-time load variations due to delays and inconsistent message transmission. In this study, we propose a novel FOG-DRL based framework that integrates Deep Reinforcement Learning (DRL) and Fog Computing (FC) to address these vehicular network challenges. A Vehicular Ad-hoc Network (VANET) is constructed, whereby the system exploits DRLs (Deep Q-Learning (DQN) and Proximal Policy Optimization (PPO)) for offloading, optimal resource allocation, and equitable task distribution within the network. Fog Computing is utilized for larger data with less latency while keeping the communication overhead lower. The DRL-based second adaptive algorithm will be in charge of optimizing the network’s performance and alleviating congestion. Energy consumption is minimized using third DRL-based optimization methods, with a Round Robin algorithm for resource allocation working within the limits of performance required for establishing effective allocation. Besides, the system incorporates mechanisms to prevent the dissemination of fraudulent messages in vehicle-to-vehicle (IoV) and vehicle-to-infrastructure (V2I) interactions, which ensure data integrity and reliability. Comprehensive performance evaluations are conducted, including metrics such as energy usage, latency, resource utilization, and throughput to varying numbers of IoV nodes. This work lays the foundation to develop safer and more competent transportation systems to meet the rising demands of IoV networking through advanced AI and distributed computing interventions.