PCO-VN: DESIGNING PREDICTIVE COMPUTATIONAL OFFLOADING MODEL FOR VEHICULAR NETWORKS USING HYBRID MOBILE EDGE AND CLOUD COMPUTING TECHNOLOGIES

Abstract

A Computational Offloading Model for Vehicular Networks leveraging Hybrid Mobile Edge and Cloud Computing Technologies presents an innovative approach to address the computational constraints of vehicles in dynamic environments. By integrating mobile edge computing (MEC) and cloud computing, this model aims to offload computational tasks from vehicles to nearby edge servers or remote cloud data centers, thereby enhancing the efficiency and scalability of vehicular networks. However, despite its promising benefits, several limitations exist. By creating a predictive offloading model with cloud and hybrid mobile edge computing technologies, this study seeks to optimize the computation offloading workflow in the vehicular cloud environment. In this study, Predictive Computation Offloading in Vehicular Network (PCO-VN) which begins by examining where RSUs should be placed about vehicles, treating RSUs as mobile edge servers. Artificial Humming Bird Optimization (AHBO), an optimization algorithm that maximizes efficiency by considering multiple variables, is used to achieve effective RSU placement. When RSUs are positioned, the surrounding cars group together to create vehicular cloudlets, which are areas where AVs and RSUs can exchange resources and data by communicating within their respective communication ranges. An Attention-based Asymmetry Encoder-Decoder Network (AEDNet) is used in this framework to forecast the trajectory of autonomous vehicles (AVs) while considering a variety of parameters to improve prediction accuracy. Afterwards, RSUs use a Multi-Criteria Decision Making (MCDM) algorithm to rate the AVs within their range from low to high moving AVs based on the trajectory prediction results. Deep Reinforcement Learning (DRL) technique is used to do virtualized compute offloading based on the ranking outcomes. To be more precise, the DRL agent finds AVs that are available for unloading first. A virtualized computing server is constructed for timely offloading, ensuring effective resource use; if all AVs are overburdened, the task is offloaded to RSUs. The proposed PCO-VN model is evaluated in terms of proving its efficiency, in which PCO-VN is outperformed other existing works.