Task Reverse Offloading with Deep Reinforcement Learning in Multi-Access Edge Computing

Abstract

The Multi-access Edge Computing (MEC) technology's quick development greatly benefits the Collaborative Mobile Infrastructure System (CMIS). To combine the data and produce tasks, crowd-sensing data will be transferred to the MEC server in CMIS. Nevertheless, if there are too many devices, it becomes extremely difficult for MEC to decide appropriately based on the data from the devices and infrastructure. This study builds a framework for reverse offloading that carefully balances the relationship between task completion time and user mobile energy consumption. Moreover, to decrease system use generally, an adaptive optimal reverse offloading method based on Deep Q-Network is created (DQN). The results of the simulations demonstrate that the suggested approach may successfully minimize energy consumption and work latency when compared to full local and fixed offloading techniques.