Green-Oriented Offloading and Resource Allocation by Reinforcement Learning in MEC

Abstract

Mobile Edge Computing (MEC) is a promising approach to satisfy the increasing demand of computationintensive applications in fifth-generation (5G) networks. In this MEC networks system, Smart Mobile Devices (SMDs) can implement tasks migration in close proximity to a MEC server via wireless channels. In this paper, we formulate a Green-Oriented Problem (GOP) in MEC networks to minimize the cost of energy consumption for all SMDs in MEC system. In order to address the problem, we jointly optimize the offloading decision, wireless communication resource allocation and computational resource allocation while meeting the latency constraints. Apart from this, noting that finding an optimal policy of GOP in this MEC system is a Mixed Integer Linear Program (MINLP) problem, we use the Reinforcement Learning (RL) method which takes a long-term goal into consideration. Our numerical experiments demonstrate that the proposed algorithm improves energy efficiency of the computation offloading in MEC system.