An Efficient Computation Offloading Approach in Multi-access Edge Computing Using Deep Reinforcement Learning

Abstract

Multi-access Edge Computing (MEC) is a promising paradigm to empower the internet of things (IoT) devices stronger ability for complicated applications and long hours of service. In this paper, we consider a binary offloading scenario with a MEC server and multiple smart devices, aiming to minimize the total energy consumption of smart devices. In order to obtain the offloading strategy quickly, we propose a deep reinforcement learning (DRL) based method which can directly output the solution without iterations that the conventional numerical optimization methods should have. To explore the offloading action space effectively, we propose a Hamming distance-based action exploration method to discover the optimal action for the update of policy networks. Numerical results show that the proposed method has good performance in the prediction accuracy and the exploring success rate.