Deep Reinforcement Learning-based UAV-assisted Mobile Edge Computing Offloading and Resource Allocation Decision

Abstract

Nowadays, the surge of user data traffic has brought great challenges to the computing and energy capacity of mobile terminals (MTs). Mobile edge computing (MEC) technology is reckoned to be an efficient method to alleviate this problem. It can transfer tasks to MEC server and improve quality of service (QoS). In case of network failure, unmanned aerial vehicle (UAV) is deployed as a data transmission hub connecting MEC server to restore the network. In this article, we consider a UAV transmission hub (UTH) to communicate with the macro base station (MBS). MTs can offload tasks to MBS for processing through UTH, and the MEC server in MBS allocates computing resources to MTs. We raise a computing offloading and resource allocation decision scheme based on deep deterministic policy gradient (DDPG). The scheme considers the continuous generation of dynamic tasks, and the optimization objectives is to minimize the long-term average system cost. The simulation experiment datas verify the performance of DDPG-Based offloading and resource allocation decision scheme. It can validly optimize the average system cost in a random dynamic environment.