A Task Offloading Strategy Based on Deep Reinforcement Learning

Abstract

Due to the limited bandwidth of Base Station (BS), without task offloading strategy in Mobile Edge Computing (MEC) scenarios, it will waste lots of resources of mobile edge devices. The greedy algorithm is an effective solution to optimize the task offloading strategy in MEC scenarios. It focuses on obtaining the maximal value, which consists of energy consumption and computation time from BS every step. However, the number of offloading tasks is another key optimized target, and it shows not ideal results with the greedy algorithm. In this paper, we aim to find a superior strategy to offload the tasks in MEC scenarios, which will fully obtain the resources from BS. Because this model can be considered as an optimization problem, we propose a task offloading strategy with deep reinforcement learning (TO-DRL). Weighted sum of task offloading number, energy consumption and computation time is the optimization target in this formulated problem. Numerical experiments demonstrate that compared with greedy algorithm, TO-DRL shows better performance in task offloading number.