A Deep Reinforcement Learning Approach Towards Computation Offloading for Mobile Edge Computing

摘要

In order to improve the quality of service for users and reduce the energy consumption of the cloud computing environment, Mobile Edge Computing (MEC) is a promising paradigm by providing computing resources which is close to the end device in physical distance. Nevertheless, the computation offloading policy to satisfy the requirements of the service provider and consumer at the same time within a MEC system still remains challenging. In this paper, we propose an offloading decision policy with three-level structure for MEC system different from the traditional two-level architecture to formulate the offloading decision optimization problem by minimizing the total cost of energy consumption and delay time. Because the traditional optimization methods could not solve this dynamic system problem efficiently, Reinforcement Learning (RL) has been used in complex control systems in recent years. We design a deep reinforcement learning (DRL) approach to minimize the total cost by applying deep Q-learning algorithm to address the issues of too large system state dimension. The simulation results show that the proposed algorithm has nearly optimal performance than traditional methods.