Dynamic Task Offloading in MEC-Enabled IoT Networks: A Hybrid DDPG-D3QN Approach
Han Hu, Dingguo Wu, Fuhui Zhou, Shi Jin, Rose Qingyang Hu
Abstract
Mobile edge computing (MEC) has recently emerged as an enabling technology to support computation-intensive and delay-critical applications for energy-constrained and computation-limited Internet of Things (IoT). Due to the time-varying channels and dynamic task patterns, there exist many challenges to make efficient and effective computation offloading decisions, especially in the multi-server multi-user IoT networks, where the decisions involve both continuous and discrete actions. In this paper, we investigate computation task offloading in a dynamic environment and formulate a task offloading problem to minimize the average long-term service cost in terms of power consumption and buffering delay. To enhance the estimation of the long-term cost, we propose a deep reinforcement learning based algorithm, where deep deterministic policy gradient (DDPG) and dueling double deep Q networks (D3QN) are invoked to tackle continuous and discrete action domains, respectively. Simulation results validate that the proposed DDPG-D3QN algorithm exhibits better stability and faster convergence than the existing methods, and the average system service cost is decreased obviously.