Computation Offloading Strategy Based on Deep Reinforcement Learning in Cloud-Assisted Mobile Edge Computing

Abstract

Mobile edge computing (MEC) is a new computing paradigm that migrates rich computing and storage resources to the edge of the network. However, compared with traditional cloud computing, mobile edge computing is constrained in computing capacity, especially under the scenario of dense population. In this paper, a Cloud-Assisted Mobile Edge (CAME) computing framework is used to study the problem of computation offloading and resource allocation. First, the transmission delay as well as computation delay that computation jobs may experience, the transmission energy as well as computation energy that the computing system would consume were modeled. Then, the weighted sum of the delay and energy-efficient minimization computation offloading problem was formulated, constrained to the maximum latency and server resources. After that, a DQN algorithm based on reinforcement learning is proposed. In order to avoid the problem of excessive state space and overestimation, a DDQN offloading algorithm is proposed. Simulation results show that the offloading algorithm DDQN proposed in this paper can reduces the weighted sum of delay and energy consumption effectively.