Task Offloading based on Deep Reinforcement Learning with LSTM for Mobile Edge Computing

Abstract

Mobile devices may offload computationally intensive tasks to edge servers for processing in Mobile Edge Computing (MEC), thereby improving the quality of experience. Heuristic algorithms are feasible for MEC offloading decisions and resource allocation, but they are not suitable for high real-time MEC systems, ignoring the impact of channel dynamic changes on the computational offloading problem. In this paper, we construct a MEC system in a time-varying fading channel scenario and propose a deep reinforcement learning algorithm based on LSTM (DR-LSTM) to solve the joint optimization problem of task offloading decision and resource allocation. The DR-LSTM is combined with an order-preserving quantization algorithm to generate offloading decision, and a linear relaxation method is used to solve the resource allocation problem. Finally, it is verified through simulations that the DR-LSTM can effectively solve the task offloading and resource allocation problem under this model.