Task Offloading of Edge Computing Network Based on Lyapunov and Deep Reinforcement Learning
Xudong Qiao, Yongxin Zhou · 2024
Reinforcement learning based task offloading is a promising research direction in edge computing. This paper proposes a Deep Reinforcement Learning (DRL) Task Offloading framework (LyDDPG) based on Lyapunov optimization, which leverages the strengths of both Lyapunov optimization and DRL. LyDDPG aims to minimize device energy consumption and reduce queue backlog under long-term data queue stability and delay constraints by decoupling the original optimization problem into an independent slot task offloading optimization problem. A multi-user edge computing network with time-varying wireless channels and random user task data arriving in a sequence time range is considered in this experiment. The simulation results show that the LyDDPG algorithm minimizes the energy consumption and queue backlog under the condition of satisfying the long-term stability constraints. The framework improves the adaptability and performance of the system in a dynamic network environment, and provides an efficient way to solve the problem of task offloading and resource allocation.