Q-learning-driven task offloading and collaborating in edge networks

Liu, Yuxin

doi:10.1016/j.ins.2026.123245

Q-learning-driven task offloading and collaborating in edge networks

2026-02-01

0

PRE

AI

DOI：10.1016/j.ins.2026.123245

Original

Original request for help

Save

Abstract

En 中文

In edge computing, it is reasonable to create different service instances on different edge servers to complete tasks efficiently and to enhance task offloading and collaboration between edge servers. However, due to the dynamic arrival of tasks, efficient task collaboration still faces challenges. In this paper, a Q-Learning-Driven Task Offloading and Collaboration (QDTOC) system is proposed to intelligently reduce the average task completion time and improve the utilization of system resources. In the QDTOC system, the edge servers form a virtual service alliance to provide task computing services, and each edge server updates its status information promptly through the edge network. Then, a Q-Learning-based framework is defined in which task offloading and collaboration actions are mapped into the Q-Learning space. Finally, a Q-Learning-based task offloading and collaboration scheme is proposed in which each edge server dynamically learns and optimizes its own service instance number and task scheduling requests to maximize system resource efficiency. We demonstrate the scheme's effectiveness through extensive experiments. Compared to those of the FR-NT, FR-T, DR-NT, and DR-T schemes, the average task network delay of the proposed scheme is decreased by 63.96%, 56.45%, 54.93%, and 41.72%, respectively, and its system resource utilization is also superior.

Keywords:

Edge computing

Mobile wireless networks

Q-Learning

Task offloading and collaboration

Service instance

Q-learning-driven task offloading and collaborating in edge networks

Abstract

Journal

Organization