Title: DHA-RL: three-tier hybrid offloading network optimisation for the Internet of Things

Authors: Sili He; Zhenjiang Zhang; Qing-An Zeng

Addresses: School of Electronic and Information Engineering, Beijing Jiaotong University, Beijing 100044, China ' School of Electronic and Information Engineering, Beijing Jiaotong University, Beijing 100044, China ' Department of Computer Systems Technology, North Carolina A&T State University, 1601 E. Market Street, Greensboro, NC 27411, USA

Abstract: With the rise of 5G and increasing demand for computing-intensive applications, efficient solutions are needed, especially in remote areas where cellular networks struggle. This paper proposes a three-tier hybrid offloading framework involving local nodes (devices), edge nodes (satellites), and cloud nodes (ground stations). It introduces a novel RL-based offloading strategy, decodable hybrid actions reinforcement learning (DHARL), to optimise latency and energy consumption within limited local and satellite resources. Using conditional variational autoencoders (VAE), the strategy learns dependencies in the hybrid action space. Constraints on the action space and supervision of representation shifts address issues like inadequate sampling and variations. Extensive simulations show that DHARL outperforms existing methods in task latency, energy consumption, and system cost, proving its potential for efficient computation offloading in Internet of Things (IoT) environments.

Keywords: computing-intensive applications; offloading framework; RL-based offloading strategy; hybrid action space; conditional variational autoencoders; latency considerations.

DOI: 10.1504/IJMNDI.2024.144008

International Journal of Mobile Network Design and Innovation, 2024 Vol.11 No.2, pp.76 - 87

Received: 21 Jun 2024
Accepted: 30 Nov 2024
Published online: 20 Jan 2025 *

Full-text access for editors Full-text access for subscribers Purchase this article Comment on this article