Energy-efficient real-time workflow scheduling in edge–cloud environments based on multi-agent reinforcement learning

Wen, M., Ning, X., Pu, M., Liu, C., Wang, Q., Chen, X. ORCID: https://orcid.org/0000-0001-9267-355X and Cheng, L. (2026) Energy-efficient real-time workflow scheduling in edge–cloud environments based on multi-agent reinforcement learning. Applied Energy, 415. 127866. ISSN 1872-9118 doi: 10.1016/j.apenergy.2026.127866

Abstract/Summary

With the widespread deployment of Internet of Things devices, the complexity of resource-intensive workloads and the associated pressures on energy consumption are steadily increasing. This trend necessitates an efficient workflow scheduling mechanism between edge nodes and the central cloud to strike a dynamic balance between low latency and low energy consumption. Traditional heuristic or meta-heuristic-based scheduling methods struggle to address the challenges posed by dynamic environments and high computational complexity. To tackle these challenges, we propose MAES, a novel real-time workflow scheduling method based on Multi-Agent Deep Reinforcement Learning for edge-cloud collaborative computing environments. MAES adopts a multi-agent system architecture, where each virtual machine acts as an agent, optimizing workflow allocation through a collaborative approach of centralized training and decentralized execution. Experimental evaluations using real scientific workflows were conducted across different scenarios, including varying task numbers and types, workflow arrival intervals, edge-cloud VM ratios, and network jitter. Specifically, MAES reduces average latency and cost, while significantly lowering total system energy consumption, providing a highly energy-efficient solution for workflow scheduling in edge-cloud collaborative environments.

Item Type	Article
URI	https://centaur.reading.ac.uk/id/eprint/129379
Identification Number/DOI	10.1016/j.apenergy.2026.127866
Refereed	Yes
Divisions	Science > School of Mathematical, Physical and Computational Sciences > Department of Computer Science
Publisher	Elsevier
Download/View statistics	View download statistics for this item