Carbon-Aware Autonomous AI Systems: Reinforcement Learning for Sustainable Cloud and Edge Computing

Notice

This is an unedited manuscript accepted for publication and provided as an Article in Press for early access at the author’s request. The article will undergo copyediting, typesetting, and galley proof review before final publication. Please be aware that errors may be identified during production that could affect the content. All legal disclaimers of the journal apply.

Year : 2026 | Volume : 16 | 02 | Page :
    By

    Anjali Rout,

  • Yashika Chandra,

  1. UG Student, Department of Computer Science Engineering, Greater Noida Institute of Technology, Greater Noida, Uttar Pradesh, India
  2. UG Student, Department of Computer Science Engineering, Greater Noida Institute of Technology, Greater Noida, Uttar Pradesh, India

Abstract

The field of communication and information technology is expanding quickly. Because of this, a significant amount of carbon emissions are produced by cloud data centres and edge computing nodes. In fact they are now responsible for 3 to 4 percent of the worlds total greenhouse gas emissions. Most of the time people who manage these resources focus on how they are working and how quickly they can get things done.. They do not think about how the carbon intensity of the electricity grid changes over timedone. is where CarbonRL comes in. CarbonRL is an intelligence system that is aware of carbon emissions and can make its own decisions to reduce the negative impact on the environment. The system uses real-time carbon intensity signals to manage workloads buy energy and reduce the carbon footprint across cloud and edge infrastructures. CarbonRL uses a kind of learning called reinforcement learning. This is based on something called Proximal Policy Optimization and a Long Short-Term Memory state encoder. This helps the system understand things that happen over time like when carbon will be high or low in the grid when energy will be available and when work will arrive.The system also has a way of figuring out what is important. It has to balance things like meeting the promises it makes to users keeping energy costs low and reducing carbon emissions. Many experiments were done using carbon intensity data from the Electricity Maps API and workload data from Google Cloud and Azure. The results show that CarbonRL reduces carbon emissions by 38.7 percent and lowers energy costs by 24.3 percent compared to systems. Additionally, it ensures that more than 99.1% of the time, it fulfils its commitments to consumers. The system can also move work to data centers that use cleaner energy. This reduces the carbon cost of sending data. CarbonRL was compared to seven systems and was tested in many ways to make sure it works well. Overall CarbonRL is a step, towards making cloud and edge computing more sustainable. By using real-time carbon data to manage resources the system is more efficient reduces energy costs and lowers the carbon footprint of cloud data centers and edge computing nodes.

Keywords: Carbon-aware computing; deep reinforcement learning; cloud resource management; edge computing; sustainable AI; Proximal Policy Optimization; marginal carbon intensity; green data centers; multi-objective optimization; SLA-constrained scheduling.

How to cite this article:
Anjali Rout, Yashika Chandra. Carbon-Aware Autonomous AI Systems: Reinforcement Learning for Sustainable Cloud and Edge Computing. Journal of Energy, Environment & Carbon Credits. 2026; 16(02):-.
How to cite this URL:
Anjali Rout, Yashika Chandra. Carbon-Aware Autonomous AI Systems: Reinforcement Learning for Sustainable Cloud and Edge Computing. Journal of Energy, Environment & Carbon Credits. 2026; 16(02):-. Available from: https://journals.stmjournals.com/joeecc/article=2026/view=249526


References

  1. Valavanidis A. Global electricity generation from renewable sources. Journal. 2016;2(5):99-110.
  2. Li B, Samsi S, Gadepally V, Tiwari D. Clover: Toward sustainable ai with carbon-aware machine learning inference service. InProceedings of the International Conference for High Performance Computing, Networking, Storage and Analysis 2023 Nov 12 (pp. 1-15).
  3. Radovanović A, Koningstein R, Schneider I, Chen B, Duarte A, Roy B, Xiao D, Haridasan M, Hung P, Care N, Talukdar S. Carbon- aware computing for datacenters. IEEE Transactions on Power Systems. 2022 May 6;38(2):1270-80.
  4. Souza A, Jasoria S, Chakrabarty B, Bridgwater A, Lundberg A, Skogh F, Ali-Eldin A, Irwin D, Shenoy P. Casper: Carbon-aware scheduling and provisioning for distributed web services. InProceedings of the 14th International Green and Sustainable Computing Conference 2023 Oct 28 (pp. 67-73).
  5. Yang J, Saad Z, Wu J, Niu X, Leung H, Drew S. A Survey on Task Scheduling in Carbon-Aware Container Orchestration. arXiv preprint arXiv:2508.05949. 2025 Aug 8.
  6. Brander M, Sood A, Wylie C, Haughton A, Lovell J. Technical Paper| Electricity-specific emission factors for grid electricity. Ecometrica, Emissionfactors. com. 2011 Aug.
  7. Wilson EL, DiGregorio AJ, Villanueva G, Grunberg CE, Souders Z, Miletti KM, Menendez A, Grunberg MH, Floyd MA, Bleacher JE, Euskirchen ES. A portable miniaturized laser heterodyne radiometer (mini-LHR) for remote measurements of column CH4 and CO2. Applied Physics B. 2019 Nov;125(11):211.
  8. Beloglazov A, Abawajy J, Buyya R. Energy-aware resource allocation heuristics for efficient management of data centers for cloud computing. Future generation computer systems. 2012 May 1;28(5):755-68.
  9. Kliazovich D, Bouvry P, Khan SU. GreenCloud: a packet-level simulator of energy-aware cloud computing data centers. The Journal of Supercomputing. 2012 Dec;62(3):1263-83.
  10. Dhawan R, Patel MS. Carbon-Aware AI Control Plane for DevOps Automation: A Reference Architecture and Next-Generation Sustainability Framework. IEEE Access. 2026 Jan 20.
  11. Mao Y, Zhang J, Letaief KB. Dynamic computation offloading for mobile-edge computing with energy harvesting devices. IEEE Journal on Selected Areas in Communications. 2016 Sep 20;34(12):3590-605.
  12. Xu J, Chen L, Zhou P. Joint service caching and task offloading for mobile edge computing in dense networks. InIEEE INFOCOM 2018-IEEE Conference on Computer Communications 2018 Apr 16 (pp. 207-215). IEEE.
  13. Roijers DM, Vamplew P, Whiteson S, Dazeley R. A survey of multi-objective sequential decision-making. Journal of Artificial Intelligence Research. 2013 Oct 18;48:67-113.
  14. Van Moffaert K, Nowé A. Multi-objective reinforcement learning using sets of pareto dominating policies. The Journal of Machine Learning Research. 2014 Jan 1;15(1):3483-512.
  15. Tessler C, Mankowitz DJ, Mannor S. Reward constrained policy optimization. arXiv preprint arXiv:1805.11074. 2018 May 28.
  16. Reiss C, Wilkes J, Hellerstein JL. Google cluster-usage traces: format+ schema. Google Inc., White Paper. 2011 Nov;1(1-14):83.
  17. Cortez E, Bonde A, Muzio A, Russinovich M, Fontoura M, Bianchini R. Resource central: Understanding and predicting workloads for improved resource management in large cloud platforms. InProceedings of the 26th Symposium on Operating Systems Principles 2017 Oct 14 (pp. 153-167).
  18. Schulman J, Wolski F, Dhariwal P, Radford A, Klimov O. Proximal policy optimization algorithms. arXiv preprint arXiv:1707.06347. 2017 Jul 20.
  19. Hochreiter S, Schmidhuber J. Long short-term memory. Neural computation. 1997 Nov 15;9(8):1735-80.
  20. Veličković P, Cucurull G, Casanova A, Romero A, Lio P, Bengio Y. Graph attention networks. arXiv preprint arXiv:1710.10903. 2017 Oct 30.
Ahead of Print Subscription Review Article
Volume 16
02
Received 05/07/2026
Accepted 07/07/2026
Published 13/07/2026
Publication Time 8 Days
1

Login


My IP

PlumX Metrics