Distributed Machine Learning for Energy Efficiency

In a global context marked by increasing digitalization and the proliferation of smart devices, efficiently managing continuously generated data has become a key challenge. In this scenario, Artificial Intelligence (AI)—particularly Machine Learning (ML)—is consolidating as a fundamental tool for extracting value from decentralized information sources. However, as millions of smart meters, IoT sensors, and distributed energy systems produce real-time data at the edge, the challenge extends beyond data processing to include sustainability, privacy preservation, and resource-aware computation. Traditional centralized ML architectures—reliant on massive data transfers and cloud-based processing—are proving increasingly impractical due to high energy consumption, communication overhead, privacy concerns, and scalability limitations.
Distributed Machine Learning (DML) leverages local computation at the edge to address these challenges, enabling collaborative model training without centralized data storage. Within this paradigm, several advanced techniques have gained relevance:

• Federated Learning (FL) allows edge devices to train models collaboratively while keeping data local, reducing privacy risks and energy usage.
• Federated Transfer Learning (FTL), which combines Federated Learning and Transfer Learning, enable the reuse of knowledge from related tasks or domains, improving convergence speed and reducing computational effort.
• Clustering techniques help group similar energy usage patterns or device characteristics, allowing for personalized and efficient model deployment.
• Additional techniques such as model compression, adaptive aggregation, energy aware scheduling and attention mechanisms contribute to optimizing performance under resource constraints.

Within this framework, two complementary lines of research address the challenges of applying distributed machine learning to energy efficiency. One line of research focuses on improving DML's energy efficiency and scalability in resource-constrained device environments. It addresses challenges such as communication overhead, data heterogeneity, and computational constraints by using FTL, promoting more sustainable, faster, and privacy-concerned collaborative training. In parallel, the second line of research explores how distributed and federated ML can enhance energy intelligence in smart metering systems and energy consumption forecasting. Rather than adopting a one-size-fits-all model, this approach leverages clustering algorithms to identify groups of users or devices with similar consumption patterns. This strategy contributes to developing adaptive and sustainable ML solutions that respond to the real-world variability of energy usage behaviors.

References

[1] M. A. Husnoo, A. Anwar, N. Hosseinzadeh, S. N. Islam, A. N. Mahmood, and R. Doss, “A Secure Federated Learning Framework for Residential Short-Term Load Forecasting,” IEEE Transactions on Smart Grid, vol. 15, no. 2, pp. 2044–2055, Mar. 2024. [Online]. Available: https://ieeexplore.ieee.org/abstract/document/10173657

[2] A. Dogra, A. Anand, and J. Bedi, “Consumers profiling based federated learning approach for energy load forecasting,” Sustainable Cities and Society, vol. 98, p. 104815, Nov. 2023. [Online]. Available: https://www.sciencedirect.com/science/article/pii/S2210670723004262

[3] H. U. Manzoor, S. Hussain, D. Flynn, and A. Zoha, “Centralised vs. decentralised federated load forecasting in smart buildings: Who holds the key to adversarial attack robustness?” Energy and Buildings, vol. 324, p. 114871, Dec. 2024. [Online]. Available: https://www.sciencedirect.com/science/article/pii/S0378778824009873

[4] Y. L. Tun, K. Thar, C. M. Thwal, and C. S. Hong, “Federated Learning based Energy Demand Prediction with Clustered Aggregation,” in 2021 IEEE International Conference on Big Data and Smart Computing (BigComp), Jan. 2021, pp. 164–167, iSSN: 2375-9356. [Online]. Available: https://ieeexplore.ieee.org/document/9373194

[5] M. Savi and F. Olivadese, “Short-Term Energy Consumption Forecasting at the Edge: A Federated Learning Approach,” IEEE Access, vol. 9, pp. 95 949–95 969, 2021. [Online]. Available: https://ieeexplore.ieee.org/document/9469923

[6] H. B. McMahan, E. Moore, D. Ramage, S. Hampson, and B. A. y. Ar-cas, “Communication-efficient learning of deep networks from decentralized data,” in Proceedings of the 20th International Conference on Artificial Intelligence and Statistics (AISTATS). PMLR, 2017, pp. 1273–1282.

[7] A. Yousefpour, S. Guo, A. Shenoy, S. Ghosh, P. Stock, K. Maeng, S.-W. Kruger, M. Rabbat, C.-J. Wu, and I. Mironov, “Green federated learning,” arXiv preprint arXiv:2303.14604, 2023.

[8] S. Saha and T. Ahmad, “Federated transfer learning: Concept and applications,” Intelligenza Artificiale, vol. 15, no. 1, pp. 35–44, 2021