Loading article...
Loading article...
Stream/Area:
-
Subject:
-
Paper ID:
IJSREM68689
Volume:
10 |
11 |
Nov 2026
Pages:
1-14
Publication Frequency:
Monthly (12 issues per year)
License:International Journal of Scientific Research in Engineering and Management follows the Creative Commons Attribution 4.0 (CC BY 4.0) License, allowing free use and distribution with proper attribution.
Department Of Computer Science & Engineering
SHUATS
Department Of Computer Science & Engineering
SHUATS
Department Of Computer Science & Engineering
SHUATS
Federated learning is a distributed learning framework for forecasting residential electricity consumption without centralizing raw client data. Thirty complete households were selected as simulated FL clients from the real-world dataset, after quality controlled preprocessing and hourly aggregation of the data. Next-hour electricity consumption is predicted using a GRU model with a 24-hour lookback and a chronological train–validation–test protocol. Centralized GRU, FedAvg, a drift-aware reliability-weighted prototype, and diagnostic conflict-aware aggregation are evaluated. FedAvg achieves a test MAE of 0.205108 kWh/hour, RMSE of 0.377878, and R² of 0.606950, while the drift-aware prototype achieves MAE of 0.208843, RMSE of 0.380713, and R² of 0.601030. Client-level analysis shows there is no significant predictive power of the static measure of heterogeneity on the forecasting MAE but update-alignment and conflict measures are associated with post-update validation error. Conflict-aware weighting does not give evidence of improvement and actually causes a 0.000170 increase in MAE compared to FedAvg.
KEYWORDS
This study presents a controlled empirical evaluation of federated residential energy forecasting using heterogeneous real-world smart-meter data. The results show that FL can achieve competitive results for one-step ahead forecasting without data pooling in raw data, but the evaluated simple adaptive aggregation mechanisms do not improve on FedAvg. FedAvg's test MAE is 0.205108, test RMSE is 0.377878, and test R² is 0.606950, which beats the tested drift-aware α=1 prototype. Measured heterogeneity does not significantly predict MAE at the client level and drift magnitude is not a good predictor of the usefulness of the updates. Alignment and conflict statistics are found to be more related to post-update validation behaviour, while the direct conflict-aware weighting does not benefit the MAE for validation. The central conclusion is therefore not that adaptive aggregation is impossible, but that single-signal adaptation is insufficiently reliable for this setting. The negative result gives a concrete methodological boundary, and inspires the future development of aggregation mechanisms that include historical, temporal, personalized and validation-based evidence.
[1] UK Power Networks, “SmartMeter Energy Consumption Data in London Households,” London Datastore, Low Carbon London project, Nov. 2011–Feb. 2014. Available: https://data.london.gov.uk/dataset/smartmeter-energy-consumption-data-in-london-households-vqm0d [2] P. Kairouz et al., “Advances and open problems in federated learning,” Foundations and Trends in Machine Learning, vol. 14, nos. 1–2, pp. 1–210, 2021, doi: 10.1561/2200000083. [3] B. McMahan, E. Moore, D. Ramage, S. Hampson, and B. A. y Arcas, “Communication-efficient learning of deep networks from decentralized data,” in Proc. 20th Int. Conf. Artificial Intelligence and Statistics (AISTATS), PMLR, vol. 54, pp. 1273–1282, 2017. [4] T. Li, A. K. Sahu, M. Zaheer, M. Sanjabi, A. Talwalkar, and V. Smith, “Federated optimization in heterogeneous networks,” Proc. Machine Learning and Systems, vol. 2, 2020. [5] S. P. Karimireddy, S. Kale, M. Mohri, S. Reddi, S. U. Stich, and A. T. Suresh, “SCAFFOLD: Stochastic controlled averaging for federated learning,” in Proc. 37th Int. Conf. Machine Learning (ICML), PMLR, vol. 119, pp. 5132–5143, 2020. [6] A. Tibermacine, I. Naidji, I. E. Tibermacine, S. Sahraoui, M. Zouai, A. Rabehi, and M. H. Habib, “Privacy-preserving load forecasting in smart grids using federated learning: A comparative analysis of aggregation strategies,” Frontiers in Energy Research, vol. 14, 2026, Art. no. 1854654, doi: 10.3389/fenrg.2026.1854654. [7] “Privacy-Preserving Electrical Load Forecasting Through Federated Learning,” in Proc. 2025 IEEE Int. Conf. on Smart Power, Energy, Renewables, and Transportation (SPERT), 2025, doi: 10.1109/SPERT67079.2025.11469461. [8] P. Das, S. Pandey, C. Pandey, and D. Kumar et al., “Heterogeneity-aware personalised federated learning for household energy forecasting on multi-source real-world smart meter data,” Scientific Reports, vol. 16, Art. no. 24957, 2026, doi: 10.1038/s41598-026-53020-6. [9] J. Wang, D. Xu, L. Xing, T. Chen, Y. Liu, M. Yan, and T. Yang, “Personalized federated learning with mixture of experts and conformal prediction for household energy forecasting,” Expert Systems with Applications, vol. 300, Art. no. 130417, 2026, doi: 10.1016/j.eswa.2025.130417. [10] Z. Zuo, H. Ye, J. Li, and Y. Ge, “A robust and heterogeneity-aware federated learning framework with knowledge distillation for cross-regional load forecasting,” IEEE Transactions on Smart Grid, early access, 2026, doi: 10.1109/TSG.2026.3660096. [11] S. H. Tindemans, “Low Carbon London smart meter data (refactored),” 4TU.ResearchData, dataset, 2023, doi: 10.4121/FBBE775B-48D8-469F-A39B-B64488BFD6FD.
Publication Date
11/5/2026
Peer Review Type
Double-Blind Peer Review
Review Rounds: 2
Documents
Article Views
0
Total views since publication
PDF Downloads
5
Total downloads since publication
Geographic Reach
5
Digital Preservation: This article is permanently archived on Zenodo and complies with all data preservation standards across major indexing platforms.
Plagiarism Screening
This manuscript has been checked for plagiarism using TurnitinPlagiarismChecker software.
Conflict of Interest
The authors declare no conflict of interest related to this research work.
Funding Information
No funding information available for this research.
Related research in the same subject area
Loading similar articles...