Abstract
Background: Smart grid systems increasingly rely on load forecasting for efficient energy management, but centralized data collection raises significant privacy concerns. Federated learning (FL) offers a promising paradigm by enabling collaborative model training without sharing raw data. Methods: This study proposes a hierarchical federated learning framework tailored for privacy-preserving smart meter data. The framework incorporates differential privacy and secure aggregation to protect individual load profiles. We evaluate the approach using a realistic dataset of residential loads, comparing performance against centralized and local baselines. Results: Experimental results demonstrate that the proposed FL framework achieves near-centralized accuracy (within 2% RMSE) while maintaining strong privacy guarantees (ε ≤ 1.0). The hierarchical structure reduces communication overhead by 40% compared to flat FL. Furthermore, we analyze the trade-off between privacy budget and model accuracy, showing that acceptable privacy can be achieved with minimal degradation. Conclusions: The integration of federated learning in smart grids enables privacy-preserving load forecasting without compromising predictive performance. The framework is scalable and adaptable to various grid configurations, supporting the transition to secure and intelligent energy systems.
Keywords
Federated Learning, Smart Grid, Load Forecasting, Privacy Preservation, Differential Privacy, Hierarchical Framework, Residential Load, Energy Management