Home // ENERGY 2025, The Fifteenth International Conference on Smart Grids, Green Communications and IT Energy-aware Technologies // View article

Federated Learning for Distributed Load Forecasting: Addressing Data Imbalance in Smart Grids

Authors:
Alexander Wallis
Sascha Hauke
Hannah Jörg
Konstantin Ziegler

Keywords: Short-Term Load Forecasting; Federated Learning; Smart Grid; Data Privacy; Distributed Data

Abstract:
The integration of renewable energy resources trans- forms traditional energy systems, introducing prosumers – entities that both produce and consume energy – as key participants in modern Smart Grids. Effective load forecasting is mandatory for optimizing energy resources and grid stability. Federated Learning (FL) has emerged as a promising approach for distributed training of Machine Learning (ML)-based forecasting models. This enables collaborative model optimization across multiple prosumers while preserving data privacy. However, the impact of unbalanced data sets across participants remains a critical challenge in terms of potentially effecting learning convergence and forecast accuracy. In this work, we define and implement a FL system based on real-world electricity consumption data from a variety of prosumers. Experimental results demonstrate the trade-off between centralized and federated learning approaches, providing insights into addressing data heterogeneity in FL systems. These insights highlight the potential of FL to support the evolution of distributed energy systems while ensuring data-privacy and scalability. Future research directions include other strategies to migrate the effect of data imbalances and further improve the efficiency of federated optimization for dynamic energy systems.

Pages: 53 to 58

Copyright: Copyright (c) IARIA, 2025

Publication date: March 9, 2025

Published in: conference

ISSN: 2308-412X

ISBN: 978-1-68558-242-5

Location: Lisbon, Portugal

Dates: from March 9, 2025 to March 13, 2025