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

Security and Attacks on Federated Energy Forecasting

Authors:
Jonas Sievers
Krupali Kumbhani
Thomas Blank
Frank Simon
Andreas Mauthe

Keywords: federated learning; poisoning attack; backdoor attack

Abstract:
Accurate energy forecasting, including load, photovoltaic generation, and prosumption prediction, is essential for the efficient operation and strategic planning of modern energy systems. Federated Learning (FL) has emerged as a promising solution for training machine learning models on decentralized data, enabling high model accuracy while maintaining data privacy. However, the decentralized nature of FL also poses security challenges, including data poisoning and backdoor attacks that compromise the integrity and reliability of forecasting models. In this study, we present a comprehensive evaluation of various data poisoning and backdoor attacks within federated energy forecasting. Our analysis explores different data distributions, varying noise scales in data poisoning attacks, and targeted manipulation of specific time intervals to assess their impact on model performance. Further, we propose robust security mechanisms, such as increased cluster sizes, local retraining, and weighted aggregation. Our results show that while our attacks can increase the Mean Absolute Error by 93-261 %, our security measures can effectively mitigate the attacks, thereby improving the security and robustness of federated energy forecasting.

Pages: 23 to 28

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