arXiv: FedGuard-DC: Privacy-Preserving Federated Load Forecasting and Cyber-Attack Detection for Data-Center Loads in Transmission Systems

This publication introduces FedGuard-DC, a new framework designed to enhance cybersecurity and operational privacy for data centers connected to transmission grids. It combines federated learning, which allows multiple entities to train AI models without sharing raw data, with a detection mechanism for cyber-attacks targeting load forecasting systems. The paper demonstrates a method to predict electricity demand and identify malicious interference while keeping sensitive operational data localized, addressing a growing vulnerability in critical energy infrastructure.

The primary audience is electricity transmission system operators, large-scale data center operators, and energy utilities that participate in demand-response programs or grid-balancing services. Any organization that shares load data with grid operators or relies on third-party forecasting tools could be affected, as the framework offers a way to meet data privacy obligations under regulations like GDPR or NIS2 while still contributing to grid stability. It is particularly relevant for entities classified as essential or important under the EU’s critical infrastructure rules.

Compliance teams should monitor this research as an indicator of emerging best practices for privacy-preserving AI in the energy sector. They should assess whether their current data-sharing agreements with grid operators expose them to unnecessary cyber risk or regulatory penalties. Next steps include reviewing internal data governance policies for load forecasting, evaluating whether federated learning could reduce data minimization burdens, and engaging with technical teams to pilot such frameworks before regulators mandate them. No immediate action is required, but proactive alignment with this approach could future-proof compliance strategies.