A new preprint introduces FedGuard-DC, a privacy-preserving framework for forecasting data-center loads and detecting cyber-attacks in transmission systems. The authors argue that the rapid growth of large data-center loads is straining power-system visibility, privacy, and cyber-physical security, and that operators need accurate short-term information about these fast-varying loads.
The framework uses federated learning, which allows models to be trained across distributed data centers without centralizing raw data. This is intended to protect sensitive operational information while still giving system operators the load forecasts they need. The same framework also incorporates cyber-attack detection, addressing the growing threat surface created by interconnected power and computing infrastructure.
Because the source is a single arXiv abstract, the article does not cover experimental results or implementation details. The contribution, as described, is the combination of federated forecasting with attack detection specifically tailored to data-center loads in transmission systems.