Adaptive Privacy Preserving Federated Learning and Robust Large Language Model for Electric Grid Cybersecurity

Junbo Zhao – Dartmouth College

This talk will first show an adaptive privacy preserving federated learning approach for detecting cyberattacks on virtual power plants. However, this approach may still be vulnerable to adversarial attacks that are difficult to detect. We then develop a robust Large Language Model that embeds electrical grid time series characteristics as well as graph information for sophisticated attack detection and localization.