Grid-air: secure and trustworthy artificial intelligence for smart grid applications

The Department of Energy Office of Science awarded the University of Maine System $2.4 million on January 15, 2026, for GRID-AIR: Secure and Trustworthy Artificial Intelligence for Smart Grid Applications, a project grant under the Office of Science Financial Assistance Program (CFDA 81.049).

The award supports development of secure, trustworthy AI systems for smart grid operations. Work is performed in Orono, Maine, with a period of performance through January 14, 2027.

The University of Maine coordinates three sub-recipient institutions. The University of Hawaii at Manoa (awarded August 20, 2026) develops secure continual learning frameworks for anomaly detection and renewable energy forecasting using adaptive transformer-based AI models, incorporating explainability through data visualization and examining economic and equity impacts. The University of North Dakota (awarded August 12, 2026) creates a self-healing AI framework utilizing Federated Edge Learning and neural network techniques including LSTM Autoencoders and Transformers for real-time detection and mitigation of cyber disruptions in grid systems, with an AI-enabled testbed integrated with AWS for validation across three fail-safe scenarios addressing data poisoning, model poisoning, and hybrid threats. The University of Puerto Rico Research and Development Center, Mayaguez Campus (awarded July 21, 2026) develops a closed-loop digital twin testbed for a distribution-level microgrid that integrates AI and machine learning for parameter estimation and power flow prediction under cybersecurity and communication scenarios. A coordinated "Mega Digital Twin" across all four institutions provides augmented capabilities for AI-based solutions, with the town of Adjuntas, Puerto Rico—which operates a unique operational microgrid—serving as the application case.