Foundation Twins: A New Generation of Power Systems Digital Twins using Foundation AI Models

Researcher Pedro P. Vergara has released a position paper on arXiv (2605.05952) titled 'Foundation Twins: A New Generation of Power Systems Digital Twins using Foundation AI Models.' The paper tackles the longstanding challenge of building digital twins (DTs) for power systems—complex networks that operate across multiple timescales, from milliseconds to years, and span diverse geographic regions. Vergara argues that despite years of research, current DTs remain theoretical and fail to provide the real-time, multi-scale decision-making support that grid operators urgently need as renewables and distributed energy resources grow.

The proposed solution, 'Foundation Twins,' leverages the generalization capabilities of large foundation AI models (similar to GPT or Claude) combined with reinforcement learning (RL) architectures. Foundation models handle broad system understanding and simulation across various conditions, while RL enables adaptive, optimal decision-making in real time. This hybrid approach could allow utilities to simulate thousands of scenarios, predict failures, optimize energy flow, and integrate intermittent renewables more effectively. While the paper is a vision statement rather than a technical implementation, it lays out a concrete roadmap for closing the gap between AI advances and practical power grid tools. If realized, Foundation Twins could transform how we manage one of the most critical infrastructures on the planet.