Digital Twins with Brains: Generative AI Meets Physics in Energy Infrastructure

Operating and planning modern energy infrastructures—electric power, natural gas, transmission and distribution networks—requires navigating complex physical flows under uncertainty, incomplete information, and increasingly volatile external conditions. At the same time, recent advances in Generative AI offer new ways to reason, sample, and make decisions in high-dimensional environments.

In this talk, I will introduce the sampling-decisions / decision-flow framework, which adapts core principles of state-of-the-art generative models of AI to energy-system challenges. The approach blends physics-based modeling with data-driven sampling, enabling digital twins that are not merely predictive but decision-ready.

I will review the methodology and present several illustrative case studies demonstrating how generative-model–based sampling can support operators in real time. A key example is the emergency response to shortfalls, such as the sudden loss of a major energy supply: the framework allows the operator to sample and rank feasible emergency unit-commitment actions, producing a dashboard of actionable strategies consistent with the underlying physics and uncertainty.

Although the examples are specific, the principles are general and scalable—offering a promising pathway for integrating Generative AI into the next generation of resilient, adaptive energy-infrastructure operations.