�鶹��Ѱ�����

Juan M Restrepo, Oak Ridge National Laboratory, Tennessee

Enhancing Resilience in Complex Systems with Digital Twins and Uncertainty Quantification

In this talk, we present a��methodology that leverages digital twins and uncertainty quantification to enhance resilience in complex systems. Given a system that has experienced a detrimental perturbation, our��method produces an adaptation strategy aimed at preventing cascading failures. This strategy consists of a sequence of recommended parameter adjustments at discrete decision times, with the goal of steering the system toward a target probability distribution. Between decision times, a digital twin simulates the system state and its uncertainty. Immediately after each decision time, data assimilation is applied to define a prior on the system state, and realizations of the digital twin are used to forecast the distribution of key observables at the next decision time. These forecasts then inform the selection of a “steering drift” that probabilistically guides the system closer to the target. We demonstrate the approach through operational policy design for resilient power grids, where a downed power line can trigger a cascade toward a blackout. Our��method generates a sequential assimilation–modeling–adaptation strategy that system operators can use to restore the grid to near-normal operating conditions. Our��method is novel avoids the computation pitfalls of approaches via control theory, is uncertainty-aware, and incorporates continuous, high-fidelity simulation of the grid’s evolving state, supported by up-to-date data.