Semi-intrusive Uncertainty Quantification for Multiscale models
share
We propose a family of semi-intrusive Uncertainty Quantification (UQ) methods for multiscale models with separated time scales. The methods are semi-intrusive in the sense that we limit the inspection of the multiscale model up to the level of the single scale systems. Our goal was to reduce the computation time spent on UQ of micro-scale models, since they usually represent the computationally most intensive part of multiscale models. Moreover, micro-scale models are usually executed at every time step of the slow dynamics, adding even more to their contribution to overall computing time. We tested the semi-intrusive algorithms on two case studies based on reaction-diffusion dynamics. Comparison of the UQ approaches shows that the proposed semi-intrusive methods can result in a significant reduction of the computational time, while still computing accurate estimates of uncertainties.
READ FULL TEXT