Data-driven operator inference for nonintrusive projection-based model reduction

What is it about?

This paper develops a method to create a reduced order model directly from simulation data, without needing to compute system matrices or their projections. The main steps are: (1) run the high-fidelity model to generate snapshots (solutions at different times), (2) compute the POD basis using an SVD of the snapshots, (3) compute the POD coefficients by projecting the snapshots onto the POD basis, (4) solve a least-squares optimization problem to fit a reduced order model to the POD coefficient data.

Featured Image

Why is it important?

Creating a POD reduced order model using the standard approach is intrusive because it requires you to access the internal matrices and flux functions of your code. Our non-intrusive approach works directly with solutions that are easily output and manipulated.

Perspectives