Data-driven quadratic modeling in the Loewner framework from input-output time-domain measurements

arXiv2211.10635MaRDI QIDQ6507100FDOQ6507100

Authors: Dimitrios Karachalios, I. V. Gosea, L. Gkimisis, A. C. Antoulas 

Abstract: In this work, we present a non-intrusive, i.e., purely data-driven method that uses the Loewner framework (LF) along with nonlinear optimization techniques to identify or reduce quadratic control systems from input-output (i/o) time-domain measurements. At the heart of this method are optimization schemes that enforce interpolation of the symmetric generalized frequency response functions (GFRFs) as derived in the Volterra series framework. We consider harmonic input excitations to infer such measurements. After reaching the steady-state profile, the symmetric GFRFs can be measured from the Fourier spectrum (phase and amplitude). By properly using these measurements, we can identify low-order nonlinear state-space models with non-trivial equilibrium state points in the quadratic form, such as the Lorenz attractor. In particular, for the multi-point equilibrium case, where measurements describe some local bifurcated models to different coordinates, we achieve global model identification after solving an operator alignment problem based on a constrained quadratic matrix equation. We test the new method for a more demanding system in terms of state dimension, i.e., the viscous Burgers' equation with Robin boundary conditions. The complexity reduction and approximation accuracy are tested. Future directions and challenges conclude this work.

Mathematics Subject Classification ID

Nonlinear systems in control theory (93C10) Control/observation systems governed by ordinary differential equations (93C15) Realizations from input-output data (93B15) Numerical methods for matrix equations (65F45)

This page was built for publication: Data-driven quadratic modeling in the Loewner framework from input-output time-domain measurements

Report a bug (only for logged in users!)Click here to report a bug for this page (MaRDI item Q6507100)