From data to reduced-order models via moment matching

TL;DR

This paper introduces a data-driven method for model reduction that computes moments directly from input-output data without system identification, enabling efficient and stable reduced models via rational interpolation.

Contribution

It develops a framework for moment computation from data using informativity concepts, allowing model reduction without explicit high-order system identification.

Findings

Enables moment matching directly from data

Provides conditions for shared moments among all systems fitting data

Demonstrates the method on an electrical circuit example

Abstract

A new method for data-driven interpolatory model reduction is presented in this paper. Using the so-called data informativity perspective, we define a framework that enables the computation of moments at given (possibly complex) interpolation points based on time-domain input-output data only, without explicitly identifying the high-order system. Instead, by characterizing the set of all systems explaining the data, necessary and sufficient conditions are provided under which all systems in this set share the same moment at a given interpolation point. Moreover, these conditions allow for explicitly computing these moments. Reduced-order models are then derived by employing a variation of the classical rational interpolation method. The condition to enforce moment matching model reduction with prescribed poles is also discussed as a means to obtain stable reduced-order models. An example of an electrical circuit illustrates this framework.