Reduced model reconstruction method for stable positive network systems

TL;DR

This paper introduces a new method for reconstructing stable positive network systems that preserves their structure and improves model accuracy using an efficient projected gradient approach with proven convergence.

Contribution

It proposes a cyclic projected gradient algorithm for $H^2$ optimal model reduction of positive systems, ensuring stability, structure preservation, and convergence without line search.

Findings

The method improves reduced models in numerical experiments.

It guarantees global convergence to a stationary point.

Applicable to large-scale systems efficiently.

Abstract

We consider a reconstruction problem of a reduced stable positive network system with the preservation of the original interconnection structure based on an $H^2$ optimal model reduction problem with constraints. To this end, we define an important set using the Perron--Frobenius theory of nonnegative matrices such that all elements of the set are stable and Metzler. Using the projection onto the set, we propose a cyclic projected gradient method to produce a better reduced model than an initial reduced model in the sense of the $H^2$ norm. In the method, we use Lipschitz constants of the gradients of our objective function to define the step sizes without a line search method whose computational complexity is large. Moreover, the existence of the Lipschitz constants guarantees the global convergence of our proposed algorithm to a stationary point. The numerical experiments demonstrate that the proposed algorithm improves a given reduced model, and can be used for large-scale systems.