On the equivalence between functionally affine LPV state-space representations and LFT models.

Mihály Petreczky, Ziad Alkhoury, Guillaume Mercère · 2025

Read On the equivalence between functionally affine LPV state-space representations and LFT models. by Mihály Petreczky, Ziad Alkhoury, Guillaume Mercère (2025) in the Tower of Babel, a free online library of public-domain works preserved in clean, searchable text. No sign-up. The full text is served as a plain-text file at a predictable URL, where text is present: https://antarctic-labs.com/tower-of-babel/arxiv-2505-14993.txt.

Introduction

Petreczky, Alkhoury, and Mercere give in 2025 an algorithm that converts one class of varying system models into another standard form. The source systems depend on parameters through fixed functions in an affine way. Such models arise in control of systems whose behavior shifts with conditions.

The target form is the linear fractional transformation, a standard way to package uncertainty. Their conversion keeps the input output behavior and minimality of the model. The uncertainty block of the result depends in a simple linear way on the source scheduling variables. The result lets control tools built for one form serve the other.

Catalog record

CREATORMihály Petreczky, Ziad Alkhoury, Guillaume Mercère
YEAR2025
COLLECTIONPAPERS
FORMATTXT
FILE SIZE85 KB
VERSION1.0
LICENSEhttp://creativecommons.org/licenses/by/4.0/
RIGHTSCC BY
SOURCEarXiv

Source: https://arxiv.org/abs/2505-14993

Browse the full library · About the Tower of Babel

Full text

The opening of the work, from the library's hosted text:

On the equivalence between functionally affine LPV state-space representations and LFT models. Mihály Petreczky, Ziad Alkhoury, Guillaume Mercère Abstract We propose a transformation algorithm for a class of Linear Parameter-Varying (LPV) systems with functional affine dependence on parameters, where the system matrices depend affinely on nonlinear functions of the scheduling variable, into Linear Fractional Transformation (LFT) systems. The transformation preserves input-output behavior and minimality, and the uncertainity block of the resulting LFT system is linear in the scheduling variables of the LPV system. I Introduction Linear Fractional Transformation (LFT) [ 1 ] and Linear Parameter-Varying (LPV) systems [ 2 , 3 , 4 ] are crucial for handling uncertainties and nonlinearities. LFT effectively represents uncertain systems, while LPV models systems with time-varying parameters. Transforming LPV systems into LFT form, where the uncertainty block depends on the scheduling signal, is common [ 2 , 5 ] . For LPV systems with affine dependence on the scheduling signal, this transformation is straightforward, but for the general case, it becomes more complex. We propose an algorithm for transforming a class of discrete-time LPV systems with a functional affine dependence on the parameters (FALPV), where the matrices depend affinely on recognizable nonlinear functions of the scheduling variable, to LFT systems. Recognizable functions are functions which can be represented by LFT models, and they include polynomial and, after scaling, rational functions. The uncertainty block of the resulting LFT system is linear in the scheduling variables of the FALPV and has the same number of uncertainty blocks as the number of scheduling variables. The transformation preserves input-output behavior and minimality. In particular, the LFTs arising from two minimal and input-output equivalent FALPVs are isomorphic. Motivation The proposed transformation is useful, because the resulting uncertainty block being linear in the scheduling variable simplifies controller synthesis and reduces conservatism. While FALPVs can be interpreted as ALPVs after a non-linear transformation of the scheduling variable, this approach results in an LFT with a non-linear dependence on the scheduling variable and possibly more uncertainty operators than the number of scheduling variables. For instance, for LPVs with a polynomial dependence the uncertainity block would contain monomials of the scheduling variables. This makes it harder to bound the uncertainty block and increases the conservatism of the controller synthesis. Preservation of minimality and isomorphism makes controller synthesis independent of the modeling approach. This is significant as LPV models often originate from system identification [ 3 , 6 ] , and different identification methods may yield different LPV models that are approximately input-output equivalent and minimal. However, naive transformations of minimal LPV systems to LFTs can result in non-isomorphic LFTs, leading to different achievable closed-loop performances [ 7 ] . Our results imply that LFTs derived from input-output equivalent minimal LPV systems are minimal, input-output equivalent and isomorphic. This suggests that there is no advantage in identifying LFTs directly [ 8 , 9 ] versus LPV models. In this paper we concentrate on discrete-time LPV systems due to their simplicity and relevance for system identification. We conjecture that the results could be extended to the continuous-time case. Related work The idea of transforming LPV systems to LFT for control design is a standard one [ 2 , 4 , 5 ] , the structural properties of this transformation, such as the preservation of minimality and input-output equivalence were not investigated before for FALPV systems. In deriving the results we use realization theory of ALPV systems [ 10 ] , and of LFTs (viewed as multidimensional systems) [ 11 , 12 ] . For LPV systems with affine dependence, the preservation of minimality and input-output equivalence for all uncertainty blocks was investigated in [ 7 ] . The results of this paper are extension of [ 7 ] to FALPV case. II Definitions and problem formulation Notation \mathbb{N} denotes the set of all natural numbers, t denotes the discrete-time instant, i.e. , t\in\mathbb{N} as well. We denote by \mathbb{I}_{i}^{j} the set \{i,i+1,\ldots,j-1,j\} , where i,j\in\mathbb{N},j>i . The n\times n identity matrix is denoted by {\mathrm{I}}_{n} . Let O_{k,l} be the k\times l zero matrix. When its dimension is clear from the context, the identity matrix is denoted by \mathrm{I} , and the zero matrix by 0 . For a set X denote by X^{\mathbb{N}} the set of functions from \mathbb{N} to X . Denote by l_{2}(\mathbb{R}^{n}) the space square summable sequences from \mathbb{R}^{n} , and by \mathcal{L}(l_{2}(\mathbb{R}^{n})) the space of all bounded linear operators on l_{2}(\mathbb{R}^{n}) . If A,B linear operators, then AB denotes their composition. If s\in(\mathbb{R}^{m})^{\mathbb{N}} , then the i th component s_{i}\in(\mathbb{R})^{\mathbb{N}} , i\in\mathbb{I}_{1}^{d} of s is such that s(t)=(s_{1}(t),\ldots,s_{m}(t))^{T} , t\in\mathbb{N} . If A is an n\times m matrix, (i,j) th entry A_{i,j} of which is a linear operator from a subspace Z\subseteq(\mathbb{R})^{\mathbb{N}} to (\mathbb{R})^{\mathbb{N}} , then A is an operator from Z^{m} to (\mathbb{R}^{p})^{\mathbb{N}} , where Z^{m} is the subspace of all s\in(\mathbb{R}^{m})^{\mathbb{N}} such that the components s_{i} of s belong to Z , i\in\mathbb{I}_{1}^{m} , and the l th component of As is \sum_{j=1}^{m}A_{l,j}s_{j} . Thus, \delta\mathrm{I}_{m} denotes the m\times m diagonal matrix of operators \mathrm{diag}(\delta,\ldots,\delta) . For a finite set \mathcal{X} let \mathcal{X}^{*} be the set of all finite sequences of elements of \mathcal{X} . An element w\in\mathcal{X}^{*} of length |w|=n\in\mathbb{N} , is a sequence of the form x_{1}x_{2}\dots x_{n} , where x_{i}\in\mathcal{X},\forall i=1,2,...,n . Denote \epsilon the empty sequence where |\epsilon|=0 , and denote by vw the concatenation of two sequence w,v\in\mathcal{X}^{*} . Thus, (\mathbb{I}_{i}^{j})^{*} denotes the set of all finite sequences of \mathbb{I}_{i}^{j} . II-A Functional-Affine LPV (FALPV) Models Let \mathbb{P}\subseteq\mathbb{R}^{{n}_{\mathrm{p}}} be the set of scheduling parameters, and let us consider a known function \psi:\mathbb{P}\rightarrow\mathbb{R}^{{n}_{\mathrm{\Psi}}} . Let us denote by \psi_{i} the i th component of \psi , i.e., \psi(\mathbf{p})=[\psi_{1}(\mathbf{p}),\ldots,\psi_{{n}_{\mathrm{\Psi}}}(\mathbf{p})]^{\top} , \forall\mathbf{p}\in\mathbb{P} . A Functional Affine Linear Parameter-Varying (FALPV) model \Sigma w.r.t \psi is a system \Sigma\left\{\begin{array}[]{lcl}x(t+1)&=&\mathscr{A}(p(t))x(t)+\mathscr{B}(p(t))u(t),\\ y(t)&=&\mathscr{C}(p(t))x(t)+\mathscr{D}(p(t))u(t).\end{array}\right. (1) where x(t)\in\mathbb{X}=\mathbb{R}^{n_{\mathrm{x}}} is the state at time t , y(t)\in\mathbb{Y}=\mathbb{R}^{n_{\mathrm{y}}} is the output at time t , u(t)\in\mathbb{U}=\mathbb{R}^{n_{\mathrm{u}}} represents the input at time t , while p(t)\in\mathbb{P}\subseteq\mathbb{R}^{n_{\mathrm{p}}} is the scheduling signal at time t . The matrix valued functions \mathscr{A} , \mathscr{B} , \mathscr{C} and \mathscr{D} are assumed to be affine functions of \psi(p(t)) . \begin{split}&\begin{bmatrix}\mathscr{A}(p(t))\!\!\!&\!\!\!\mathscr{B}(p(t))\\ \mathscr{C}(p(t))\!\!\!&\!\!\!\mathscr{D}(p(t))\end{bmatrix}\!\!=\!\!\begin{bmatrix}\mathscr{A}_{0}\!\!&\!\!\mathscr{B}_{0}\\ \mathscr{C}_{0}\!\!&\!\!\mathscr{D}_{0}\end{bmatrix}+\sum\limits_{i=1}^{{\color[rgb]{1,0,0}{n}_{\mathrm{\Psi}}}}\begin{bmatrix}\mathscr{A}_{i}\!\!&\!\!\mathscr{B}_{i}\\ \mathscr{C}_{i}\!\!&\!\!\mathscr{D}_{i}\end{bmatrix}\psi_{i}(p(t))\\ \end{split} for constant matrices \mathscr{A}_{i}\in\mathbb{R}^{n_{\mathrm{x}}\times n_{\mathrm{x}}} , \mathscr{B}_{i}\in\mathbb{R}^{n_{\mathrm{x}}\times n_{\mathrm{u}}} , \mathscr{C}_{i}\in\mathbb{R}^{n_{\mathrm{y}}\times n_{\mathrm{x}}} , \mathscr{D}_{i}\in\mathbb{R}^{n_{\mathrm{y}}\times n_{\mathrm{u}}} , i\in\mathbb{I}_{1}^{{n}_{\mathrm{\Psi}}} , Without loss of generality we assume that \mathbb{P}=[-1,1]^{{n}_{\mathrm{p}}} , and the functions 1,\psi_{1},\ldots,\psi_{{n}_{\mathrm{\Psi}}} are linearly independent , i.e. no \psi_{i} can be expressed as a linear combination of \{1\}\cup\{\psi_{j}\}_{j=1,i\neq j}^{{n}_{\mathrm{\Psi}}} , i\in\mathbb{I}_{1}^{{n}_{\mathrm{\Psi}}} . If the functions \{1\}\cup\{\psi_{i}\}_{i=1}^{{n}_{\mathrm{p}}} are not linearly independent, we can choose a linear basis from them and express the system matrices using only this linear basis. The scheduling variables are usually assumed to be bounded, and we can always scale them to [-1,1]^{{n}_{\mathrm{p}}} . Note that \psi is not necessarily a polynomial/affine function. In principle, for any LPV system when can choose a function \psi (formed by the entries of the matrices of the LPV) for which we can rewrite that LPV as an FALPV. The dimension of \Sigma , denoted by \dim(\Sigma) is the dimension {n}_{\mathrm{x}} of its state-space. By a solution of \Sigma we mean a tuple of trajectories (x,y,u,p)\in(\mathcal{X},\mathcal{Y},\mathcal{U},\mathcal{P}) satisfying ( 1 ) for

Read the complete text