A Koopman Operator Approach to Data-Driven Control of Semilinear Parabolic Systems

Joachim Deutscher, Tarik Enderes · 2025

Read A Koopman Operator Approach to Data-Driven Control of Semilinear Parabolic Systems by Joachim Deutscher, Tarik Enderes (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-2504-11959.txt.

Introduction

Joachim Deutscher and Tarik Enderes set out a way to steer a class of heat-like systems from data in this two thousand twenty-five paper. The systems are semilinear parabolic partial differential equations with control at the edge. Their true rules are not known. The team lifts the curved dynamics with a small set of Koopman operator eigenfunctionals. That gives a model with a set size that standard design tools can make stable.

A bilinear form of the lifted model links how eigenfunctionals are set to feedback linearization. The paper checks that the closed loop stays stable in an exponential way. That holds even with errors from the data-based sums for the eigenfunctionals. The controller comes from generalized eDMD used on state data. A test case is an unstable reaction-diffusion system that would blow up in finite time. It shows the method at work.

Catalog record

CREATORJoachim Deutscher, Tarik Enderes
YEAR2025
COLLECTIONPAPERS
FORMATTXT
FILE SIZE62 KB
VERSION1.0
LICENSEhttp://creativecommons.org/licenses/by/4.0/
RIGHTSCC BY
SOURCEarXiv

Source: https://arxiv.org/abs/2504-11959

Browse the full library · About the Tower of Babel

Full text

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

A Koopman Operator Approach to Data-Driven Control of Semilinear Parabolic Systems Joachim Deutscher, Tarik Enderes Abstract This paper is concerned with the data-driven stabilization of unknown boundary controlled semilinear parabolic systems. The nonlinear dynamics of the system are lifted using a finite number of eigenfunctionals of the Koopman operator related to the autonomous semilinear PDE. This results in a novel data-driven finite-dimensional model of the lifted dynamics, which is amenable to apply design procedures for finite-dimensional systems to stabilize the semilinear parabolic system. In order to facilitate this, a bilinearization of the lifted dynamics is considered and feedback linearization is applied for the data-driven stabilization of the semilinear parabolic PDE. This reveals a novel connection between the assignment of eigenfunctionals to the closed-loop Koopman operator and feedback linearization. By making use of a modal representation, exponential stability of the closed-loop system in the presence of errors resulting from the data-driven computation of eigenfunctionals and the bilinearization is verified. The data-driven controller directly follows from applying generalized eDMD to state data available for the semilinear parabolic PDE. An example of an unstable semilinear reaction-diffusion system with finite-time blow up demonstrates the novel data-driven stabilization approach. I Introduction I- 1 Background and Motivation The stabilization of boundary controlled semilinear parabolic PDE systems arises in many applications. For example, thermal chemical-reacting and plasma systems or crystal growth processes to name but a few. If the controller has to be designed in a late-lumping fashion, i.e., directly taking the nonlinear PDE system for the design into account, several methods are available in the literature. This includes nonlinear backstepping designs (see [ 38 , 39 ] ), optimal control (see [ 21 ] ) or discontinuous control (see [ 31 ] ). Another wildely applied approach is to determine an approximation of the semilinear parabolic system and then design a controller for the resulting finite-dimensional model, i.e., applying early lumping . Prominent methods are the Galerkin approach (see [ 4 ] for semilinear parabolic systems with distributed input) and the proper orthogonal decomposition (POD) (see [ 2 , Ch. 12] ). All these methods, however, require the knowledge of the PDE for the design. Recently, data-driven control methods have become more and more important due to the increasing availability of data. They allow to extend the controller design to complex system classes, where the first principle modeling is very tedious (see [ 2 ] ). In addition, no time consuming system modeling and parameter identification are necessary. In this respect, the Koopman operator approach has emerged as a ubiquitous tool to deal with the data-driven analysis and synthesis of nonlinear dynamical systems (see [ 27 ] ). This new viewpoint lifts the nonlinear autonomous system dynamics on an infinite-dimensional space of observables, where the evolution of the latter can be described by the linear Koopman operator globally. Hence, linear system and operator theory can be applied for the analysis of the underlying nonlinear dynamics. In addition, systematic methods exist for the data-driven approximation of the Koopman operator with guaranteed finite-data bounds (see [ 34 ] ) in form of the linear regression-based extended dynamic mode decomposition (eDMD) (see [ 22 , 2 ] ). Currently, the focus of the Koopman approach for data-driven control has been on finite-dimensional nonlinear systems (see [ 27 ] for examples), while only few results are available for nonlinear PDE systems. In particular, the contributions [ 33 , 1 ] apply the Koopman theory to nonlinear PDE systems in order to solve data-driven optimal control problems. Therein, however, the approximation of the Koopman operator is based on a spatial discretization of the underlying PDE dynamics. In [ 28 ] it has been shown, that this is not necessary by introducing functionals as observables. This allows a direct spectral analysis of the Koopman operator and thus of the nonlinear PDE system. First applications of these new ideas include the solution of identification problems in [ 26 ] and the Koopman modal control of linear parabolic systems in [ 7 ] . So far, however, the results in [ 28 ] have not been used for the data-driven control of nonlinear PDE systems, which is the obvious application of the Koopman theory. I- 2 Contribution This paper considers the data-driven state feedback control for unknown semilinear parabolic systems, for which only state data is available. For this, the underlying nonlinear dynamics are lifted by using a finite number of open-loop eigenfunctionals of the Koopman operator for the semilinear parabolic system. As a result, a Koopman modal representation in form of a finite-dimensional nonlinear system with linear eigendynamics and a state-dependent input vector field is obtained, which globally describes the system dynamics. Hence, the dominant dynamics are captured with a lower order model, when compared to a Galerkin-type approximation using the eigenfunctions related only to the linear part of the semilinear parabolic PDE (see [ 4 , Ch. 4] ). The new Koopman lifting extends previous results in [ 12 , 16 , 36 ] for finite-dimensional nonlinear systems to infinite dimensions. For finite-dimensional systems [ 16 ] directly uses the Koopman model with state dependent input vector field for the design, while the results in [ 12 ] rely on a subsequent bilinearization for the controller and observer design. In the paper the optimal bilinearization method in [ 12 ] is generalized to the semilinear parabolic systems in question. The new data-driven bilinear system description provides a firm basis to apply various design methods for finite-dimensional nonlinear systems to stabilize unknown semilinear parabolic PDE systems. In the paper, feedback linearization is utilized to facilitate the design of a stabilizing controller assigning a finite number of closed-loop eigenvalues. In particular, the single-step approach proposed in [ 17 ] is applied, which circumvents the restrictive involutivity conditions necessary to map the system into nonlinear controller form (see [ 15 , Th. 4.2.3] ). Hence, feedback linearization under weak conditions and thus for a large class of semilinear parabolic PDE systems is possible. More importantly, it is shown that if the Koopman modal representation can be reformulated with an input vector field depending only on the open-loop eigenfunctionals without error, then this feedback linearizing controller also assigns a set of desired eigenfunctionals to the closed-loop system. This extends the results in [ 7 ] for linear parabolic systems to the semilinear case. This pleasing result directly exploits the inherent property of the Koopman operator that the system dynamics have a nice linear representation in the coordinates of the eigenfunctionals and relates it to feedback linearization. It is demonstrated that this design method can be reformulated as a nonlinear change of coordinates using modal coordinates w.r.t. the eigenfunctions of the linear part in the semilinear parabolic PDE. This allows to systematically verify closed-loop stability also in the presence of the errors appearing from the data-driven computation of the open-loop eigenfunctionals and in the system simplification. The design only requires a finite number of Koopman eigenvalues and the corresponding eigenfunctionals. They are determined from the available state data by making use of eDMD, which is generalized in [ 26 ] to PDE systems using functionals as observables. This results in a general and new data-driven approach for

Read the complete text