pith. sign in

arxiv: 2605.23524 · v1 · pith:TK2MUC2Knew · submitted 2026-05-22 · 📡 eess.SY · cs.SY

Beyond Shrinkage: Foundations of Data-Driven Control for Piecewise Affine Systems

Gianluca Giacomelli , Victor G. Lopez , Simone Formentin , Matthias A. M\"uller , Valentina Breschi This is my paper

Pith reviewed 2026-05-25 03:47 UTC · model grok-4.3

classification 📡 eess.SY cs.SY
keywords data-enabled predictive controlpiecewise affine systemsWillems fundamental lemmabehavioral controlDeePCshrinkage regularizationmisclassification errorshybrid systems
0
0 comments X

The pith

Piecewise affine systems need an extended behavioral lemma for data-driven control, as linear predictors with shrinkage fall short.

A machine-rendered reading of the paper's core claim, the machinery that carries it, and where it could break.

This paper lays the foundations for data-enabled predictive control of piecewise affine systems by first characterizing their behavior and then extending Willems' Fundamental Lemma to represent trajectories directly from raw data. It shows through analysis that DeePC strategies built on linear predictors and shrinkage regularizers produce predictions that are incoherent with the piecewise structure. The work further examines how errors in classifying the different affine regions affect the organization of data for prediction. A sympathetic reader would care because PWA systems are widely used to approximate nonlinear and hybrid dynamics, yet existing data-driven methods have not been adapted to them, limiting explainability and effectiveness in practice.

Core claim

The authors provide a behavioral characterization of piecewise affine systems and extend Willems' Fundamental Lemma to represent their input-output behavior from raw data. They demonstrate that DeePC approaches using a linear predictor together with shrinkage regularizers lack coherence with PWA dynamics. They also analyze the impact of misclassification errors when structuring data for prediction, concluding that effective and explainable control requires moving beyond regularized linear DeePC to methods that respect the piecewise affine nature of the system.

What carries the argument

The extension of Willems' Fundamental Lemma to piecewise affine systems, which represents their behavior from raw input-output data without explicit model identification.

If this is right

  • Piecewise affine systems admit a behavioral characterization that supports data-driven methods.
  • An extended Willems' Fundamental Lemma enables representation of PWA behavior directly from raw data.
  • DeePC strategies based on linear predictors and shrinkage regularizers are incoherent with PWA system behavior.
  • Misclassification errors in identifying affine regions affect data structuring and subsequent prediction quality.
  • Control actions for PWA systems must incorporate PWA-specific structure to remain effective and explainable.

Where Pith is reading between the lines

These are editorial extensions of the paper, not claims the author makes directly.

  • The coherence results suggest that data-driven predictors for hybrid systems will generally need to incorporate mode information rather than relying on a single linear model.
  • The misclassification analysis indicates potential sensitivity of data-driven PWA control to uncertainties in region identification.
  • Similar behavioral extensions could be pursued for other classes of switched or hybrid systems that admit piecewise descriptions.
  • The simple numerical validation leaves open whether the proposed foundations scale to higher-dimensional or more complex PWA examples.

Load-bearing premise

The coherence analysis of linear predictors with shrinkage and the misclassification study, supported by a simple numerical example, suffice to establish the need to extend beyond regularized linear DeePC for PWA systems.

What would settle it

A concrete case where a linear predictor with shrinkage regularization generates predictions that exactly match all possible PWA trajectories from the data, or raw data for which the extended lemma fails to capture the observed PWA behavior.

read the original abstract

Data-enabled predictive control (DeePC) has recently attracted attention as a promising approach for controlling systems directly from raw data, without requiring an explicit identification step. However, DeePC has not yet been extended to piecewise affine (PWA) systems, despite their extensive use in the (predictive) control literature and their universal approximation capabilities. To address this gap, in this work, we lay the foundations for data-enabled predictive control of PWA systems, providing: $(i)$ their behavioral characterization; $(ii)$ an extension of Willems' Fundamental Lemma to represent their behavior from raw data; $(iii)$ an analysis of the coherence of DeePC strategies using a linear predictor and shrinkage regularizers; and $(iv)$ a study of the impact of misclassification errors on structuring data for prediction. Our theoretical findings are validated by numerical results on a simple example, emphasizing the need to extend beyond a regularized version of the foundational DeePC framework to design control actions that are both effective and coherent with a PWA system's behavior, thus ensuring the controller's explainability.

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, simulated authors' rebuttal, and a circularity audit. Tearing a paper down is the easy half of reading it; the pith above is the substance, this is the friction.

Referee Report

2 major / 1 minor

Summary. The paper lays foundations for data-enabled predictive control (DeePC) of piecewise affine (PWA) systems. It provides (i) a behavioral characterization of PWA systems, (ii) an extension of Willems' Fundamental Lemma to represent PWA behavior from raw data, (iii) an analysis of coherence for DeePC strategies that employ linear predictors and shrinkage regularizers, and (iv) a study of misclassification errors when structuring data for prediction. Theoretical findings are validated on a single simple numerical example, which is used to argue that regularized linear DeePC must be extended to achieve control actions coherent with PWA dynamics.

Significance. If the behavioral characterization and Fundamental Lemma extension are rigorously derived, and if the coherence analysis demonstrates quantifiable limitations of linear shrinkage for PWA switching, the work would provide a useful starting point for data-driven control of a practically relevant system class. The explicit treatment of misclassification effects on data matrices is a constructive contribution. However, the single-example validation does not yet establish generality of the 'beyond shrinkage' conclusion.

major comments (2)
  1. [§7] §7 (Numerical Example): The validation consists of one low-dimensional PWA system. To support the central claim that linear DeePC with shrinkage must be extended, the example must exhibit PWA-specific phenomena (mode-dependent trajectories or switching-induced rank deficiencies) where linear shrinkage produces measurably incoherent predictions or higher closed-loop cost, with explicit quantitative comparison to a PWA-aware predictor. A single case leaves open whether the observed incoherence is general or an artifact of the chosen regions.
  2. [§5] §5 (Coherence Analysis): The analysis of linear-predictor coherence under shrinkage must include a concrete metric (e.g., prediction error norm or data-matrix rank drop across modes) showing that shrinkage cannot restore coherence for PWA trajectories; without this, the necessity of moving 'beyond shrinkage' rests on an extrapolation rather than a demonstrated insufficiency.
minor comments (1)
  1. [Abstract] Abstract: The phrase 'simple numerical example' should be accompanied by the system order and number of affine regions to allow readers to assess the scope of the validation immediately.

Simulated Author's Rebuttal

2 responses · 0 unresolved

We thank the referee for the constructive and detailed comments. We address each major comment point by point below, indicating the revisions we will incorporate to strengthen the manuscript.

read point-by-point responses

  1. Referee: [§7] §7 (Numerical Example): The validation consists of one low-dimensional PWA system. To support the central claim that linear DeePC with shrinkage must be extended, the example must exhibit PWA-specific phenomena (mode-dependent trajectories or switching-induced rank deficiencies) where linear shrinkage produces measurably incoherent predictions or higher closed-loop cost, with explicit quantitative comparison to a PWA-aware predictor. A single case leaves open whether the observed incoherence is general or an artifact of the chosen regions.

    Authors: We agree that reliance on a single low-dimensional example limits the strength of the central claim. In the revised version we will augment §7 with explicit quantitative comparisons (prediction error norms and closed-loop costs) between the linear shrinkage DeePC and a PWA-aware predictor, while explicitly highlighting mode-dependent trajectories and switching-induced rank deficiencies in the data matrices. These additions will demonstrate that the observed incoherence arises from the linear predictor’s inability to capture PWA switching rather than from the specific choice of regions. revision: yes

  2. Referee: [§5] §5 (Coherence Analysis): The analysis of linear-predictor coherence under shrinkage must include a concrete metric (e.g., prediction error norm or data-matrix rank drop across modes) showing that shrinkage cannot restore coherence for PWA trajectories; without this, the necessity of moving 'beyond shrinkage' rests on an extrapolation rather than a demonstrated insufficiency.

    Authors: We will revise §5 to include the requested concrete metrics—specifically the prediction error norm and the rank drop of the data matrix across modes—thereby showing quantitatively that shrinkage regularizers cannot restore coherence when trajectories cross PWA switching surfaces. This will replace the current extrapolation with a direct demonstration of insufficiency. revision: yes

Circularity Check

0 steps flagged

No circularity: claims rest on independent theoretical extensions and external validation

full rationale

The abstract and provided text outline four distinct contributions—behavioral characterization of PWA systems, an extension of Willems' Fundamental Lemma, coherence analysis of linear predictors with shrinkage, and misclassification impact—without any quoted equations, self-citations, or fitted parameters that reduce a 'prediction' or result to the inputs by construction. The numerical example is presented as validation rather than a fitted input renamed as output. No self-definitional loops, ansatzes smuggled via citation, or uniqueness theorems imported from the authors' prior work appear in the load-bearing steps. This is the normal self-contained case; the derivation chain does not collapse to its own definitions or data fits.

Axiom & Free-Parameter Ledger

0 free parameters · 0 axioms · 0 invented entities

Insufficient detail in abstract to identify free parameters, axioms, or invented entities; no equations or modeling choices are shown.

pith-pipeline@v0.9.0 · 5734 in / 1085 out tokens · 21971 ms · 2026-05-25T03:47:31.530628+00:00 · methodology

discussion (0)

Sign in with ORCID, Apple, or X to comment. Anyone can read and Pith papers without signing in.

Reference graph

Works this paper leans on

55 extracted references · 45 canonical work pages · 1 internal anchor

  1. [1]

    H. Akaike. Fitting autoregressive models for predictio n. Annals of the Institute of Statistical Mathematics , 21(2):243–247, 1969

    1969

  2. [2]

    Alsalti, V

    M. Alsalti, V . G. Lopez, J. Berberich, F. Allg¨ ower, and M . A. M¨ uller. Data-driven nonlinear predictive control for feedback lin earizable sys- tems. IF AC-PapersOnLine, 56(2):617–624, 2023

    work page 2023

  3. [3]

    Beerwerth and B

    J. Beerwerth and B. Alrifaee. Less is more: Contextual sa mpling for nonlinear data-enabled predictive control. arXiv e-prints , pages arXiv– 2503, 2025. 16 GENERIC COLORIZED JOURNAL, VOL. XX, NO. XX, XXXX 2026

    2025

  4. [4]

    Bemporad, G

    A. Bemporad, G. Ferrari-Trecate, and M. Morari. Observa bility and con- trollability of piecewise affine and hybrid systems. IEEE Transactions on Automatic Control , 45(10):1864–1876, 2000

    2000

  5. [5]

    Bemporad, A

    A. Bemporad, A. Garulli, S. Paoletti, and A. Vicino. A gre edy approach to identification of piecewise affine models. In International W orkshop on Hybrid Systems: Computation and Control , pages 97–112. Springer, 2003

    2003

  6. [6]

    Bemporad, A

    A. Bemporad, A. Garulli, S. Paoletti, and A. Vicino. A bou nded-error approach to piecewise affine system identification. IEEE Transactions on Automatic Control , 50(10):1567–1580, 2005

    2005

  7. [7]

    Berberich and F

    J. Berberich and F. Allg¨ ower. An overview of systems-th eoretic guarantees in data-driven model predictive control. Annual Review of Control, Robotics, and Autonomous Systems , 8(1):77–100, 2025

    2025

  8. [8]

    Berberich, J

    J. Berberich, J. K¨ ohler, M. A. M¨ uller, and F. Allg¨ ower . Data-driven model predictive control with stability and robustness gua rantees. IEEE Transactions on Automatic Control , 66(4):1702–1717, 2020

    work page 2020

  9. [9]

    Berberich, J

    J. Berberich, J. K¨ ohler, M. A. M¨ uller, and F. Allg¨ ower . Linear tracking MPC for nonlinear systems-part II: The data-drive n case. IEEE Transactions on Automatic Control , 67(9):4406–4421, 2022

    work page 2022

  10. [10]

    Bianchi, S

    M. Bianchi, S. Grammatico, and J. Cort´ es. Data-driven stabilization of switched and constrained linear systems. Automatica, 171:111974, 2025

    work page 2025

  11. [11]

    Borrelli, A

    F. Borrelli, A. Bemporad, and M. Morari. Predictive Control for Linear and Hybrid Systems . Cambridge University Press, 2017

    work page 2017

  12. [12]

    L. Breiman. Hinging hyperplanes for regression, class ification, and function approximation. IEEE Transactions on Information Theory , 39(3):999–1013, 2002

    work page 2002

  13. [13]

    Breschi, A

    V . Breschi, A. Chiuso, and S. Formentin. Data-driven pr edictive control in a stochastic setting: a unified framework. Automatica, 152:110961, 2023

    2023

  14. [14]

    Breschi, D

    V . Breschi, D. Piga, and A. Bemporad. Piecewise affine re gression via recursive multiple least squares and multicategory dis crimination. Automatica, 73:155–162, 2016

    work page 2016

  15. [15]

    Breschi, A

    V . Breschi, A. Sassella, and S. Formentin. On the design of regularized explicit predictive controllers from input-output data. IEEE Transactions on Automatic Control , 68(8):4977–4983, 2023

    work page 2023

  16. [16]

    Chiuso, M

    A. Chiuso, M. Fabris, V . Breschi, and S. Formentin. Harn essing uncertainty for a separation principle in direct data-driv en predictive control. Automatica, 173:112070, 2025

    work page 2025

  17. [17]

    Coulson, J

    J. Coulson, J. Lygeros, and F. D¨ orfler. Data-enabled pr edictive control: In the shallows of the deepc. In 2019 18th European Control Conference (ECC), pages 307–312. IEEE, 2019

    work page 2019

  18. [18]

    Coulson, J

    J. Coulson, J. Lygeros, and F. D¨ orfler. Regularized and distributionally robust data-enabled predictive control. In 2019 IEEE 58th Conference on Decision and Control (CDC) , pages 2696–2701. IEEE, 2019

    work page 2019

  19. [19]

    de Jong, V

    T. de Jong, V . Breschi, M. Schoukens, and M. Lazar. Koopm an data- driven predictive control with robust stability and recurs ive feasibility guarantees. In 2024 IEEE 63rd Conference on Decision and Control (CDC), pages 140–145. IEEE, 2024

    work page 2024

  20. [20]

    D¨ orfler, J

    F. D¨ orfler, J. Coulson, and I. Markovsky. Bridging dire ct and indirect data-driven control formulations via regularizations and relaxations. IEEE Transactions on Automatic Control , 68(2):883–897, 2022

    2022

  21. [21]

    Eising, S

    J. Eising, S. Liu, S. Mart´ ınez, and J. Cort´ es. Data-dr iven mode detection and stabilization of unknown switched linear sys tems. IEEE Transactions on Automatic Control , 70(6):3830–3845, 2024

    2024

  22. [22]

    Faye-Bedrin, P

    A. Faye-Bedrin, P . Chauchat, S. Aranovskiy, and R. Bour dais. Willems’ lemma reformulations: which operators preserve LTI system behavior? IEEE Control Systems Letters , 9:258–263, 2025

    work page 2025

  23. [23]

    Fazzi and I

    A. Fazzi and I. Markovsky. Addition and intersection of linear time- invariant behaviors. IF AC Journal of Systems and Control , 26:100233, 2023

    work page 2023

  24. [24]

    Ferrari-Trecate, M

    G. Ferrari-Trecate, M. Muselli, D. Liberati, and M. Mor ari. A clustering technique for the identification of piecewise affine systems . Automatica, 39(2):205–217, 2003

    2003

  25. [25]

    A note on the group lasso and a sparse group lasso

    J. Friedman, T. Hastie, and R. Tibshirani. A note on the g roup lasso and a sparse group lasso. arXiv preprint arXiv:1001.0736 , 2010

    work page internal anchor Pith review Pith/arXiv arXiv 2010

  26. [26]

    Giacomelli, S

    G. Giacomelli, S. Formentin, V . G. Lopez, M. A. M¨ uller, and V . Breschi. Insights into the explainability of lasso-based deepc for n onlinear systems. In 2025 IEEE 64th Conference on Decision and Control (CDC) , pages 1036–1041. IEEE, 2025

    work page 2025

  27. [27]

    J. A. Hartigan, M. A. Wong, et al. A k-means clustering al gorithm. Applied statistics , 28(1):100–108, 1979

    work page 1979

  28. [28]

    Hastie, R

    T. Hastie, R. Tibshirani, and M. Wainwright. Statistic al learning with sparsity. Monographs on statistics and applied probability , 143(143):8, 2015

    work page 2015

  29. [29]

    T. He. Lasso and general L1-regularized regression und er linear equality and inequality constraints. PhD. Dissertation, Purdue University , 2011

    work page 2011

  30. [30]

    Hjalmarsson

    H. Hjalmarsson. From experiment design to closed-loop control. Automatica, 41(3):393–438, 2005

    work page 2005

  31. [31]

    Hu and T

    K. Hu and T. Liu. Data-driven output feedback control fo r unknown piecewise affine systems. IEEE Transactions on Neural Networks and Learning Systems , 2026

    work page 2026

  32. [32]

    L. C. Iacob, R. T ´ oth, and M. Schoukens. Koopman form of n onlinear systems with inputs. Automatica, 162:111525, 2024

    work page 2024

  33. [33]

    Kl¨ adtke and M

    M. Kl¨ adtke and M. S. Darup. Implicit predictors in regu larized data- driven predictive control. IEEE Control Systems Letters , 7:2479–2484, 2023

    work page 2023

  34. [34]

    Kl¨ adtke and M

    M. Kl¨ adtke and M. S. Darup. On data usage and predictive behavior of data-driven predictive control with 1-norm regularizatio n. IEEE Control Systems Letters , 9:943–948, 2025

    work page 2025

  35. [35]

    Kl¨ adtke and M

    M. Kl¨ adtke and M. Schulze Darup. Towards explainable d ata-driven predictive control with regularizations. at-Automatisierungstechnik, 73(6):365–382, 2025

    work page 2025

  36. [36]

    B. Li, T. Wu, B. Cai, L. Zhang, and H. Chen. Online data-dr iven MPC for unknown switched linear systems: A hybrid sampling appr oach. IEEE Transactions on Automatic Control , 2026

    work page 2026

  37. [37]

    J. Li, S. Li, and D. Chen. Datamodel-based data selectio n for nonlinear data-enabled predictive control. arXiv preprint arXiv:2512.00276 , 2025

    work page arXiv 2025

  38. [38]

    Lin and R

    J.-N. Lin and R. Unbehauen. Canonical piecewise-linea r approximations. IEEE Transactions on Circuits and Systems I: Fundamental Th eory and Applications, 39(8):697–699, 2002

    work page 2002

  39. [39]

    Markovsky

    I. Markovsky. Data-driven simulation of generalized b ilinear systems via linear time-invariant embedding. IEEE Transactions on Automatic Control, 68(2):1101–1106, 2022

    work page 2022

  40. [40]

    Markovsky and F

    I. Markovsky and F. D¨ orfler. Identifiability in the beha vioral setting. IEEE Transactions on Automatic Control , 68(3):1667–1677, 2022

    work page 2022

  41. [41]

    Markovsky, J

    I. Markovsky, J. Eising, and A. Padoan. How to represent and identify affine time-invariant systems? IEEE Control Systems Letters , 9:1207– 1212, 2025

    work page 2025

  42. [42]

    Martinelli, M

    A. Martinelli, M. Gargiani, M. Draskovic, and J. Lygero s. Data-driven optimal control of affine systems: A linear programming pers pective. IEEE Control Systems Letters , 6:3092–3097, 2022

    work page 2022

  43. [43]

    L. P . Miao Y u, Federico Bianchi. Identification of piece wise affine systems using a cluster refinement technique. European Journal of Control, 83:101204, 2025

    work page 2025

  44. [44]

    N¨ af, K

    J. N¨ af, K. Moffat, J. Eising, and F. D¨ orfler. Choose wis ely: Data-driven predictive control for nonlinear systems using online data selection. arXiv preprint arXiv:2503.18845 , 2025

    work page arXiv 2025

  45. [45]

    Paoletti, J

    S. Paoletti, J. Roll, A. Garulli, and A. Vicino. On the in put-output representation of piecewise affine state space models. IEEE Transactions on Automatic Control , 55(1):60–73, 2009

    work page 2009

  46. [46]

    Pasini, M

    L. Pasini, M. Bruschetta, and A. Chiuso. Taming non-lin earity with local data-driven predictive control. IF AC Journal of Systems and Control , 35:100396, 2026

    work page 2026

  47. [47]

    Rosolia and F

    U. Rosolia and F. Borrelli. Learning how to autonomousl y race a car: A predictive control approach. IEEE Transactions on Control Systems Technology, 28(6):2713–2719, 2019

    work page 2019

  48. [48]

    Rotulo, C

    M. Rotulo, C. De Persis, and P . Tesi. Online learning of d ata- driven controllers for unknown switched linear systems. Automatica, 145:110519, 2022

    work page 2022

  49. [49]

    V erdult and M

    V . V erdult and M. V erhaegen. Subspace identification of piecewise linear systems. In 43rd IEEE Conference on Decision and Control , volume 4, pages 3838–3843. IEEE, 2004

    work page 2004

  50. [50]

    V erhoek, I

    C. V erhoek, I. Markovsky, S. Haesaert, and R. T ´ oth. A be havioral approach for LPV data-driven representations. IEEE Transactions on Automatic Control, 71(3):1616–1629, 2026

    work page 2026

  51. [51]

    Wang, K.-Z

    Z.-M. Wang, K.-Z. Liu, X.-L. Cheng, and X.-M. Sun. Onlin e data-driven model predictive control for switched linear systems. IEEE Transactions on Automatic Control , 70(9):6222–6229, 2025

    work page 2025

  52. [52]

    J. C. Willems. From time series to linear system—part I. finite dimensional linear time invariant systems. Automatica, 22(5):561–580, 1986

    work page 1986

  53. [53]

    J. C. Willems, P . Rapisarda, I. Markovsky, and B. L. De Mo or. A note on persistency of excitation. Systems & Control Letters , 54(4):325–329, 2005

    work page 2005

  54. [54]

    Y uan and Y

    M. Y uan and Y . Lin. Model selection and estimation in reg ression with grouped variables. Journal of the Royal Statistical Society Series B: Statistical Methodology, 68(1):49–67, 2006

    work page 2006

  55. [55]

    P . Zhao, G. Rocha, and B. Y u. Grouped and hierarchical mo del selection through composite absolute penalties. Department of Statistics, UC Berkeley, Tech. Rep, 703, 2006

    work page 2006