Pith. sign in

REVIEW 5 major objections 8 minor 56 references

Fine-tuning for Data-enabled Predictive Control of Noisy Systems by Reinforcement Learning

T0 review · 5 major / 8 minor · reviewed 2026-08-07 · deepseek-v4-flash

Challenge this review Re-run · record.json Download PDF Read on arXiv ↗

Pith's one-line read This paper claims that a Q-table trained offline can select near-optimal DeePC regularization hyperparameters online, letting data-enabled predictive control stay effective as noise changes.

desk verdict Plausible SARSA-based λ_g tuner for DeePC with useful simulations, but the 'near-optimal' claim is unverified and the theory section needs a rewrite. read the letter →

arxiv 2505.24572 v1 pith:3EYFTLID submitted 2025-05-30 math.OC

Jinbao Wang , Shiliang Zhang , Jun Liu , Xuehui Ma , Haolin Liu This is my paper · ORCID
classification math.OC MSC 93C4093E2093C05
keywords data-enabledpredictivecontrolhyperparametertuningreinforcementlearningsequentialdecision-makingSARSADeePCnoiserobustness
verification ladder T0 review T1 audit T2 compute T3 formal

The pith

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

The reading

The paper proposes S-DeePC, a version of data-enabled predictive control (DeePC) whose regularization parameter $\lambda_g$ is adjusted in real time by a reinforcement learning agent rather than fixed in advance or tuned by exhaustive search. The authors argue that the optimal $\lambda_g$ for a noisy system drifts as noise and operating conditions change, and that a Q-table trained offline on input–output data can be used online to select near-optimal $\lambda_g$ at each step. They model the tuning problem as a sequential decision problem whose state is the pair $(\|u\|_{2,n}, M)$, where the first is control energy and $M$ is root-mean-square tracking error, and whose reward is the negative of a weighted difference of the two. In numerical comparisons against fixed-parameter DeePC, the Hanke rule, and DeePC-Hunt, the proposed method adapts faster to a sudden doubling of Gaussian or uniform noise and stabilizes with competitive tracking performance.

What carries the argument

The load-bearing object is the Q-table learned by SARSA, with state $x_k = (\|u\|_{2,n}, M_k)$, action $\lambda_g$, and reward $R_k = -|a x_k(1) - b x_k(2)|$. The authors justify this choice by an empirical curve (Fig. 2) linking $\lambda_g$ to input energy and tracking error, which exhibits a critical inflection point $C$. The Bellman-style recursion (Eq. 17) and the update rule $Q^{\text{new}} \leftarrow Q^{\text{old}} + \alpha(R + \gamma Q^{\text{old}}(\text{next}) - Q^{\text{old}})$ (Eq. 20) carry the learning; the online stage loads the trained Q-table and switches $\lambda_g$ with probability $1-\varepsilon$ to the argmax action, with $\varepsilon$ allowing exploration. The offline/online split keeps the online computation light enough for real-time control.

What would settle it

A direct test: train the Q-table offline with Gaussian noise of variance $10^{-6}$, then run S-DeePC online while the noise variance is ramped continuously from $10^{-6}$ to $10^{-2}$, well outside the training range. If the Markov-state and transfer assumptions hold, tracking error $M$ should stay near its minimum; if the method's adaptivity is only interpolation within the training distribution, $M$ should rise sharply and the selected $\lambda_g$ should fail to track the optimal value. Running the same experiment on a different plant with the same training data would test whether the table encodes plant-specific behavior.

Watch

Extended reading notes

Core claim

The central discovery, as the authors present it, is that a DeePC controller's regularization hyperparameter $\lambda_g$ can be mapped to measurable I/O quantities—control input energy $\|u\|_{2,n}$ and output error $M$—and that this mapping has a consistent shape: increasing $\lambda_g$ first sharply reduces input energy, then saturates, while output error $M$ dips slightly and then rises. This empirical relationship lets the tuning problem be recast as a Markov decision process, where the state $x_k = (\|u\|_{2,n}, M_k)$, the action is the next $\lambda_g$, and the reward is $R_k = -|a\|u\|_{2,n} - bM|$. The authors show that SARSA with an $\varepsilon$-greedy policy, trained offline on noisy trajectories and deployed online, updates a Q-table via the standard recursion and selects $\lambda_g$ that tracks the near-optimal value when noise intensity changes. The claimed result is not a theorem but a demonstrated mechanism: an offline-trained table gives real-time adaptivity without system identification or prior knowledge of the noise distribution.

Load-bearing premise

The method assumes that the pair (input energy, tracking error) together with the current $\lambda_g$ is a sufficient state for choosing the next $\lambda_g$, and that a Q-table trained offline on one noise level still selects good actions when noise changes.

Editorial extensions

If this is right

  • If the method works as claimed, DeePC can be deployed with a pre-trained Q-table and maintain performance when noise statistics change, without re-solving an optimization or re-identifying the system.
  • The tuning framework is not tied to DeePC's specifics; any controller whose performance depends on a scalar regularization weight in a similar I/O-responsive way could use the same state-action-reward design.
  • Because the expensive learning happens offline, the online controller needs only table look-ups and the DeePC update, avoiding the overhead of methods that re-optimize $\lambda_g$ online.
  • The state is built from quantities already computed during control, so the method needs no extra sensors or models.
  • The approach is demonstrated to regain stability within about ten seconds after a noise doubling, suggesting a practical adaptation timescale for slowly varying disturbances.

Reading between the lines

Editorial extensions of the paper, not claims the author makes directly.

  • A testable extension: train the Q-table on one plant and run it on a second plant with different dynamics; the paper's claim of operating without prior knowledge predicts transfer, but the Markov assumption would likely break, causing degraded performance.
  • The reward $J = |a\|u\| - bM|$ with $a=b=1$ fixes the trade-off at one point; the paper leaves open how to set $a$ and $b$ systematically when the control objective is strict output tracking with a hard input constraint.
  • The empirical $\lambda_g$-to-I/O curve has a threshold structure, so an agent that learns which side of the inflection point it is on could be replaced by a simple switching rule; the paper does not test whether the Q-table actually encodes that structure.
  • Because the Q-table is trained on a fixed noise distribution, the claimed robustness across noise levels requires the table to generalize beyond its training distribution, which is untested.
Share X Bluesky LinkedIn Reddit HN

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, and a circularity audit.

Referee Report

5 major / 8 minor

Summary. The paper proposes S-DeePC, a reinforcement-learning-based adaptive tuning scheme for the regularization parameter λ_g in data-enabled predictive control (DeePC). The method trains a SARSA Q-table offline and then adjusts λ_g online using a two-dimensional state (the control input norm and the root-mean-square output error), aiming to maintain performance when noise conditions change without prior system knowledge. The authors claim near-optimal hyperparameter identification and robustness, supported by numerical simulations on a second-order LTI system and a triple-mass-spring system, with comparisons against baseline DeePC, Hanke, and DeePC-Hunt. The central claims are that the learned policy tracks the optimal λ_g and that the offline-trained model transfers to online operation under varying noise.

Significance. If the central claims held, the paper would offer a practical, low-computation way to replace manual or exhaustive DeePC hyperparameter tuning and to adapt to changing noise conditions online. The offline/online decomposition and the use of a Q-table for λ_g selection are sensible engineering directions, and the comparison with Hanke and DeePC-Hunt addresses a relevant literature. However, the theoretical derivation contains mathematical errors, the empirical evaluation does not report the selected λ_g values or compare them to an oracle optimum, and the hyperparameters of the RL scheme are tuned on the same test scenario used for evaluation. These issues mean that the current evidence does not establish near-optimal λ_g tracking or the claimed robustness; the paper's contribution is therefore not yet substantiated, though the underlying idea is salvageable.

major comments (5)
  1. [Section IV-B and Abstract/Conclusion] The central claim that S-DeePC identifies 'near-optimal' λ_g is never directly tested. The paper plots the composite objective J of Eq. (8) but never reports the λ_g values chosen online or their gap to the oracle optimal λ_g for each noise condition. An improvement in J over fixed baselines does not establish near-optimal hyperparameter tracking. Please add a plot of the online-selected λ_g(t) versus the optimal λ_g found by exhaustive search for each noise level, and quantify the gap.
  2. [Section III, Eqs. (15)-(18)] The Bellman recursion contains an unjustified interchange of max and expectation. In Eq. (18), the third equality rewrites max over future policies of an expectation as the expectation of a max, which is not valid in general; the max over future policies should be conditional on the realized next state, not taken outside the expectation. In addition, Eq. (15) defines Q_{z→c}(x, λ_g) with an argmax over λ_g while also taking λ_g as an argument, which is circular. These issues undermine the theoretical foundation for the Q-learning/SARSA update in Eq. (20).
  3. [Section III, Eq. (10)] The stochastic optimization problem (10a)-(10g) introduces probability densities pf_{x,k}, pf_k, pf_{λ,k} and variables λ̂_{g,k} that are never defined or used in the rest of the paper. The paragraph after Eq. (10) also refers to functions f_{u,k}, μ_k, and û_k that are undefined. This makes the claimed 'probabilistic inference' step vacuous, and the connection between Eq. (10) and the SARSA algorithm is not established.
  4. [Section IV-A] The RL meta-parameters and horizon n are selected by minimizing the same objective J on the same experimental setup later used for performance evaluation (Figs. 5-6). This in-sample tuning means the reported superiority of S-DeePC may partly reflect overfitting to the test scenario. Please use a separate validation set, or report performance across a range of hyperparameters without selecting on the test data.
  5. [Section III, Eqs. (10e)-(11) and Algorithm 1] The state x_k = (||u||_{2,n}, M_k) together with λ_{g,k} is assumed to be a Markov state for the controlled system, and the offline-trained Q-table is assumed to transfer to online operation with different noise levels. Neither assumption is verified. The transition model (11) is unknown, and the Bellman recursion (17) requires Markovianity; the paper should test the validity of the state representation, for example by checking whether the learned Q-values yield consistent λ_g choices for the same state under different noise histories.
minor comments (8)
  1. [Section IV, introductory paragraph] The text refers to 'baseline PeePC'; this should read 'baseline DeePC.'
  2. [Section III, paragraph after Eq. (9)] The sentence 'We would like to not that random additive noise...' contains a typo; it should read 'note.'
  3. [Section II, Eq. (6)] The DeePC formulation in Eq. (6) is not the standard formulation used in the simulations; the paper does not explain how this bilevel rank-constrained problem is implemented in MATLAB/OSQP. Please clarify the actual optimization problem solved.
  4. [Section II, Notation] The notation ||e||_n is ambiguous: it is defined as sqrt(e^T n e) where n is a matrix, but n is also used as the number of samples later. Use a different symbol for the weighting matrix.
  5. [Section III, Eqs. (17)-(20)] The paper calls the method SARSA, but Eqs. (17)-(18) are Q-learning Bellman optimality equations, while Eq. (20) is the SARSA update rule. Please clarify the relationship between the two and which update is actually used.
  6. [Algorithm 1] The lines 'Hankel matrix←(3)' and 'λ_g,k, x_k ← Discretize U_p, Y_p, λ_g, using (11)' are vague; specify the discretization procedure and how Eq. (11) is used to construct the state and action.
  7. [Section IV-C] Reference [46] is cited for the triple-mass-spring system, but [46] is a paper on the relationship between DeePC and subspace predictive control and does not describe that system; the reference appears to be incorrect.
  8. [Section V, Conclusion] The conclusion states that the method 'can effectively identify optimal DeePC hyperparameter', but no selected λ_g values are reported; please align the conclusion with the available evidence.

Circularity Check

0 steps flagged · score 0.0 of 10

No significant circularity: S-DeePC's RL optimization is self-contained; the missing oracle comparison is an evidence gap, not a circular reduction.

full rationale

The paper's derivation chain does not reduce to its inputs. The control cost J in Eq. (8) is defined as |a||u||_{2,n} - bM|, and the RL reward in Eq. (12) is set to R_k = -|a x_k(1)-b x_k(2)|; this is the standard identification of reward with negative cost, not a circularity, because the paper's claim is that the SARSA Q-table trained on data can select λ_g that keeps J small, which is an empirical assertion tested by simulation against baseline DeePC, Hanke, and DeePC-Hunt. The Q-table is constructed from noisy I/O trajectories and updated by Eq. (20), with no fitted parameter that is later renamed as a prediction of λ_g. The RL algorithm's own hyperparameters (α, γ, ε, n) are tuned in Section IV-A by minimizing the same J used in evaluation, and the Q-table is trained and tested on the same second-order system and noise family; this raises an in-sample/overfitting concern and means the 'near-optimal' claim is not verified against an oracle optimal λ_g, but it is a correctness/validation gap rather than a logical equivalence between input and output. Self-citations (e.g., [56] for SARSA/Q-learning) are for standard background and are not load-bearing. No circular step satisfying the evidentiary standard was found.

Assumptions & free parameters 4 free parameters · 4 assumptions · 0 invented entities

The central design rests on several unverified premises: exact-rank data representation in the presence of noise, a Markov state encoded by two scalars, transfer of an offline-trained Q-table to changing noise, and the assumed shape of the lambda_g performance curves in Fig 2. These are domain assumptions or ad hoc choices rather than quantities derived in the paper. No new physical entities are introduced.

free parameters (4)
  • n = 40
    Chosen by scanning n=1..100 and minimizing the average objective J in Fig 5; the paper calls this the optimal choice.
  • RL meta-parameters (alpha, gamma, epsilon) = alpha=0.53, gamma=0.86, epsilon=0.35
    Selected in Section IV-A2 by polynomial or exponential lower-envelope fitting of average J on the same simulation setup used later for evaluation.
  • objective weights a, b = a=1, b=1
    User-specified weights in Eq (8) define the objective and reward; no justification or sensitivity analysis is provided.
  • lambda_g action grid = step 0.006 over [0.006,0.606]; step 0.01 over (0,1] in the spring study
    Discretization of the action space used to generate training data and the Q-table; this design choice directly limits the achievable near-optimality.
assumptions (4)
  • domain assumption The collected noisy input-output data satisfies the fundamental lemma rank condition (Eq 4), so the Hankel matrix fully represents the system behavior.
    The paper assumes noisy raw data still yields a usable non-parametric representation model, but the rank condition is stated for exact trajectories and is an idealization under noise.
  • domain assumption The underlying system is controllable, LTI, and the output disturbance is additive Gaussian or uniform with known parametric form.
    The problem formulation in Section II and all simulations use these restrictions; transfer to other noise types or nonlinear systems is untested.
  • ad hoc to paper The two-scalar state x_k = (||u||_{2,n}, M_k) together with lambda_g forms a Markov state, making the Bellman recursion (17)-(18) and Q-learning update (20) valid.
    This is not proven and is central to the proposed RL method. It enters at Eq (10e) and Eq (11)-(14), and is assumed without verification.
  • ad hoc to paper A well-defined near-optimal lambda_g exists for each noise condition, and the qualitative shape in Fig 2 (monotone decrease of input energy, non-monotone M) holds for the tested systems.
    Section III uses Fig 2 to justify the objective J and the search over lambda_g, but no data or derivation is provided for this relation.

how reviews work

0 comments
Cite this review

Pith. "Pith review of Fine-tuning for Data-enabled Predictive Control of Noisy Systems by Reinforcement Learning." pith.science (2026). https://pith.science/paper/3EYFTLID

@misc{pith2026250524572,
  author       = {Pith},
  title        = {Pith review of: Fine-tuning for Data-enabled Predictive Control of Noisy Systems by Reinforcement Learning},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/3EYFTLID}},
  note         = {Machine review of arXiv:2505.24572}
}
read the original abstract

Data-enabled predictive control (DeePC) leverages system measurements in characterizing system dynamics for optimal control. The performance of DeePC relies on optimizing its hyperparameters, especially in noisy systems where the optimal hyperparameters adapt over time. Existing hyperparameter tuning approaches for DeePC are more than often computationally inefficient or overly conservative. This paper proposes an adaptive DeePC where we guide its hyperparameters adaption through reinforcement learning. We start with establishing the relationship between the system I/O behavior and DeePC hyperparameters. Then we formulate the hyperparameter tuning as a sequential decision-making problem, and we address the decision-making through reinforcement learning. We implement offline training to gain a reinforcement learning model, and we integrate the trained model with DeePC to adjust its hyperparameters adaptively in real time. We conduct numerical simulations with diverse noisy conditions, and the results demonstrate the identification of near-optimal hyperparameters and the robustness of the proposed approach against noises in the control.

Figures

Figures reproduced from arXiv: 2505.24572 by the authors.

Figure 1
Figure 1. Control from A DeePC Perspective. ωini and ωf denote initial and future input-output behavior, respectively [PITH_FULL_IMAGE:figures/full_fig_p002_1.png] view at source ↗
Figure 2
Figure 2. Relationship between DeePC Control input Energy, M (measured by [PITH_FULL_IMAGE:figures/full_fig_p003_2.png] view at source ↗
Figure 3
Figure 3. Block diagram for the fine-tuning of DeePC. [PITH_FULL_IMAGE:figures/full_fig_p004_3.png] view at source ↗
Figures from the paper (7 more)
Figure 4
Figure 4. Figure 4: SARSA architecture in DeePC tuning. Under the reinforcement learning framework, equation (10) is a sequential decision-making problem. Specifically, in sce￾narios where the system’s output is subject to Gaussian noise, the objective is to analyze the system state x, de…
Figure 6
Figure 6. Figure 6: Objective function under difference values of parameter [PITH_FULL_IMAGE:figures/full_fig_p007_6.png]
Figure 5
Figure 5. Figure 5: Objective function under different values of parameter [PITH_FULL_IMAGE:figures/full_fig_p007_5.png]
Figure 7
Figure 7. Figure 7: Convergence comparison between Baseline DeePC, S-DeePC, DeePC [PITH_FULL_IMAGE:figures/full_fig_p007_7.png]
Figure 8
Figure 8. Figure 8: Control performance comparison between baseline DeePC, S-DeePC, DeePC-Hunt, and Hanke under normal and doubled Gaussian noise conditions. [PITH_FULL_IMAGE:figures/full_fig_p008_8.png]
Figure 9
Figure 9. Figure 9: Control performance comparison between baseline DeePC, S-DeePC, DeePC-Hunt, and Hanke under normal and doubled uniform noise conditions. [PITH_FULL_IMAGE:figures/full_fig_p009_9.png]
Figure 11
Figure 11. Figure 11: Control for the middle disc of the triple-mass-spring system. The [PITH_FULL_IMAGE:figures/full_fig_p009_11.png]

Discussion (0). Continue with ORCID to comment.

Reference graph

Works this paper leans on

56 extracted references · 55 canonical work pages

  1. [2]

    Ensemble Control for Stochastic Systems with Asymmetric Laplace Noises

    Yajie Yu, Xuehui Ma, Shiliang Zhang, Zhuzhu Wang, Xubing Shi, Yushuai Li, and Tingwen Huang. Adaptive ensemble control for stochastic systems with mixed asymmetric laplace noises.arXiv preprint arXiv:2405.09973, 2024

    work page Pith review arXiv 2024

  2. [3]

    De- centralized backstepping control for interconnected systems with non- triangular structural uncertainties.IEEE Transactions on Automatic Control, 68(3):1692–1699, 2023

    Jianping Cai, Changyun Wen, Lantao Xing, and Qiuzhen Yan. De- centralized backstepping control for interconnected systems with non- triangular structural uncertainties.IEEE Transactions on Automatic Control, 68(3):1692–1699, 2023

    work page 2023

  3. [4]

    Robust quadratic optimal control of linear systems with ellipsoid-set learning

    Xuehui Ma, Yutong Chen, Shiliang Zhang, Yushuai Li, Fucai Qian, and Zhiyong Sun. Robust quadratic optimal control of linear systems with ellipsoid-set learning. In2024 European Control Conference (ECC), pages 2125–2131, 2024

    work page 2024

  4. [5]

    A nonlinear MPC scheme for output tracking without terminal ingredients.IEEE Transactions on Automatic Control, 68(4):2368–2375, 2023

    Raffaele Soloperto, Johannes K ¨ohler, and Frank Allg ¨ower. A nonlinear MPC scheme for output tracking without terminal ingredients.IEEE Transactions on Automatic Control, 68(4):2368–2375, 2023

    work page 2023

  5. [6]

    Adaptive robust tracking control with active learning for linear systems with ellipsoidal bounded uncertainties.IEEE Transactions on Automatic Control, 69(11):8096–8103, 2024

    Xuehui Ma, Shiliang Zhang, Yushuai Li, Fucai Qian, Zhiyong Sun, and Tingwen Huang. Adaptive robust tracking control with active learning for linear systems with ellipsoidal bounded uncertainties.IEEE Transactions on Automatic Control, 69(11):8096–8103, 2024

    work page 2024

  6. [7]

    Adaptive dual control with online outlier detection for uncertain systems.ISA transactions, 129:157–168, 2022

    Xuehui Ma, Fucai Qian, Shiliang Zhang, Li Wu, and Lei Liu. Adaptive dual control with online outlier detection for uncertain systems.ISA transactions, 129:157–168, 2022

    work page 2022

  7. [8]

    Adaptive quantile control for stochastic system.ISA transactions, 123:110–121, 2022

    Xuehui Ma, Fucai Qian, Shiliang Zhang, and Li Wu. Adaptive quantile control for stochastic system.ISA transactions, 123:110–121, 2022

    work page 2022

  8. [9]

    Active learning for anti-disturbance dual control of unknown nonlinear systems

    Xuehui Ma, Shiliang Zhang, Fucai Qian, Jinbao Wang, and Yushuai Li. Active learning for anti-disturbance dual control of unknown nonlinear systems.arXiv preprint arXiv:2212.08934, 2022

    work page Pith review arXiv 2022

Show all 56 references
  1. [10]

    Dual control for stochastic systems with multiple uncertainties

    Xuehui Ma, Fucai Qian, and Shiliang Zhang. Dual control for stochastic systems with multiple uncertainties. In2020 39th Chinese Control Conference (CCC), pages 1001–1006. IEEE, 2020

    2020

  2. [11]

    Data-driven optimization framework for nonlinear model predictive control.Mathematical Problems in Engineering, 2017(1):9402684, 2017

    Shiliang Zhang, Hui Cao, Yanbin Zhang, Lixin Jia, Zonglin Ye, and Xiali Hei. Data-driven optimization framework for nonlinear model predictive control.Mathematical Problems in Engineering, 2017(1):9402684, 2017

    2017

  3. [12]

    Nonlinear prediction model for ventilation of ball mill pulverizing system

    Yiwei Yuan, Yanbin Zhang, Hui Cao, Gangquan Si, Shiliang Zhang, and Qian Xie. Nonlinear prediction model for ventilation of ball mill pulverizing system. In2016 35th Chinese Control Conference (CCC), pages 2025–2028. IEEE, 2016

    2025

  4. [13]

    A nonlinear model predictive controller based on the linguistic model for biochemical continuous sterilization

    Shiliang Zhang, Hui Cao, Yanbin Zhang, Yiwei Yuan, and Xiaoyan Ma. A nonlinear model predictive controller based on the linguistic model for biochemical continuous sterilization. In2016 35th Chinese Control Conference (CCC), pages 4354–4359. IEEE, 2016

    2016

  5. [14]

    Energy- efficient dynamic matrix control for biochemical continuous sterilization

    Shiliang Zhang, Hui Cao, Yanbin Zhang, and ZongZhen Xiong. Energy- efficient dynamic matrix control for biochemical continuous sterilization. In2016 Chinese Control and Decision Conference (CCDC), pages 612–

  6. [15]

    A predictive controller with self-renewal model observer for continuous sterilization of biochemical process

    Shiliang Zhang, Hui Cao, Yanbin Zhang, and Lixin Jia. A predictive controller with self-renewal model observer for continuous sterilization of biochemical process. In2015 34th Chinese Control Conference (CCC), pages 3998–4003. IEEE, 2015

    2015

  7. [16]

    Distributed stochastic model predictive control for cyber–physical systems with multiple state delays and probabilistic saturation constraints.Automatica, 129:109574, 2021

    Langwen Zhang, Bohui Wang, Yuanlong Li, and Yang Tang. Distributed stochastic model predictive control for cyber–physical systems with multiple state delays and probabilistic saturation constraints.Automatica, 129:109574, 2021

    2021

  8. [17]

    Control strategy for biopharmaceu- tical production by model predictive control.Biotechnology Progress, 40(2):e3426, 2024

    Touraj Eslami and Alois Jungbauer. Control strategy for biopharmaceu- tical production by model predictive control.Biotechnology Progress, 40(2):e3426, 2024

    2024

  9. [18]

    Distributionally robust chance constrained data-enabled predictive control.IEEE Trans- actions on Automatic Control, 67(7):3289–3304, 2022

    Jeremy Coulson, John Lygeros, and Florian D ¨orfler. Distributionally robust chance constrained data-enabled predictive control.IEEE Trans- actions on Automatic Control, 67(7):3289–3304, 2022

    2022

  10. [19]

    Ramirez-Mendoza, and Yang Xu

    Jianhong Wang, Ricardo A. Ramirez-Mendoza, and Yang Xu. Nonlinear direct data-driven control for UA V formation flight system.Journal of Systems Engineering and Electronics, 34(6):1409–1418, 2023

    2023

  11. [20]

    van Waarde, Jaap Eising, Harry L

    Henk J. van Waarde, Jaap Eising, Harry L. Trentelman, and M. Kanat Camlibel. Data informativity: A new perspective on data-driven analysis and control.IEEE Transactions on Automatic Control, 65(11):4753– 4768, 2020

    2020

  12. [21]

    Adaptive fault-tolerant control of underwater vehicles with thruster failures.arXiv preprint arXiv:2504.16037, 2025

    Haolin Liu, Shiliang Zhang, Shangbin Jiao, Xiaohui Zhang, Xuehui Ma, Yan Yan, Wenchuan Cui, and Youmin Zhang. Adaptive fault-tolerant control of underwater vehicles with thruster failures.arXiv preprint arXiv:2504.16037, 2025

    2025 arXiv

  13. [22]

    A bionic data-driven approach for long-distance underwater navigation with anomaly resistance.IEEE Transactions on Instrumentation and Measurement, 74:1–14, 2025

    Songnan Yang, Xiaohui Zhang, Shiliang Zhang, Xuehui Ma, Wenqi Bai, Yushuai Li, and Tingwen Huang. A bionic data-driven approach for long-distance underwater navigation with anomaly resistance.IEEE Transactions on Instrumentation and Measurement, 74:1–14, 2025

    2025

  14. [23]

    Exploring the generalizability of geomagnetic navigation: A deep reinforcement learning approach with policy distillation.arXiv preprint arXiv:2502.05069, 2025

    Wenqi Bai, Shiliang Zhang, Xiaohui Zhang, Xuehui Ma, Songnan Yang, Yushuai Li, and Tingwen Huang. Exploring the generalizability of geomagnetic navigation: A deep reinforcement learning approach with policy distillation.arXiv preprint arXiv:2502.05069, 2025

    2025 arXiv

  15. [24]

    Hardware-in-the-loop simulation testbed for geomagnetic navigation.arXiv preprint arXiv:2412.11882, 2024

    Songnan Yang, Shiliang Zhang, Qianyun Zhang, Xiaohui Zhang, and Xuehui Ma. Hardware-in-the-loop simulation testbed for geomagnetic navigation.arXiv preprint arXiv:2412.11882, 2024

    2024 arXiv

  16. [25]

    Long-distance geomagnetic navigation in gnss- denied environments with deep reinforcement learning.arXiv preprint arXiv:2410.15837, 2024

    Wenqi Bai, Xiaohui Zhang, Shiliang Zhang, Songnan Yang, Yushuai Li, and Tingwen Huang. Long-distance geomagnetic navigation in gnss- denied environments with deep reinforcement learning.arXiv preprint arXiv:2410.15837, 2024

    2024 arXiv

  17. [26]

    The impact of integration of renewable energy on imbalance settlement: Resilience analysis

    Shiliang Zhang, Sabita Maharjan, Raul Shahi, and Xuehui Ma. The impact of integration of renewable energy on imbalance settlement: Resilience analysis. In2024 IEEE International Conference on Com- munications, Control, and Computing Technologies for Smart Grids (SmartGridComm)...

    2024

  18. [27]

    Extended vehicle energy dataset (eved): an enhanced large- scale dataset for deep learning on vehicle trip energy consumption.arXiv preprint arXiv:2203.08630, 2022

    Shiliang Zhang, Dyako Fatih, Fahmi Abdulqadir, Tobias Schwarz, and Xuehui Ma. Extended vehicle energy dataset (eved): an enhanced large- scale dataset for deep learning on vehicle trip energy consumption.arXiv preprint arXiv:2203.08630, 2022

    2022 arXiv

  19. [28]

    An energy consumption model for electrical vehicle networks via extended federated-learning

    Shiliang Zhang. An energy consumption model for electrical vehicle networks via extended federated-learning. In2021 IEEE Intelligent V ehicles Symposium (IV), pages 354–361. IEEE, 2021

    2021

  20. [29]

    Willems, I

    J.C. Willems, I. Markovsky, P. Rapisarda, and B.L.M. De Moor. A note on persistency of excitation. In2004 43rd IEEE Conference on Decision and Control (CDC) (IEEE Cat. No.04CH37601), volume 3, pages 2630–2631 V ol.3, 2004

    2004

  21. [30]

    I. A. M S. Exact and approximate modeling of linear systems: A behavioral approach.Exact and approximate modeling of linear systems :, 2006

    2006

  22. [31]

    Data-driven simulation and control.International Journal of Control, (12), 2008

    Ivan Markovsky and Paolo Rapisarda. Data-driven simulation and control.International Journal of Control, (12), 2008. 11

    2008

  23. [32]

    A missing data approach to data-driven filtering and control.IEEE Transactions on Automatic Control, 62(4):1972–1978, 2017

    Ivan Markovsky. A missing data approach to data-driven filtering and control.IEEE Transactions on Automatic Control, 62(4):1972–1978, 2017

    1972

  24. [33]

    Behavioral systems theory in data-driven analysis, signal processing, and control.Annual Reviews in Control, 52:42–64, 2021

    Ivan Markovsky and Florian D ¨orfler. Behavioral systems theory in data-driven analysis, signal processing, and control.Annual Reviews in Control, 52:42–64, 2021

    2021

  25. [34]

    Data-driven dynamic interpolation and approximation.Automatica, 135:110008, 2022

    Ivan Markovsky and Florian D ¨orfler. Data-driven dynamic interpolation and approximation.Automatica, 135:110008, 2022

    2022

  26. [35]

    Robust and kernelized data-enabled predictive control for nonlinear systems.IEEE Transac- tions on Control Systems Technology, 32(2):611–624, 2024

    Linbin Huang, John Lygeros, and Florian D ¨orfler. Robust and kernelized data-enabled predictive control for nonlinear systems.IEEE Transac- tions on Control Systems Technology, 32(2):611–624, 2024

    2024

  27. [36]

    Decentralized data-enabled predictive control for power system oscil- lation damping.IEEE Transactions on Control Systems Technology, 30(3):1065–1077, 2022

    Linbin Huang, Jeremy Coulson, John Lygeros, and Florian D ¨orfler. Decentralized data-enabled predictive control for power system oscil- lation damping.IEEE Transactions on Control Systems Technology, 30(3):1065–1077, 2022

    2022

  28. [37]

    Data-driven continuous-set predictive current control for synchronous motor drives.IEEE Transactions on Power Electronics, 37(6):6637–6646, 2022

    Paolo Gherardo Carlet, Andrea Favato, Saverio Bolognani, and Flo- rian D ¨orfler. Data-driven continuous-set predictive current control for synchronous motor drives.IEEE Transactions on Power Electronics, 37(6):6637–6646, 2022

    2022

  29. [38]

    Kerrigan, Paola Falugi, Marta Zagorowska, and Nilay Shah

    Edward O’Dwyer, Eric C. Kerrigan, Paola Falugi, Marta Zagorowska, and Nilay Shah. Data-driven predictive control with improved per- formance using segmented trajectories.IEEE Transactions on Control Systems Technology, 31(3):1355–1365, 2023

    2023

  30. [39]

    A data-driven model predictive control for alleviating thermal overloads in the presence of possible false data.IEEE Transactions on Industry Applications, 57(2):1872–1881, 2021

    Rui Ma, Sagnik Basumallik, Sara Eftekharnejad, and Fanxin Kong. A data-driven model predictive control for alleviating thermal overloads in the presence of possible false data.IEEE Transactions on Industry Applications, 57(2):1872–1881, 2021

    2021

  31. [40]

    An outlier detection scheme for dynamical sequential datasets.Commu- nications in Statistics-Simulation and Computation, 48(5):1450–1502, 2019

    Shiliang Zhang, Hui Cao, Zonglin Ye, Yanbin Zhang, and Xiali Hei. An outlier detection scheme for dynamical sequential datasets.Commu- nications in Statistics-Simulation and Computation, 48(5):1450–1502, 2019

    2019

  32. [41]

    Sequential outlier criterion for sparsification of online adaptive filter- ing.IEEE Transactions on Neural Networks and Learning Systems, 29(11):5277–5291, 2018

    Shiliang Zhang, Hui Cao, Shuo Yang, Yanbin Zhang, and Xiali Hei. Sequential outlier criterion for sparsification of online adaptive filter- ing.IEEE Transactions on Neural Networks and Learning Systems, 29(11):5277–5291, 2018

    2018

  33. [42]

    Robust data-enabled predictive control: Tractable formulations and performance guarantees.IEEE Transactions on Automatic Control, 68(5):3163–3170, 2023

    Linbin Huang, Jianzhe Zhen, John Lygeros, and Florian D ¨orfler. Robust data-enabled predictive control: Tractable formulations and performance guarantees.IEEE Transactions on Automatic Control, 68(5):3163–3170, 2023

    2023

  34. [43]

    On the equivalence of direct and indirect data-driven predictive control approaches.IEEE Control Systems Letters, 2024

    Per Mattsson, Fabio Bonassi, Valentina Breschi, and Thomas B Sch ¨on. On the equivalence of direct and indirect data-driven predictive control approaches.IEEE Control Systems Letters, 2024

    2024

  35. [44]

    Data-driven predictive control with improved per- formance using segmented trajectories.IEEE Transactions on Control Systems Technology, 31(3):1355–1365, 2022

    Edward O’Dwyer, Eric C Kerrigan, Paola Falugi, Marta Zagorowska, and Nilay Shah. Data-driven predictive control with improved per- formance using segmented trajectories.IEEE Transactions on Control Systems Technology, 31(3):1355–1365, 2022

    2022

  36. [45]

    Data-driven control based on the behavioral approach: From theory to applications in power systems.IEEE Control Systems Magazine, 43(5):28–68, 2023

    Ivan Markovsky, Linbin Huang, and Florian D ¨orfler. Data-driven control based on the behavioral approach: From theory to applications in power systems.IEEE Control Systems Magazine, 43(5):28–68, 2023

    2023

  37. [46]

    On the relationship between data-enabled predictive control and subspace predictive control

    Felix Fiedler and Sergio Lucia. On the relationship between data-enabled predictive control and subspace predictive control. In2021 European Control Conference (ECC), pages 222–229. IEEE, 2021

    2021

  38. [47]

    Physics-augmented data-enabled predictive control for eco-driving of mixed traffic considering diverse human behaviors.IEEE Transactions on Control Systems Technology, 2024

    Dongjun Li, Kaixiang Zhang, Haoxuan Dong, Qun Wang, Zhaojian Li, and Ziyou Song. Physics-augmented data-enabled predictive control for eco-driving of mixed traffic considering diverse human behaviors.IEEE Transactions on Control Systems Technology, 2024

    2024

  39. [48]

    Cloud- based computational data-enabled predictive control.IEEE Internet of Things Journal, 9(24):24949–24962, 2022

    Li Dai, Teng Huang, Runze Gao, Yuan Zhang, and Yuanqing Xia. Cloud- based computational data-enabled predictive control.IEEE Internet of Things Journal, 9(24):24949–24962, 2022

    2022

  40. [49]

    An extended kalman filter for data-enabled predictive control.IEEE Control Systems Letters, 4(4):994–999, 2020

    Daniele Alpago, Florian D ¨orfler, and John Lygeros. An extended kalman filter for data-enabled predictive control.IEEE Control Systems Letters, 4(4):994–999, 2020

    2020

  41. [50]

    Subspace predictive control of flexible structures actuated by piezoelectric elements

    Gijs Van Der Veen and Jan-Willem van Wingerden. Subspace predictive control of flexible structures actuated by piezoelectric elements. In 2024 IEEE Conference on Control Technology and Applications (CCTA), pages 316–322. IEEE, 2024

    2024

  42. [51]

    Decentralized data-enabled predictive control for power system oscil- lation damping.IEEE Transactions on Control Systems Technology, 30(3):1065–1077, 2021

    Linbin Huang, Jeremy Coulson, John Lygeros, and Florian D ¨orfler. Decentralized data-enabled predictive control for power system oscil- lation damping.IEEE Transactions on Control Systems Technology, 30(3):1065–1077, 2021

    2021

  43. [52]

    Harnessing uncertainty for a separation principle in direct data- driven predictive control.Automatica, 173:112070, 2025

    Alessandro Chiuso, Marco Fabris, Valentina Breschi, and Simone For- mentin. Harnessing uncertainty for a separation principle in direct data- driven predictive control.Automatica, 173:112070, 2025

    2025

  44. [53]

    Offset–free data–driven predictive control

    Mircea Lazar and PCN Verheijen. Offset–free data–driven predictive control. In2022 IEEE 61st Conference on Decision and Control (CDC), pages 1099–1104. IEEE, 2022

    2022

  45. [54]

    Handbook of linear data-driven predictive control: Theory, implementation and design

    PCN Verheijen, Valentina Breschi, and Mircea Lazar. Handbook of linear data-driven predictive control: Theory, implementation and design. Annual Reviews in Control, 56:100914, 2023

    2023

  46. [55]

    DeePC-Hunt: Data-enabled predictive control hy- perparameter tuning via differentiable optimization.arXiv preprint arXiv:2412.06481, 2024

    Michael Cummins, Alberto Padoan, Keith Moffat, Florian Dorfler, and John Lygeros. DeePC-Hunt: Data-enabled predictive control hy- perparameter tuning via differentiable optimization.arXiv preprint arXiv:2412.06481, 2024

    2024 arXiv

  47. [56]

    Q- learning based linear quadratic regulator with balanced exploration and exploitation for unknown systems

    Xuehui Ma, Shiliang Zhang, Fucai Qian, Jinbao Wang, and Lin Yan. Q- learning based linear quadratic regulator with balanced exploration and exploitation for unknown systems. In2022 China Automation Congress (CAC), pages 446–451. IEEE, 2022

    2022

  48. [57]

    Reinforcement learning: An introduction

    Barto Andrew and Sutton Richard S. Reinforcement learning: An introduction. 2018

    2018

Pith tools

Reviewed August 7, 2026 · model on record in the stance chip above.