Pith. sign in

REVIEW 4 major objections 4 minor 35 references

Physics-Guided Spectral Parametric Reduced-Order Modeling for Transient Prediction of Controlled Dynamical Systems

T0 review · 4 major / 4 minor · reviewed 2026-08-01 · deepseek-v4-flash

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

Pith's one-line read A physics-guided reduced-order framework predicts transients at unseen parameters with sub-1% error.

desk verdict A credible extrapolation claim built on an underspecified alignment step; worth reviewing, but only with code. read the letter →

arxiv 2607.18133 v1 pith:S55PBUVZ submitted 2026-07-20 math.DS

Ao Zhang , Tian Zhang , Antonio Cammi , Xiang Wang This is my paper
classification math.DS MSC 37M9993B11
keywords parametricreduced-ordermodelingdynamicmodedecompositionwithcontrolspectraloperatortransferextrapolativetransientpredictionphysics-guidedcoordinatetransformationcontrolleddynamicalsystemsBraytoncycleKármánvortexstreet
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

This paper proposes a method for predicting the transient response of controlled dynamical systems at parameter values and operating conditions never seen in training. The key move is to identify a reduced linear model with control (DMDc), align its spectral components across sampled parameters after a physics-informed rescaling of the parameter axis, and then propagate those components as if they were a time series using a secondary DMD. On three test systems—a mechanical transmission, a closed Brayton cycle, and a Kármán vortex street—the method extrapolates beyond the sampled parameter range with relative norm errors below 1% for the first two, and preserves dominant vortex shedding for the third. The work matters because existing parametric reduced-order models typically interpolate; this one aims to extrapolate, which is what real predictive use requires.

What carries the argument

The central object is the set of aligned reduced spectral quantities — eigenvalues Ω, eigenvectors Φ, mode amplitudes b, and the factorized control operator B = U_B Σ_B V_B^T — extracted by DMDc from transient snapshots. The framework's key step is the physics-guided parameter transformation µ' = f(µ) (e.g., 1/J for inertia ratio, Reynolds number for viscosity) that makes these quantities vary more linearly along the parameter axis. Once ordered, the training samples become a pseudo-temporal sequence, and a Secondary DMD learns a propagation operator in parameter space. This lets the reduced operator be reconstructed at target parameters, even outside the sampled range, and then advanced in

What would settle it

Take a controlled dynamical system whose parameter interval contains a supercritical Hopf bifurcation, train on three sampling points on one side, and extrapolate to a point on the other side. If the predicted transient does not exhibit the new oscillatory regime (e.g., amplitude or frequency misses by more than 5%), the coherence assumption behind secondary-DMD extrapolation is falsified.

Watch

Extended reading notes

Core claim

The paper's central claim is that parameter-dependent reduced-order operators for controlled systems can be transferred across parameter values by treating the parameter direction as a pseudo-time axis. After extracting intrinsic dynamics (A) and control effects (B) from transient snapshots via DMDc, the authors align eigenvalues, eigenvectors, and amplitudes across training conditions using a physics-guided coordinate transformation µ' = f(µ). The aligned spectral quantities then evolve coherently along this axis, so a secondary DMD can propagate them to unseen parameters, including extrapolation beyond the sampled interval. Baseline regularization (subtracting steady offsets) removes opera

Load-bearing premise

The method assumes that once the parameter is rescaled in a physics-informed way, the reduced model's eigenvalues, modes, and amplitudes evolve smoothly enough along the parameter axis that extrapolating that evolution predicts new conditions; a sudden change in the system's dominant behavior (like a mode onset) in the extrapolation interval would break this.

Editorial extensions

If this is right

  • For controlled linear and nonlinear transient systems, extrapolative transient prediction at unseen parameter values can be achieved with relative norm errors below 1%, avoiding repeated high-fidelity simulations.
  • Parameter-domain Secondary DMD gives more accurate extrapolation than linear or radial-basis-function regression because it models the structured evolution of spectral quantities rather than pointwise fitting.
  • Including measured auxiliary inputs (e.g., controller outputs) in the control channel keeps intrinsic dynamics in the state operator, improving prediction under closed-loop operating scenarios.
  • Baseline regularization is necessary for extrapolation; without subtracting steady offsets, error grows substantially in the Brayton-cycle extrapolation.
  • For nonlinear periodic systems, a nondimensional time coordinate based on Strouhal number lets one reduced operator represent vortex-shedding transients across Reynolds numbers.

Reading between the lines

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

  • The reliance on a hand-crafted parameter transformation suggests that automated discovery of such coordinates (e.g., via neural networks or manifold learning) could broaden the framework to systems where physics intuition is lacking; the authors hint at this as future work.
  • The reported sub-1% errors are averaged over system-level variables; local field errors (as in the vortex-street case) may be larger, so users should carry out the paper's confidence-assessment procedure before trusting pointwise quantities.
  • Because Secondary DMD extrapolates spectral evolution, the method will likely struggle near bifurcations or regime changes; a natural test is to apply it to a system whose dominant mode changes sign within the extrapolation interval.
  • The separation of intrinsic dynamics from control via DMDc could be combined with other input-output techniques, such as balanced truncation or resolvent analysis, to extend the extrapolation claim to frequency-domain responses.
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

4 major / 4 minor

Summary. The paper proposes a physics-guided spectral parametric reduced-order modeling (pROM) framework for controlled dynamical systems. Reduced operators are identified with DMDc from transient snapshots, then spectral quantities (eigenvalues, eigenvectors, amplitudes, and factorized control operators) are aligned across parameter samples and propagated along a physics-guided parameter coordinate μ'=f(μ) by a Secondary DMD regression. The method is tested on three systems: a mechanical transmission, a Helium–Xenon closed Brayton cycle, and a Kármán vortex street. For the mechanical and Brayton cases the authors report relative norm errors below 1% for extrapolative prediction at unseen parameter values and new operating conditions; for the vortex street the method preserves dominant shedding structures. Additional comparisons with LSTM and an auxiliary error-prediction model are provided.

Significance. If the central claim holds, the paper would be a valuable step beyond interpolation-focused pROM: it explicitly targets extrapolative transient prediction by treating parameter-domain evolution of aligned spectral operators as a pseudo-temporal sequence. The paper is honest in using held-out targets, reporting rank-sensitivity analyses, and comparing with an LSTM baseline. The main risk is that the extrapolation claim rests on a spectral-alignment procedure and a low-dimensional-coherence assumption that are not fully specified or independently validated. Given the potential impact, the paper deserves revision rather than rejection, but the underspecified alignment step and the unstated treatment of steady-state offsets at target parameters are load-bearing and must be clarified.

major comments (4)
  1. [§2.2.2] The alignment step—'eigenvalue matching, phase correction, and normalization'—is described in a single sentence without an algorithm, tolerance, or treatment of ambiguous cases such as eigenvalue crossings or near-degenerate modes. Because the Secondary DMD step in §2.2.3 operates on the aligned sequence, any mismatch in this step directly corrupts the extrapolated operators. Please provide a precise algorithmic specification (or a pointer to code) for the alignment, and demonstrate its behavior on the test cases, e.g., by showing mode trajectories across the parameter axis.
  2. [§2.2.3] The core extrapolation claim requires that after the transformation μ'=f(μ), the aligned spectral quantities evolve as 'a coherent, low-dimensional sequence' along the parameter axis so that Secondary DMD can propagate them beyond the sampled interval. For the mechanical case this is supported by the M⁻¹K and 1/J structure. For the Brayton and vortex cases it is asserted from results. This assumption is load-bearing for the headline sub-1% extrapolation errors. The paper should supply a diagnostic test for coherence (e.g., evaluation of the Secondary DMD residual or a mode-tracking plot) and explicitly state what happens if a regime change or mode crossing occurs between the last training point and the farthest target, such as T_min = 400 K in §3.2.
  3. [§2.2.2 and §3.2] The baseline regularization in Eq. (3) removes steady-state offsets X_ss and β_ss, which are said to be 'regressed separately.' The paper never states how X_ss and β_ss are obtained for unseen target parameters. For the mechanical case the offsets may be trivial, but for the Brayton cycle the steady-state offsets depend on the minimum-cycle temperature and the operating scenario. If these offsets are predicted using the same Secondary DMD/regression, then the reported 'below 1%' errors include that regression, and its contribution should be reported separately. If they are known exactly, that should be stated. This is necessary to interpret the extrapolation error.
  4. [§3.1 and §3.2] The text repeatedly states that relative norm errors remain 'below 1%' in extrapolation, but the exact numerical values are not given in the body; only figures and pointwise ±3% errors at the farthest Brayton target are shown. Please provide a table with the exact relative norm errors for every target point, interpolation and extrapolation, in both scenarios, so that the sub-1% claim can be verified without reading values off plots.
minor comments (4)
  1. [§3.4.2] The LSTM comparison lacks essential details: architecture, number of layers/units, sequence length, training/validation split, and the hyperparameter search space. Without these, the comparison is difficult to reproduce and the claim of 'more accurate predictions than LSTM' should be interpreted cautiously.
  2. [§3.4.3] The confidence-assessment surrogate is described as requiring 'a few additional extrapolation points' as test points. This is a reasonable post-hoc validation approach, but it is not a deployable confidence bound for a scenario in which no test points exist. The paper should distinguish between 'validation-based error estimation' and 'online confidence estimation'; the current wording overstates the latter.
  3. [§2.3.2] The Brayton-cycle reference model is validated in previous work (Zhang and Wang, 2024; Zhang et al., 2025a). Since the current paper's sub-1% claim depends on that reference data, the paper should either summarize the validation in an appendix or state that the governing equations and validation are available only in the cited references, which limits self-containedness.
  4. [Nomenclature] Several subscripts and symbols appear garbled in the extracted text (e.g., 'X௥', '𝛽௞', '𝜇௡௘௪'). This may be a formatting artifact, but the final version should be carefully typeset. Also, the Kármán street is called 'nonlinear transient and periodic' in the abstract but 'nonlinear periodic' in §2.2.1; please make the terminology consistent.

Circularity Check

0 steps flagged · score 1.0 of 10

No significant circularity found; extrapolation targets are genuine holdouts, and self-citations are limited to benchmark infrastructure and ancillary guidance.

full rationale

Circularity would require a predicted quantity to be reused as its own fitting input, or a load-bearing appeal to a self-cited uniqueness/validity result. No such reduction is present. The three validation cases use reserved target parameters and operating scenarios: Section 3.1 states that 'target points are reserved for prediction'; Section 3.2 uses 'target points in both the original and new scenarios' to evaluate prediction under unseen parameter values and operating conditions; Section 3.3 extrapolates in Reynolds number. The self-citations (Zhang and Wang 2024; Zhang et al. 2025a) provide the Brayton reference model and prior DMD/noise discussion, i.e., benchmark infrastructure and ancillary guidance, not the pROM's fitted outputs and not a prohibition of alternatives. The physics-guided coordinate mu'=f(mu) (Eq. 6) is an assumed modeling ansatz with physical motivation; Secondary DMD (Section 2.2.3) is learned from training samples only and is genuinely extrapolative. The main risks — unspecified alignment tolerances, assumed coherent spectral evolution beyond the training interval, and steady-offset regression details — are correctness/transparency risks, not definitional circularity, because they concern extrapolative robustness rather than identity between input and output. Therefore no significant circularity is found.

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

No new physical entities are postulated. 'Secondary DMD' is a new method name (a second DMD applied along the parameter axis), not an entity; beta+ auxiliary inputs are measured closed-loop signals (PID output), not invented quantities.

free parameters (4)
  • temporal reduced rank r_t = r_t=3 (mechanical), r_t=2 (Brayton, vortex)
    Selected per case via sensitivity analysis (§3.1–§3.3). High ranks degrade extrapolation robustness; the headline errors depend on the tuned value.
  • parametric reduced rank r_p = r_p=2
    Rank of the Secondary DMD parameter-domain regression, chosen per case (Fig. 16 caption, §3.3).
  • steady-state offsets X_ss, beta_ss = regressed from training data
    Regressed from training data (Eq. 3); extrapolated to target parameters and added back to reconstruct absolute responses, so part of the 'prediction' is an extrapolated fitted offset.
  • Strouhal number St and amplitude envelope
    Identified from the reduced linear evolution per condition and extrapolated separately in the vortex case (§2.3.3); used in the nondimensional time mapping Eq. (8).
assumptions (4)
  • domain assumption A finite-dimensional linear reduced operator (A,B) identified by DMDc captures the dominant dynamics of the nonlinear systems, including beyond the training window.
    Stated in §2.1: 'the linear form of DMDc does not imply globally linear controlled dynamics... can approximate dominant nonlinear temporal behavior.' Standard DMD premise, but an assumption for extrapolation.
  • domain assumption After physics-guided transformation mu'=f(mu), aligned spectral quantities and operator components evolve as a coherent low-dimensional pseudo-temporal sequence that Secondary DMD can propagate beyond the sampled interval.
    Introduced in §2.2.3 ('the ordered training samples can be treated as a pseudo-temporal sequence'); load-bearing for extrapolation and only empirically validated on three systems.
  • domain assumption Eigenvalue matching, phase correction, and normalization yield correct modal correspondences across training conditions.
    Stated in §2.2.2 without an algorithm or error analysis; misalignment would corrupt the spectral sequence feeding Secondary DMD.
  • domain assumption The reference simulation models are valid: mechanical model matches the Modelica Standard Library example; the He-Xe Brayton model is validated in self-cited prior work; the LBM solver matches CFDBench (St=0.2188 vs 0.2219 at Re=640).
    §2.3.1–2.3.3. The Brayton validation is not independently checkable in this paper.

how reviews work

0 comments
Cite this review

Pith. "Pith review of Physics-Guided Spectral Parametric Reduced-Order Modeling for Transient Prediction of Controlled Dynamical Systems." pith.science (2026). https://pith.science/paper/S55PBUVZ

@misc{pith2026260718133,
  author       = {Pith},
  title        = {Pith review of: Physics-Guided Spectral Parametric Reduced-Order Modeling for Transient Prediction of Controlled Dynamical Systems},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/S55PBUVZ}},
  note         = {Machine review of arXiv:2607.18133}
}
read the original abstract

Efficient parametric transient prediction at unseen parameter values and under new operating conditions remains challenging because repeated high-fidelity simulations are computationally prohibitive. Existing data-driven surrogates and parametric reduced-order models perform well within sampled ranges but often lose reliability beyond them. This study proposes a physics-guided spectral parametric reduced-order modeling framework for controlled dynamical systems. Parameter-dependent reduced spectral operators are identified from transient snapshots using Dynamic Mode Decomposition with control, separating intrinsic dynamics from external control effects. After physics-guided parameter transformation, aligned spectral quantities and reduced operator components are propagated across parameter conditions using Secondary Dynamic Mode Decomposition, with linear and radial basis function regressions as baselines. Baseline regularization and nondimensional time mapping improve robustness. Validation uses a mechanical transmission system, a Helium-Xenon closed Brayton cycle, and a Karman vortex street, covering linear transient, nonlinear transient, and nonlinear periodic dynamics. For the mechanical and Brayton systems, system-level multivariable responses at unseen parameter values and under new operating conditions are predicted with relative norm errors below 1%. For the vortex-street system, nondimensional time mapping preserves dominant vortex-shedding structures across Reynolds numbers. Further analyses compare the framework with an LSTM surrogate and assess extrapolation confidence using an auxiliary error-prediction model. Overall, the framework extends parametric reduced-order modeling from in-domain approximation toward extrapolative transient prediction of controlled dynamical systems under unseen conditions.

Figures

Figures reproduced from arXiv: 2607.18133 by the authors.

Figure 1
Figure 1. Review of existing approaches and novelty positioning of this work. 1.5 Contributions and Organization To address the demand for efficient and interpretable transient prediction beyond sampled parameter values and operating conditions, this paper proposes a physics-guided spectral pROM framework. The main methodological advance lies in extending DMDc-based control-separated reduced-order modeling to parametric trans… view at source ↗
Figure 2
Figure 2. Physics-guided spectral pROM framework. 2.2.1 Dynamical Archetypes To evaluate the proposed framework under representative dynamical behaviors, the systems considered in this study are grouped into three representative dynamical archetypes based on their behaviors: linear transient systems, nonlinear transient systems, and nonlinear periodic systems.  Linear transient systems obey the superposition principle, with … view at source ↗
Figure 3
Figure 3. Modelica-based model of the mechanical transmission. 2.3.2 Nonlinear Transient System: Closed Brayton Cycle Energy-conversion systems are representative controlled dynamical systems in the energy domain, and the closed Brayton cycle is a typical example involving strong thermodynamic coupling and active load regulation. In this case, the minimum cycle temperature 𝑇௖ is treated as the system parameter defining differ… view at source ↗
Figures from the paper (17 more)
Figure 4
Figure 4. Figure 4: Modelica-based visual model of the Helium-Xenon closed Brayton cycle [PITH_FULL_IMAGE:figures/full_fig_p012_4.png]
Figure 5
Figure 5. Figure 5: Geometric structure of the Kármán vortex street (Luo et al., 2024). Owing to its transient periodic nature, a nondimensional time mapping (Eq. (4)) is introduced to reduce the inconsistency of the identified spectral operator caused by frequency variation. For the [PI…
Figure 6
Figure 6. Figure 6: Setup of the scenarios and parameter points [PITH_FULL_IMAGE:figures/full_fig_p015_6.png]
Figure 7
Figure 7. Figure 7: Prediction error comparison of pROMs using different regression methods [PITH_FULL_IMAGE:figures/full_fig_p016_7.png]
Figure 8
Figure 8. Figure 8: Time-domain prediction of a representative system variable (𝑟௧ = 3). 3.2 Closed Brayton Cycle: Load Transients under Minimum-Cycle-Temperature Variation For the Modelica-based Helium-Xenon closed Brayton cycle, two operating scenarios are constructed, and the parameter…
Figure 9
Figure 9. Figure 9: Setup of the scenarios and parameter points [PITH_FULL_IMAGE:figures/full_fig_p017_9.png]
Figure 10
Figure 10. Figure 10: Prediction error comparison of pROMs using different regression methods [PITH_FULL_IMAGE:figures/full_fig_p018_10.png]
Figure 11
Figure 11. Figure 11: Time-resolved relative errors of all predicted system variables at 𝑇௖ = 400 K (𝑟௧,௣ = 2) [PITH_FULL_IMAGE:figures/full_fig_p018_11.png]
Figure 12
Figure 12. Figure 12: Time-domain prediction of a representative system variable (𝑟௧ = 2) [PITH_FULL_IMAGE:figures/full_fig_p019_12.png]
Figure 13
Figure 13. Figure 13: Effect of measured auxiliary inputs 𝛽 ା (𝑟௧ = 2). 3.3 Kármán Vortex Street: Inlet-Velocity Transients under Fluid-Viscosity Variation In the Kármán vortex-street case, the reduced-order model is constructed from full-field velocity snapshots of the cylinder wake under…
Figure 14
Figure 14. Figure 14: Setup of the scenarios and parameter points [PITH_FULL_IMAGE:figures/full_fig_p020_14.png]
Figure 15
Figure 15. Figure 15: Effect of nondimensional time mapping on transient periodic prediction [PITH_FULL_IMAGE:figures/full_fig_p020_15.png]
Figure 16
Figure 16. Figure 16: Prediction error comparison of pROMs using different regression methods [PITH_FULL_IMAGE:figures/full_fig_p021_16.png]
Figure 17
Figure 17. Figure 17: Time-domain prediction of the Kármán vortex street (Secondary DMD, 𝑟௣ = 2) [PITH_FULL_IMAGE:figures/full_fig_p021_17.png]
Figure 18
Figure 18. Figure 18: Predictions of velocity magnitude at selected probes. 3.4 Further Discussion Building on the predictive capability demonstrated across linear, nonlinear, and periodic archetypes, this section further discusses the data requirements of the proposed physics-guided spect…
Figure 19
Figure 19. Figure 19: Comparison between pROM and LSTM [PITH_FULL_IMAGE:figures/full_fig_p023_19.png]
Figure 20
Figure 20. Figure 20: Extrapolation confidence assessment for the Helium-Xenon closed Brayton cycle (Secondary DMD, 𝑟௧,௣ = 2). 4. Conclusions This paper establishes a DMDc-centered physics-guided spectral pROM framework for interpretable system identification and extrapolative transient pr…

Discussion (0). Sign in to comment.

Reference graph

Works this paper leans on

35 extracted references · 10 canonical work pages

  1. [8]

    Presented at the The American Modelica Conference 2018, October 9-10, Somberg Conference Center, Cambridge MA, USA, pp

    The OpenModelica Integrated Modeling, Simulation, and Optimization Environment. Presented at the The American Modelica Conference 2018, October 9-10, Somberg Conference Center, Cambridge MA, USA, pp. 206–219. https://doi.org/10.3384/ecp18154206 Gabbai, R.D., Benaroya, H.,

    work page doi:10.3384/ecp18154206 2018

  2. [11]

    Progress in Nuclear Energy 183, 105684

    Fast prediction and inverse modeling for parameterized depressurization from a CO2 vessel using POD-BPNN reduced-order method. Progress in Nuclear Energy 183, 105684. https://doi.org/10.1016/j.pnucene.2025.105684 Krüger, T., Kusumaatmaja, H., Kuzmin, A., Shardt, O., Silva, G., Viggen, E.M.,

    arXiv 2025

  3. [12]

    Springer International Publishing, Cham

    The Lattice Boltzmann Method: Principles and Practice, Graduate Texts in Physics. Springer International Publishing, Cham. https://doi.org/10.1007/978-3-319-44649-3 Ljung, L., Andersson, C., Tiels, K., Schön, T.B.,

    doi:10.1007/978-3-319-44649-3

  4. [13]

    IFAC-PapersOnLine 53, 1175–1181

    Deep Learning and System Identification. IFAC-PapersOnLine 53, 1175–1181. https://doi.org/10.1016/j.ifacol.2020.12.1329 Luo, Y ., Chen, Y ., Zhang, Z.,

    work page doi:10.1016/j.ifacol.2020.12.1329 2020

  5. [14]

    https://doi.org/10.48550/arXiv.2310.05963 Lusch, B., Kutz, J.N., Brunton, S.L.,

    CFDBench: A Large-Scale Benchmark for Machine Learning Methods in Fluid Dynamics. https://doi.org/10.48550/arXiv.2310.05963 Lusch, B., Kutz, J.N., Brunton, S.L.,

    arXiv doi:10.48550/arxiv.2310.05963

  6. [16]

    Progress in Nuclear Energy 191, 106032

    A novel reduced-order model and two-stage optimization strategy for high-efficiency predicting thermal-hydraulic behavior in a 6×6 fuel bundle. Progress in Nuclear Energy 191, 106032. https://doi.org/10.1016/j.pnucene.2025.106032 Mo, J., Dong, X., Xu, Y ., Xu, A., Yu, Y ., Yang, M.,

    arXiv 2025

  7. [17]

    Progress in Nuclear Energy 184, 105723

    Research on fast prediction of boiling flow parameters in rod bundle of NPP based on efficient CFD-ROM methods. Progress in Nuclear Energy 184, 105723. https://doi.org/10.1016/j.pnucene.2025.105723 Muñoz, M.C., Peñalba, M.A., González, A.E.S.,

    arXiv 2025

  8. [18]

    Energy Reports 12, 1000–1013

    Analysis of aggregated load consumption forecasting in short, medium and long term horizons using Dynamic Mode Decomposition. Energy Reports 12, 1000–1013. https://doi.org/10.1016/j.egyr.2024.06.040 Proctor, J.L., Brunton, S.L., Kutz, J.N.,

    work page doi:10.1016/j.egyr.2024.06.040 2024

Show all 35 references
  1. [20]

    Energy Conversion and Management 326, 119462

    Review of system design and operation control technology of supercritical CO2 power cycle. Energy Conversion and Management 326, 119462. https://doi.org/10.1016/j.enconman.2024.119462 Riva, S., Missaglia, A., Introini, C., Bang, I.C., Cammi, A.,

    2024

  2. [21]

    Nuclear Technology 1–25

    A Comparative Study of Parametric Dynamic Mode Decomposition Algorithms on Thermal-Hydraulics Case Studies. Nuclear Technology 1–25. https://doi.org/10.1080/00295450.2025.2575701 Sayadi, T., Schmid, P.J., Richecoeur, F., Durox, D.,

    2025

  3. [22]

    Physics of Fluids 27, 037102

    Parametrized data-driven decomposition for bifurcation analysis, with application to thermo-acoustically unstable systems. Physics of Fluids 27, 037102. https://doi.org/10.1063/1.4913868 Schmid, P.J.,

    doi

  4. [28]

    Sci Rep 15, 23489

    Dynamic mode decomposition for analysis and prediction of metabolic oscillations from time-lapse imaging of cellular autofluorescence. Sci Rep 15, 23489. https://doi.org/10.1038/s41598-025-07255-4 Xie, H.-R., Hua, Y ., Li, Y .-B., Aubry, N., Wu, W.-T., He, Y ., Peng, J.-Z.,

    doi

  5. [29]

    Ocean Engineering 293, 116684

    Estimation of sequential transient flow around cylinders using recurrent neural network coupled graph convolutional network. Ocean Engineering 293, 116684. https://doi.org/10.1016/j.oceaneng.2024.116684 Yu, Y ., Zhang, Y ., Qian, S., Wang, S., Hu, Y ., Yin, B.,

    2024

  6. [31]

    Progress in Nuclear Energy 172, 105205

    Development of Modelica-based one-dimensional thermodynamic cycle library and its application in simulation and multi-objective optimization of a He–Xe closed- Brayton-cycle system. Progress in Nuclear Energy 172, 105205. https://doi.org/10.1016/j.pnucene.2024.105205 Zhang, K....

    2024

  7. [32]

    Applied Thermal Engineering 289, 129763

    Surrogate-assisted dynamic response prediction and intelligent fault management for gas–cooled reactor–Brayton energy systems. Applied Thermal Engineering 289, 129763. https://doi.org/10.1016/j.applthermaleng.2026.129763 Zhang, Y ., Li, H., Li, K., Yang, Y ., Zhou, Y ., Zhang,...

    2026

  8. [33]

    Applied Energy 328, 120190

    Dynamic characteristics and control strategies of the supercritical CO2 Brayton cycle tailored for the new generation concentrating solar power. Applied Energy 328, 120190. https://doi.org/10.1016/j.apenergy.2022.120190 Zhang, Y ., Li, P., Qiao, H., 2025b. Fast prediction of t...

    2022

  9. [34]

    Energy 275, 127525

    High-fidelity wind turbine wake velocity prediction by surrogate model based on d-POD and LSTM. Energy 275, 127525. https://doi.org/10.1016/j.energy.2023.127525 Zhu, M., Zhang, H., Jiao, A., Karniadakis, G.E., Lu, L.,

    2023

  10. [35]

    Computer Methods in Applied Mechanics and Engineering 412, 116064

    Reliable extrapolation of deep neural operators informed by physics or sparse observations. Computer Methods in Applied Mechanics and Engineering 412, 116064. https://doi.org/10.1016/j.cma.2023.116064

    2023

  11. [254]

    https://doi.org/10.1146/annurev-fluid-030121-015835 Schmid, P.J.,

    doi

  12. [2005]

    Journal of Sound and Vibration 282, 575–616

    An overview of modeling and experiments of vortex-induced vibration of circular cylinders. Journal of Sound and Vibration 282, 575–616. https://doi.org/10.1016/j.jsv.2004.04.017 Huhn, Q.A., Tano, M.E., Ragusa, J.C., Choi, Y .,

    2004 doi

  13. [2010]

    Dynamic mode decomposition of numerical and experimental data. J. Fluid Mech. 656, 5–28. https://doi.org/10.1017/S0022112010001217 Sutton, C., Boley, M., Ghiringhelli, L.M., Rupp, M., Vreeken, J., Scheffler, M.,

    doi

  14. [2015]

    SIAM Rev

    A Survey of Projection-Based Model Reduction Methods for Parametric Dynamical Systems. SIAM Rev. 57, 483–531. https://doi.org/10.1137/130932715 Brunton, S.L., Kutz, J.N.,

    doi

  15. [2016]

    Dynamic Mode Decomposition with Control. SIAM J. Appl. Dyn. Syst. 15, 142–161. https://doi.org/10.1137/15M1013857 Qin, T., Yan, X., Yuan, C., Sun, Y .,

    doi

  16. [2017]

    Energy 125, 643–653

    Multi-objective optimization of the design and operating point of a new external axis wind turbine. Energy 125, 643–653. https://doi.org/10.1016/j.energy.2017.01.070 Fritzson, P., Pop, A., Asghar, A., Bachmann, B., Braun, W., Braun, R., Buffoni, L., Casella, F., Castro, R., Da...

    2017 doi

  17. [2018]

    Applied Energy 212, 109–121

    A systematic comparison of different S-CO2 Brayton cycle layouts based on multi-objective optimization for applications in solar power tower plants. Applied Energy 212, 109–121. https://doi.org/10.1016/j.apenergy.2017.12.031 Wüstner, D., Gundestrup, H.H., Thaysen, K.,

    2017 doi

  18. [2019]

    Cambridge University Press

    Data-Driven Science and Engineering: Machine Learning, Dynamical Systems, and Control, 1st ed. Cambridge University Press. https://doi.org/10.1017/9781108380690 Brunton, S.L., Noack, B.R., Koumoutsakos, P.,

    doi

  19. [2020]

    Annual Review of Fluid Mechanics

    Machine Learning for Fluid Mechanics. Annual Review of Fluid Mechanics. https://doi.org/10.1146/annurev-fluid-010719-060214 Chen, G., Qi, B., Hu, W., Zhang, Y ., Wei, J.,

    doi

  20. [2021]

    IEEE Trans

    A Low Rank Dynamic Mode Decomposition Model for Short-Term Traffic Flow Prediction. IEEE Trans. Intell. Transport. Syst. 22, 6547–6560. https://doi.org/10.1109/TITS.2020.2994910 Zhang, A., Cammi, A., Wang, X., 2025a. System-level data-driven identification of complex closed Br...

    2020

  21. [2022]

    IEEE Trans

    Developing a Ball Screw Drive System of High-Speed Machine Tool Considering Dynamics. IEEE Trans. Ind. Electron. 69, 4966–4976. https://doi.org/10.1109/TIE.2021.3083200 Wang, K., Li, M.-J., Guo, J.-Q., Li, P., Liu, Z.-B.,

    2021

  22. [2023]

    Journal of Computational Physics 475, 111852

    Parametric dynamic mode decomposition for reduced order modeling. Journal of Computational Physics 475, 111852. https://doi.org/10.1016/j.jcp.2022.111852 Kang, Y ., Tian, G., Huang, Y ., Qin, S., Huang, Q., Yuan, Y ., Zhou, Y .,

    2022

  23. [2024]

    International Journal of Hydrogen Energy 49, 1391–1404

    A fast POD prediction method for hydrogen leakage at different pressures. International Journal of Hydrogen Energy 49, 1391–1404. https://doi.org/10.1016/j.ijhydene.2023.09.282 Du, L., Zhang, Shengqi, Zhang, R., Zhang, Shibin,

    2023 doi

  24. [2025]

    Energy 335, 138047

    Data-driven modeling and prediction of PEM fuel cell voltage response to load transients for energy applications. Energy 335, 138047. https://doi.org/10.1016/j.energy.2025.138047 Ferdoues, M.S., Ebrahimi, S., Vijayaraghavan, K.,

    2025

  25. [2026]

    Journal of Computational Physics 544, 114436

    Interpolation-based parametric reduced-order models with dynamic mode decomposition. Journal of Computational Physics 544, 114436. https://doi.org/10.1016/j.jcp.2025.114436 Elkholy, M., Boureima, A., Kim, J., Aziz, M.,

    2025

  26. [4428]

    https://doi.org/10.1038/s41467-020-17112-9 Wang, D., Zhang, S., Wang, L., Liu, Y .,

    doi

  27. [4950]

    https://doi.org/10.1038/s41467-018-07210-0 Min, G., Shen, X., Wang, L., Jiang, N.,

    doi

Pith tools

Reviewed August 1, 2026 · model on record in the stance chip above.