Pith. sign in

REVIEW 4 major objections 5 minor 1 cited by

Improving Low-Fidelity Models of Li-ion Batteries via Hybrid Sparse Identification of Nonlinear Dynamics

T0 review · 4 major / 5 minor · reviewed 2026-08-12 · deepseek-v4-flash

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

Pith's one-line read The paper claims that a low-fidelity Li-ion battery model can be made nearly as accurate as a high-fidelity model by adding a sparse, data-driven correction for its voltage error, and demonstrates this across multiple driving cycles.

desk verdict Useful incremental method for correcting reduced-order battery models, but the pivotal implementation detail—recursive vs teacher-forced error propagation—is unstated, so the headline accuracy numbers are hard to interpret. read the letter →

arxiv 2411.12935 v1 pith:MUX5XG54 submitted 2024-11-20 eess.SY cs.LGcs.NEcs.SY

Samuel Filgueira da Silva , Mehmet Fatih Ozkan , Faissal El Idrissi , Prashanth Ramesh , Marcello Canova This is my paper · ORCID
classification eess.SYcs.LGcs.NEcs.SY
keywords Li-ionbatterymodelingreduced-ordermodelsparseidentificationgeneticalgorithmvoltageerrorcorrectiondrivingcycleselectrochemicalhybrid
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 tries to show that a fast, reduced-order battery model can be made nearly as accurate as a full electrochemical model by learning a sparse correction for its voltage error. The correction is a dynamical function of the low-fidelity model's states, the applied current, and the previous error, discovered with genetic-algorithm-guided sparse regression. On multiple driving cycles, the augmented model reduces voltage root-mean-square error by 53% to 66% relative to the baseline low-fidelity model, while cutting computation time by 51% to 63% relative to the high-fidelity model. A sympathetic reader would care because it offers a path to real-time-accurate battery simulation without abandoning physics-based structure.

What carries the argument

The load-bearing object is the sparse error-dynamics model of equation (11), a one-step-ahead predictor of the voltage error as a linear combination of basis functions built from the previous error, the applied current, and the low-fidelity model's solid-phase surface concentrations. The identification procedure, GA-STRidge, couples a genetic algorithm to sequentially thresholded ridge regression: the genetic algorithm selects which basis functions enter the library and sets the two regularization thresholds, while the ridge regression fits and sparsifies the coefficient vector. This machinery augments the low-fidelity equations with a small, interpretable correction without requiring re-derivation of the electrochemical model.

What would settle it

Train the same genetic-algorithm sparse regression on the same low-fidelity model but with the 'truth' data coming from physical experiments on a real cell instead of the high-fidelity virtual battery, then compare voltage RMSE on the same driving cycles; if the correction no longer yields a comparable error reduction (roughly 53–66%), the sparse error-map assumption fails when the high-fidelity model itself is imperfect.

Watch

Extended reading notes

Core claim

The central claim is that the discrepancy between a low-fidelity and a high-fidelity Li-ion battery model can be represented by a sparse, interpretable dynamical map of the voltage error, learned from data rather than derived from first principles. Specifically, the paper models the one-step-ahead error as $\hat{e}_r[k+1] = \Theta(e_r[k], I[k], c_{s,p}[k], c_{s,n}[k])\xi$, where $\Theta$ is a library of candidate basis functions chosen by a genetic algorithm and $\xi$ is a sparse coefficient vector obtained by sequentially thresholded ridge regression. The hybrid model built this way reproduces the voltage behavior of the high-fidelity model substantially better than the low-fidelity model alone, with a relative RMSE reduction between 53.06% and 66.64% over validation and test cycles.

Load-bearing premise

The load-bearing premise is that the one-step-ahead voltage error can be captured accurately as a sparse function of the low-fidelity model's states, the current, and the previous error, an assumption checked only against the authors' high-fidelity model and not against experimental cell data.

Editorial extensions

If this is right

  • Battery management systems could run near-high-fidelity voltage prediction at a fraction of the computational cost, enabling more accurate state-of-charge and power-limit estimation in real time.
  • The learned correction terms are explicit functions of physical states, so the same procedure can identify which approximations in a reduced-order model dominate the error.
  • The automated selection of basis functions removes a key manual step in sparse-regression identification, making the method easier to apply to other reduced-order models.
  • Because the method was validated on unseen driving cycles, the learned correction appears to generalize across current profiles rather than merely memorizing the training sequence.

Reading between the lines

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

  • If the approach transfers from virtual high-fidelity data to real cell measurements, it could turn cheap reduced-order models into accurate digital twins for aging and diagnostics.
  • The recurrence of the error state in the discovered model suggests the correction acts like a lightweight observer; a natural extension is testing closed-loop stability when the hybrid model is used in feedback control.
  • The singular-value ranking points to solid-phase concentration terms as the main error drivers, so a physical interpretation is that the reduced-order model's diffusion approximation is the largest source of discrepancy; this could guide targeted improvements to the model reduction itself.
  • Because the error map includes the previous error as an input, training data must sufficiently excite the error's own dynamics; profiles with very different frequency content than the training set may require re-identification.
Share X Bluesky LinkedIn Reddit HN

Signed reviews

No signed human review yet.

Editorial analysis

A structured set of objections, weighed in public.

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

Referee Report

4 major / 5 minor

Summary. The paper proposes a hybrid modeling approach for Li-ion batteries in which a sparse nonlinear autoregressive error model, identified via a Genetic Algorithm guided Sequentially Thresholded Ridge Regression (GA-STRidge), predicts the voltage discrepancy between a low-fidelity electrochemical-equivalent-circuit model (LFM) and a high-fidelity extended single particle model (HFM). The learned error dynamics (Eq. 11) is intended to be added to the LFM terminal voltage, and the approach is evaluated on several driving cycles. The authors report 53-66% RMSE reduction relative to the LFM and 51-63% computation-time reduction versus the HFM, and analyze the selected basis functions via SVD ranking.

Significance. If the evaluation protocol is a standalone recursive simulation, the result is a practically useful and interpretable method for augmenting reduced-order battery models, with promising generalization across unseen cycles. The use of GA for library selection and SVD for feature interpretability is a methodological strength, as is the multi-cycle validation. However, because the training and evaluation data come from the same HFM, the paper does not demonstrate physical fidelity to a real cell; the reported improvements are relative to the virtual battery and therefore should be interpreted as a demonstration of the method's internal consistency rather than of real-cell predictive accuracy.

major comments (4)
  1. [Section IV-B, Eq. (11)] The manuscript does not state whether the validation and testing results in Fig. 3b and Table II were obtained by recursively simulating Eq. (11) from er[0]=0 with predicted errors fed back, or by teacher-forcing the true er[k] from the HFM at every sample. This distinction is critical: under teacher-forcing, the 'hybrid model' would require the HFM voltage at the previous time step, so the reported RRR and computational-time reductions would not describe a deployable standalone model; under recursive simulation, the one-step training loss (16) does not optimize multi-step rollout accuracy and no stability or error-accumulation analysis is provided. The authors should state the protocol explicitly, and if the results are recursive, add a multi-step stability analysis and report results from both modes.
  2. [Algorithm 1] The coefficient update line in Algorithm 1 contains a typo: the expression (Θ_lc^T Θ_lc + λ1 I)^{-1} Θ_lc^{-1} er should instead be (Θ_lc^T Θ_lc + λ1 I)^{-1} Θ_lc^T er. In addition, the regression target in Algorithm 1 is written as er[k] while Eq. (11) and Eq. (14) define the target as er[k+1]; this inconsistency must be fixed for reproducibility.
  3. [Section IV-B, Table II] The computational-efficiency claim is supported only by percentage reductions relative to the HFM, with no absolute computation times reported. Since the hybrid model's runtime is described as 'slightly higher' than the LFM but no quantitative baseline is given, the reader cannot assess whether the 51-63% time reduction is meaningful in absolute terms or how it scales with simulation horizon. Please report wall-clock times (or per-step times) for all three models.
  4. [Sections III and IV] All results are obtained with the HFM as a virtual battery; no experimental cell data are used. The abstract states that data are 'generated either from testing or a high-fidelity model,' but the experiments use only HFM data. Because the training data and evaluation target come from the same HFM, the reported RMSE reductions partly measure the quality of the error-model fit on the training distribution. The authors acknowledge experimental extension in the conclusions, but the current claims of robustness under various operating conditions and of physical fidelity should be tempered, or an experimental validation should be added.
minor comments (5)
  1. [Eq. (1)] There is a stray '+' before the minus sign in the overpotential term (ηp(cs,p,t)+ −ηn(cs,n,t)); please correct the formatting.
  2. [Section IV-C, Eq. (20)] The matrix S is described with columns representing predictors, but the SVD analysis would be clearer if the dimensions and the construction of the weighted feature matrix were stated explicitly.
  3. [Fig. 4] The cumulative information curve is described as ranking features, but the figure's horizontal axis is 'Rank' and the curve is not monotonic in the text; please clarify how the ranking is computed and why the cumulative quantity can decrease.
  4. [General] The paper does not mention data or code availability; adding a statement would improve reproducibility.
  5. [Eq. (16)] The use of '...' in the displayed loss function is nonstandard; write the two MSE terms and the complexity term as a single expression.

Circularity Check

0 steps flagged · score 1.0 of 10

No significant circularity: the learned error correction is tested on held-out cycles, and the self-cited LFM/HFM only define the benchmark.

full rationale

The reported improvement is not equivalent to its own inputs by construction. The error signal e_r = V_HFM - V_LFM (Eq. 10) is a supervised target; the hybrid model adds a sparse regression \hat e_r (Eq. 11) whose coefficients are fit by GA-STRidge on a cascade of training driving cycles and, via the validation term in Eq. (16), on a WLTC validation profile. Table II then evaluates LFM + \hat e_r against the same HFM on WLTC, Artemis Urban, JC08, and SC03. Because the GA loss (Eq. 16) includes the validation error, WLTC is not a fully independent holdout, and the JC08/SC03/Artemis cycles are the genuinely unseen benchmarks; this weakens the strength of the generalization claim but does not make the result definitionally equal to the training fit. The RRR formula (Eq. 18) is algebraically 1 - RMSE(e_r - \hat e_r)/RMSE(e_r), so the reported voltage improvement is exactly the error model's generalization quality, which is the appropriate metric for a data-driven correction. The autoregressive term e_r[k] in Eq. (11), together with the mismatch between Eq. (14) (e_r[k+1]) and Algorithm 1 (e_r[k]), creates a genuine ambiguity about whether Table II is a closed-loop rollout or a teacher-forced one-step fit; that is a correctness risk affecting the computational-time comparison, not a circular reduction. The LFM [19] and HFM [20] come from prior work with overlapping authors, but those citations supply the benchmark models rather than evidence for the identified correction, so the self-citation is not load-bearing. In sum, no prediction in the paper reduces to its input by construction.

Assumptions & free parameters 5 free parameters · 4 assumptions · 0 invented entities

The method introduces no new physical entities. Its free parameters are the GA-selected sparsification and regularization coefficients plus the sparse model coefficients, none of which are reported numerically. The central assumption is that a sparse nonlinear map of LFM states and current captures the model discrepancy, and that the LFM and HFM from previous work are valid base models.

free parameters (5)
  • lambda_1 (regularization) = not reported
    L2 regularization coefficient in Eq. (14), optimized by GA; value not disclosed.
  • lambda_2 (sparsification threshold) = not reported
    Coefficient magnitude threshold in Eq. (14), optimized by GA; value not disclosed.
  • GA hyperparameters = not reported
    Population size, number of generations, crossover and mutation rates are not specified, so the optimization cannot be replicated.
  • Training MSE constraint epsilon = not reported
    Constraint in Eq. (17) used as a stop criterion in the GA; value not provided.
  • Sparse coefficients xi = not reported
    The 32 active basis-function coefficients are fit by STRidge, but the full coefficient vector is not given in the paper.
assumptions (4)
  • domain assumption The low-fidelity model equations from [19] and high-fidelity ESPM from [20] are accurate enough to serve as base and reference models.
    The entire correction is trained against the HFM; if the HFM is inaccurate, the corrected model inherits its errors. These models are the authors' own prior work.
  • ad hoc to paper The voltage error depends only on current I, LFM states cs,p and cs,n, and previous error er.
    Introduced in Eq. (11) without derivation; other states or temperature could matter.
  • domain assumption Sparse regression with GA-selected basis functions generalizes to unseen driving cycles.
    Central methodological premise tested on a few cycles but without theoretical guarantee.
  • standard math Standard linear algebra and least-squares solutions apply.
    Used in STRidge update in Algorithm 1; standard.

how reviews work

0 comments
Cite this review

Pith. "Pith review of Improving Low-Fidelity Models of Li-ion Batteries via Hybrid Sparse Identification of Nonlinear Dynamics." pith.science (2026). https://pith.science/paper/MUX5XG54

@misc{pith2026241112935,
  author       = {Pith},
  title        = {Pith review of: Improving Low-Fidelity Models of Li-ion Batteries via Hybrid Sparse Identification of Nonlinear Dynamics},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/MUX5XG54}},
  note         = {Machine review of arXiv:2411.12935}
}
read the original abstract

Accurate modeling of lithium ion (li-ion) batteries is essential for enhancing the safety, and efficiency of electric vehicles and renewable energy systems. This paper presents a data-inspired approach for improving the fidelity of reduced-order li-ion battery models. The proposed method combines a Genetic Algorithm with Sequentially Thresholded Ridge Regression (GA-STRidge) to identify and compensate for discrepancies between a low-fidelity model (LFM) and data generated either from testing or a high-fidelity model (HFM). The hybrid model, combining physics-based and data-driven methods, is tested across different driving cycles to demonstrate the ability to significantly reduce the voltage prediction error compared to the baseline LFM, while preserving computational efficiency. The model robustness is also evaluated under various operating conditions, showing low prediction errors and high Pearson correlation coefficients for terminal voltage in unseen environments.

Figures

Figures reproduced from arXiv: 2411.12935 by the authors.

Figure 1
Figure 1. Schematic of the GA-STRidge algorithm. IV. RESULTS AND DISCUSSION A. Training and Validation Methodology The current profile in the training dataset is shown in [PITH_FULL_IMAGE:figures/full_fig_p003_1.png] view at source ↗
Figure 3
Figure 3. Predicted model performance under training and validation tests. [PITH_FULL_IMAGE:figures/full_fig_p005_3.png] view at source ↗
Figure 4
Figure 4. SVD ranking of the discovered basis function by GA-STRidge [PITH_FULL_IMAGE:figures/full_fig_p006_4.png] view at source ↗

Discussion (0). Continue with ORCID to comment.

Forward citations

Cited by 1 Pith paper

Reviewed papers in the Pith corpus that reference this work. Sorted by Pith novelty score. Full citation record

  1. Augmented Physics-Based Li-ion Battery Model via Adaptive Ensemble Sparse Learning and Conformal Prediction

    eess.SY 2025-07 conditional novelty 5.0 of 10

    A hybrid battery model that learns sparse corrections to a reduced-order physics model reduces unseen-cycle voltage MSE by up to 46%, with conformal intervals above 90% coverage.

Reference graph

Works this paper leans on

21 extracted references · 14 canonical work pages · cited by 1 Pith paper

  1. [1]

    A review of lithium-ion battery for electric vehicle applications and beyond,

    W. Chen, J. Liang, Z. Yang, and G. Li, “A review of lithium-ion battery for electric vehicle applications and beyond,” Energy Procedia, vol. 158, pp. 4363–4368, 2019. Innovative Solutions for Energy Transitions

    work page 2019

  2. [2]

    Batteries and fuel cells for emerging electric vehicle markets,

    Z. P. Cano, D. Banham, S. Ye, A. Hintennach, J. Lu, M. Fowler, and Z. Chen, “Batteries and fuel cells for emerging electric vehicle markets,” Nature Energy, vol. 3, no. 4, pp. 279–289, 2018

    2018

  3. [3]

    A comparison between physics-based li-ion battery models,

    H. A. A. Ali, L. H. Raijmakers, K. Chayambuka, D. L. Danilov, P. H. Notten, and R.-A. Eichel, “A comparison between physics-based li-ion battery models,” Electrochimica Acta, vol. 493, p. 144360, 2024

    2024

  4. [4]

    Lithium-ion battery multi-scale modeling coupled with simplified electrochemical model and kinetic monte carlo model,

    H. Yu, L. Zhang, W. Wang, K. Yang, Z. Zhang, X. Liang, S. Chen, S. Yang, J. Li, and X. Liu, “Lithium-ion battery multi-scale modeling coupled with simplified electrochemical model and kinetic monte carlo model,” iScience, vol. 26, no. 9, p. 107661, 2023

    2023

  5. [5]

    Evaluation and observability analysis of an improved reduced-order electrochemical model for lithium-ion battery,

    L. Wu, K. Liu, and H. Pang, “Evaluation and observability analysis of an improved reduced-order electrochemical model for lithium-ion battery,” Electrochimica Acta, vol. 368, p. 137604, 2021

    work page 2021

  6. [6]

    Integrating electrochemical modeling with machine learning for lithium-ion batteries,

    H. Tu, S. Moura, and H. Fang, “Integrating electrochemical modeling with machine learning for lithium-ion batteries,” in 2021 American Control Conference (ACC), pp. 4401–4407, 2021

    work page 2021

  7. [7]

    State-of- charge estimation of li-ion batteries using deep neural networks: A machine learning approach,

    E. Chemali, P. J. Kollmeyer, M. Preindl, and A. Emadi, “State-of- charge estimation of li-ion batteries using deep neural networks: A machine learning approach,” Journal of Power Sources , vol. 400, pp. 242–255, 2018

    work page 2018

  8. [8]

    State-of-charge estimation of lithium-ion batteries using lstm and ukf,

    F. Yang, S. Zhang, W. Li, and Q. Miao, “State-of-charge estimation of lithium-ion batteries using lstm and ukf,” Energy, vol. 201, p. 117664, 2020

    work page 2020

Show all 21 references
  1. [9]

    Dis- covering governing equations of li-ion batteries pertaining state of charge using input-output data,

    R. Rodriguez, O. Ahmadzadeh, Y . Wang, and D. Soudbakhsh, “Dis- covering governing equations of li-ion batteries pertaining state of charge using input-output data,” in 2023 American Control Conference (ACC), pp. 3081–3086, IEEE, 2023

    2023

  2. [10]

    Data driven battery modeling and management method with aging phenomenon considered,

    S. Li, H. He, C. Su, and P. Zhao, “Data driven battery modeling and management method with aging phenomenon considered,” Applied Energy, vol. 275, p. 115340, 2020

    2020

  3. [11]

    Data-driven nonparametric li-ion battery ageing model aiming at learning from real operation data - part b: Cycling operation,

    M. Lucu, E. Martinez-Laserna, I. Gandiaga, K. Liu, H. Camblong, W. Widanage, and J. Marco, “Data-driven nonparametric li-ion battery ageing model aiming at learning from real operation data - part b: Cycling operation,” Journal of Energy Storage , vol. 30, p. 101410, 2020

    2020

  4. [12]

    A data- driven approach with uncertainty quantification for predicting future capacities and remaining useful life of lithium-ion battery,

    K. Liu, Y . Shang, Q. Ouyang, and W. D. Widanage, “A data- driven approach with uncertainty quantification for predicting future capacities and remaining useful life of lithium-ion battery,” IEEE Transactions on Industrial Electronics, vol. 68, no. 4, pp. 3170–3180, 2021

    2021

  5. [13]

    A hybrid data-driven approach for multistep ahead prediction of state of health and remaining useful life of lithium-ion batteries,

    M. U. Ali, A. Zafar, H. Masood, K. D. Kallu, M. A. Khan, U. Tariq, Y . J. Kim, and B. Chang, “A hybrid data-driven approach for multistep ahead prediction of state of health and remaining useful life of lithium-ion batteries,” Computational Intelligence and Neuroscience , vol....

    2022

  6. [14]

    Sparse identification of nonlinear dynamics with control (SINDYc),

    S. L. Brunton, J. L. Proctor, and J. N. Kutz, “Sparse identification of nonlinear dynamics with control (SINDYc),” IFAC-PapersOnLine, vol. 49, no. 18, pp. 710–715, 2016. 10th IFAC Symposium on Nonlinear Control Systems NOLCOS 2016

    2016

  7. [15]

    Ai feynman: A physics-inspired method for symbolic regression,

    S.-M. Udrescu and M. Tegmark, “Ai feynman: A physics-inspired method for symbolic regression,” Science Advances , vol. 6, no. 16, p. eaay2631, 2020

    2020

  8. [16]

    Contemporary symbolic regression methods and their relative performance,

    W. La Cava, B. Burlacu, M. Virgolin, M. Kommenda, P. Orzechowski, F. O. de Franc ¸a, Y . Jin, and J. H. Moore, “Contemporary symbolic regression methods and their relative performance,” Advances in neural information processing systems, vol. 2021, no. DB1, p. 1, 2021

    2021

  9. [17]

    A physics-inspired machine learning nonlinear model of li-ion batteries,

    O. Ahmadzadeh, R. Rodriguez, Y . Wang, and D. Soudbakhsh, “A physics-inspired machine learning nonlinear model of li-ion batteries,” in 2023 American Control Conference (ACC) , pp. 3087–3092, IEEE, 2023

    2023

  10. [18]

    A data-driven frame- work for learning governing equations of li-ion batteries and co- estimating voltage and state-of-charge,

    O. Ahmadzadeh, Y . Wang, and D. Soudbakhsh, “A data-driven frame- work for learning governing equations of li-ion batteries and co- estimating voltage and state-of-charge,” Journal of Energy Storage , vol. 84, p. 110743, 2024

    2024

  11. [19]

    Physics-based equivalent circuit model for lithium-ion cells via reduction and ap- proximation of electrochemical model,

    D. Seals, P. Ramesh, M. D’Arpino, and M. Canova, “Physics-based equivalent circuit model for lithium-ion cells via reduction and ap- proximation of electrochemical model,” SAE International Journal of Advances and Current Practices in Mobility, vol. 4, no. 2022-01-0701, pp. 11...

    2022

  12. [20]

    Modeling of li-ion cells for fast simulation of high c-rate and low temperature operations,

    G. Fan, K. Pan, M. Canova, J. Marcicki, and X. G. Yang, “Modeling of li-ion cells for fast simulation of high c-rate and low temperature operations,” Journal of The Electrochemical Society , vol. 163, no. 5, p. A666, 2016

    2016

  13. [21]

    Discovering governing equations from data by sparse identification of nonlinear dynamical systems,

    S. L. Brunton, J. L. Proctor, and J. N. Kutz, “Discovering governing equations from data by sparse identification of nonlinear dynamical systems,” Proceedings of the national academy of sciences , vol. 113, no. 15, pp. 3932–3937, 2016

    2016

Pith tools

Reviewed August 12, 2026 · model on record in the stance chip above.