Pith. sign in

REVIEW 3 major objections 5 minor 41 references

State of health prediction of lithium-ion batteries for driving conditions based on full parameter domain sparrow search algorithm and dual-module bidirectional gated recurrent unit

T0 review · 3 major / 5 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 A sparrow-optimized bidirectional GRU, using one incremental-capacity feature, predicts battery state of health better than six rival models and reaches RMSE under 0.003 on real EV data.

desk verdict A plausible incremental engineering combination whose reported accuracy is likely inflated by bidirectional look-ahead in the test protocol; needs a causal re-evaluation before the numbers can be believed. read the letter →

arxiv 2505.17405 v1 pith:4CCED5M4 submitted 2025-05-23 eess.SY cs.SY

Jie Wen , Chenyu Jia , Guangshu Xia This is my paper
classification eess.SYcs.SY
keywords lithium-ionbatteriesstateofhealthbidirectionalgatedrecurrentunitsparrowsearchalgorithmincrementalcapacityindicatorelectricvehicledrivingconditionshyperparameteroptimization
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 tries to establish that the state of health of electric-vehicle lithium-ion batteries can be predicted accurately using only a single health indicator extracted from charging data, rather than direct capacity measurements or physical battery models. The indicator is the margin factor of the incremental capacity curve, normalized and denoised by singular value decomposition, and selected by Spearman correlation and ablation tests. This one-dimensional series is fed to a dual-module bidirectional gated recurrent unit, and the sparrow search algorithm tunes every hyperparameter of that network independently over the full parameter domain. On a laboratory dataset that simulates driving conditions, the resulting SSA-BiGRU model reports the lowest root-mean-square error in 11 of 12 comparisons against six other fusion models, with errors as low as 0.0027; on a real fleet of 20 EVs, it predicts state of health with RMSE between 0.00037 and 0.0029 after training on one vehicle's complete history. If the claim holds, battery management systems gain a low-cost route to early degradation and replacement-timing warnings that does not require internal battery knowledge.

What carries the argument

The load-bearing object is the one-dimensional health indicator: the margin factor of the incremental capacity curve, normalized to [0,1], denoised with singular value decomposition, and chosen by Spearman correlation and ablation. The predictor is a dual-module BiGRU, a serial stack of two bidirectional gated recurrent units in which each module reads the sequence forward and backward (the backward path uses Flip layers around a GRU), giving every output access to past and future context. The optimizer is the sparrow search algorithm, a swarm method that models producers, scroungers, and warners moving through the search space; here it independently sets the number of GRU cells in each of four layers, the four dropout rates, learning rate, max epochs, and minibatch size, which the authors call full-parameter-domain optimization. The argument runs through these three components: the health indicator condenses degradation into a scalar trace, the bidirectional stack propagates trend information across the whole sequence, and the optimizer removes manual hyperparameter choice.

What would settle it

Run the same SSA-BiGRU model in a strictly causal evaluation on the same laboratory cells: at each cycle, supply only health-indicator values up to that cycle (via masking or recursive one-step-ahead feeding) and recompute RMSE at the 25% starting point. If the causal RMSE stays near the reported 0.0027, the claim survives; if it rises to the level of the runner-up models or above, the reported superiority is an artifact of the bidirectional network seeing future data.

Watch

Extended reading notes

Core claim

On its own terms, the paper shows that one denoised dimensionless feature of the incremental capacity curve contains enough degradation information to drive an accurate SOH predictor. The chosen feature, the margin factor after SVD denoising, is selected by Spearman correlation and ablation experiments, and it is used as the sole input to a dual-module BiGRU. The network stacks two bidirectional GRU modules; each module has a forward GRU and a backward GRU (implemented with Flip layers), so the output at every cycle is conditioned on both earlier and later values of the input sequence. The sparrow search algorithm then optimizes the four GRU layer sizes, four dropout rates, learning rate, maximum epochs, and minibatch size independently, rather than sharing parameters symmetrically as in a conventional BiGRU. The reported evidence is numerical: across four cells and three prediction starting points, SSA-BiGRU has the lowest RMSE in 11 of 12 comparisons, and on real EV charging data its RMSE stays below 0.003 for all 20 vehicles when the model is trained on the complete history of a single other vehicle. The paper also reports a sharp benefit from more training data, with RMSE dropping by nearly two orders of magnitude when the prediction starting point moves from 5% to 25% of the sequence.

Load-bearing premise

The reported test accuracy assumes the model sees the entire test sequence at once, so each SOH prediction can draw on future health-indicator values, and the paper gives no causal masking or recursive one-step-ahead evaluation; if the intended use is true forecasting with only past data, the stated errors are optimistic and the advantage over causal models may shrink or disappear.

Editorial extensions

If this is right

  • A battery management system could flag replacement timing from charging-curve features alone, without capacity measurements or physical models, using the reported error ranges as a guide.
  • The chosen margin-factor health indicator, after SVD denoising, is sufficient as a single input feature; the paper's ablation shows it reduces RMSE by 70% and 86.76% compared with the raw margin factor at the two prediction starting points.
  • Full-parameter-domain SSA optimization matters most when training data reaches 25% of the sequence, where RMSE falls by almost two orders of magnitude relative to 5% starting-point experiments; the optimized model also outperforms an IWOA-optimized BiGRU in most settings.
  • On real EV data, the model generalizes across vehicles of the same model when trained on one vehicle's complete degradation history, and it is better suited to early degradation prediction (e.g., second month) than to later stages.
  • The method transfers from a simulated-driving laboratory dataset to lower-resolution on-road charging data, suggesting the health indicator survives measurement noise and reduced sampling rates.

Reading between the lines

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

  • If the reported accuracy is confirmed under a causal protocol (no future data at prediction time), the practical deployment case becomes much stronger: an onboard estimator could produce a replacement warning from real-time charging data. The current protocol, which feeds the full test sequence through the bidirectional network, leaves this deployment claim untested.
  • The real-fleet experiment is a same-model transfer task; a natural extension is cross-batch or cross-chemistry testing to see whether the full-parameter-domain optimization, which re-tunes hyperparameters to each training set, preserves the margin over simpler baselines.
  • Reporting point errors without uncertainty bounds leaves end-of-life decisions exposed; adding prediction intervals or quantile outputs would let a battery management system trade off false alarms against missed replacements, a direct testable extension of the same architecture.
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

3 major / 5 minor

Summary. The manuscript proposes a fusion model (SSA-BiGRU) for lithium-ion battery state-of-health (SOH) prediction, combining a dual-module bidirectional GRU with sparrow search algorithm (SSA) hyperparameter optimization and an incremental-capacity (IC)-curve-derived health indicator (HI). The authors first extract candidate HIs from IC curves, use Spearman correlation and ablation on battery Cell2 to select the best one, then optimize the BiGRU hyperparameters with SSA. They evaluate the model on the Oxford battery dataset (Cell1, Cell3, Cell7, Cell8) at three prediction starting points and on a real-road EV charging dataset of 20 vehicles, reporting very low RMSE/MAE/MAPE values and claiming superior accuracy, robustness, and generalization compared with six alternative fusion models.

Significance. If the reported accuracy were obtained under a causally valid forecasting protocol, the paper would offer a useful engineering contribution: the HI extraction via IC curves with correlation/ablation screening is methodologically reasonable, the dual-module BiGRU architecture is clearly specified, and the real-EV validation on 20 vehicles is valuable for practical deployment. However, the central claim is currently undermined by a bidirectional look-ahead in the test protocol, which makes the predictions appear artificially accurate, and by a self-normalized SOH definition in the real-EV experiment that makes errors trivially small. The paper does not release code or provide machine-checked proofs; its value rests entirely on the experimental comparison, so the validity of that comparison is decisive.

major comments (3)
  1. [3.1.1, Eq. (8); 4.3; 4.4] Section 3.1.1, Eq. (8) defines the BiGRU output as the concatenation of forward and reverse hidden-state sequences, so that at every test time step the model has access to future test-sequence HI values. The experimental protocols in Sections 4.3 and 4.4 feed the complete test sequence (or the full SOH curve from month 2 onward) through this model, and the paper describes no causal masking, no recursive one-step-ahead strategy, and no statement that future test data are excluded. The reported RMSE values (e.g., 0.0026961 for Cell8 at 25% starting point in Table 4, and 0.00037008 for V15 in Table 5) may therefore reflect interpolation of the full SOH curve rather than forecasting, and the margin over unidirectional baselines could be an artifact of the evaluation protocol. The authors must re-run the evaluation with a causally valid protocol, such as one-step-ahead recursive prediction or masked future inputs, and report the resulting errors.
  2. [4.1, 4.2, 4.3] Sections 4.1 and 4.2 select the health indicator (MF-SVD) and tune the SSA-optimized hyperparameters using battery Cell2, and Section 4.3 then evaluates the same model on Cell1, Cell3, Cell7, and Cell8. The manuscript does not state whether SSA is re-run on each test cell or whether the Cell2-optimized parameters are transferred unchanged. If the parameters are transferred, the 'full parameter domain' claim in Section 3.2 (Table 2) applies only to Cell2, and the cross-cell results are a transfer-learning test rather than a demonstration of the optimization method's generality; if SSA is re-run per cell, the comparison with baselines that are not similarly re-tuned is not fair. The protocol must be clarified, and the authors should specify exactly which parameters were optimized, on which data, and how the comparison models were tuned.
  3. [4.4; Conclusions] In Section 4.4, the SOH for each real EV is computed with the maximum of that vehicle's own monthly median as the denominator, so every vehicle's SOH curve starts at 1 by construction. The reported RMSE values are therefore normalized relative to each vehicle's own starting capacity and cannot be meaningfully compared across vehicles or against the Oxford results; the Conclusions' statement that 'the highest is not more than 0.0025' is expressed in this self-referential unit. The authors should report errors in absolute capacity (Ah) or against a common denominator, and should state the corresponding absolute errors so the practical significance of the prediction accuracy can be assessed.
minor comments (5)
  1. [Throughout] The manuscript contains numerous typos and language errors, including 'Sigmod' in Section 3.1.1, 'Filp layer' in Section 3.2 and Figure 12, and 'pre- and post-textual information' in the abstract.
  2. [Figure 10] Figure 10 contains embedded Chinese text, which is not appropriate for an English-language journal and should be replaced with an English-language schematic.
  3. [Eq. (2)] Equation (2) is poorly formatted, and the definitions of the crest factor, pulse factor, margin factor, and kurtosis are not fully consistent with the surrounding text; please rewrite these formulas with clear variable definitions.
  4. [2.2; [28]] The HI extraction procedure refers to the authors' previous work [28] for technical details, but the current manuscript should be self-contained; please provide a complete description of the extraction and SVD denoising steps.
  5. [4.3, qualitative analysis] The text states that FM1 is below the average error 'in all cases except for battery Cell1 at 5% prediction starting point,' but Table 4 shows FM1's RMSE for Cell1 at 5% (0.032169) is below the average (0.0759); this sentence appears to be inaccurate and should be corrected.

Circularity Check

3 steps flagged · score 6.0 of 10

The BiGRU test protocol feeds full future HI sequences, so the reported 'predictions' are non-causal interpolations; HI extraction is delegated to a self-cited preprint, and Cell2 is used for both tuning and ablation, making several results circular by construction.

  1. fitted input called prediction [Section 3.1.1 (Eq. 8) and Section 4.1 prediction starting point protocol]
    "The BiGRU consists of two layers of GRUs that can use both past and future information... processing the sequences in forward and reverse direction respectively, and finally generates the output Y_t of the whole network from the forward and reverse outputs... Since both the HI sequence and the actual SOH sequence are one-dimensional and of equal length, the first 15% and 25% of the sequential data of the HI sequence are selected as the training set, and the experiments are performed on the data after 15% and 25% of the SOH sequence, respectively."

    At test time the model receives the entire post-cut HI sequence and, because it is bidirectional, each output depends on the reverse hidden state computed from future test HIs. Thus the 'prediction' at every test cycle is a function of future health-indicator values from the same battery, so the reported RMSE measures non-causal interpolation rather than forecasting. The advantage over unidirectional baselines, which cannot see the future, is therefore built into the evaluation protocol and not a genuine predictive improvement.

  2. self citation load bearing [Section 2.2, sentence after Eq. (2)]
    "The more technical details of extracting HI based on ICA used in this paper can be found in our previous work [28]."

    The paper's key input feature, the HI selected from the IC curve, is defined by a method whose technical details are relegated to a self-cited preprint by overlapping authors (Jia et al., arXiv:2503.23858). The present paper does not independently specify how Boundary1/Boundary2 are chosen or how the HI is computed, so the claimed 'first-principles' HI extraction is imported via a self-citation rather than from an externally verified independent source. This makes the feature-construction step load-bearing on the authors' own prior work.

1 more flagged steps
  1. fitted input called prediction [Section 3.3 Step 2 and Section 4.2 model ablation first paragraph]
    "taking Cell2 cell as an example, one can take these three HIs and the margin factor without SVD treatment as training data respectively, and performs SOH prediction through the dual-module BiGRU model, and the best HI is screened out according to the prediction error... we sets up a model ablation experiment on SSA-BiGRU based on the data of battery Cell2 and the HI selection strategy derived in subsection 4.1."

    The same Cell2 data are used to select the HI, to tune SSA hyperparameters, and then to demonstrate SSA-BiGRU superiority in the ablation experiments. The model and feature are chosen to minimize error on Cell2's test portion, so the Cell2 ablation results are in-sample and cannot independently support the generalization claim. The later Cell1/3/7/8 and real-EV experiments provide some outside evidence, but the Cell2-based conclusions are circular.

full rationale

The paper's headline accuracy claim for SSA-BiGRU is supported primarily by the Oxford-battery experiments (Table 4) and the real-EV experiments (Tables 5-6). The most serious circularity is the bidirectional test protocol: Eq. (8) defines each BiGRU output as a function of both forward and reverse hidden sequences, and the experiments feed the full post-training-cut HI sequence into the model. Each test output therefore uses future HI values from the same battery, making the reported RMSE a measure of non-causal interpolation rather than forecasting. This affects every reported experiment and inflates the comparison with unidirectional baselines by construction. A second circular element is the HI extraction, whose technical details are delegated to the authors' own prior work [28], so the central input feature is imported via self-citation. Third, Cell2 is used both to select the HI and to tune SSA hyperparameters and then to demonstrate SSA-BiGRU superiority in the ablation; those Cell2 numbers are in-sample. The later held-out results on Cell1/3/7/8 and V2-V20 provide some independent content, so the paper does not reduce entirely to a fitted constant, but the non-causal protocol means the claimed 'predictions' reduce in part to interpolating the test sequence. Score 6 reflects partial circularity in the central prediction claim.

Assumptions & free parameters 5 free parameters · 7 assumptions · 0 invented entities

The model relies on a small number of tunable choices (SSA hyperparameters, HI selection, SVD rank, IC boundaries, SOH normalization) and on several domain assumptions about what the datasets represent. No new physical entities are introduced.

free parameters (5)
  • SSA-tuned hyperparameters of dual-module BiGRU = not reported (search ranges in Table 2)
    SSA with 6 populations and 10 iterations optimizes GRU cell counts, dropout rates, learning rate, batch size, and epochs on Cell2; the resulting values are not given, so the model cannot be reproduced exactly.
  • Health indicator selection (MF-SVD) = margin factor after SVD denoising
    Chosen based on Spearman correlation and ablation experiments on Cell2; other candidate HIs (PF, Kur) are discarded, so the reported accuracy is conditioned on this in-sample choice.
  • SVD denoising truncation = not specified
    The paper says SVD denoising is applied to each normalized HI, but the retained rank or energy threshold is not reported (Section 2.2 and 2.3.1).
  • IC curve integration boundaries = Boundary1, Boundary2, Position1 from prior work [28]
    The area-under-curve features are defined relative to voltage boundaries inherited from the authors' previous paper; these are not derived here.
  • Real-EV SOH denominator = each vehicle's maximum monthly median capacity
    Section 4.4 uses the vehicle-specific max median as the denominator so the initial SOH is 1; this rescales the target per vehicle and affects the error metric.
assumptions (7)
  • standard math GRU and BiGRU update equations (Section 3.1.1)
    The network definitions are taken from the literature and are used as building blocks.
  • standard math Sparrow search algorithm update rules (Section 3.1.2)
    The optimizer is cited from Xue and Shen [37] and used as given.
  • domain assumption IC curves computed from constant-current charging data capture degradation
    The method maps IC-derived features to SOH; if IC features do not track degradation outside this dataset, the input is uninformative.
  • domain assumption The Oxford dataset's drive-cycle protocol approximates real driving conditions
    The paper chooses this dataset specifically to simulate driving, but the transfer to real fleets is only tested on 20 vehicles of one model.
  • domain assumption Monthly median charging capacity represents SOH for real EVs
    Section 2.3.2 selects the monthly median to suppress outliers; this aggregation choice affects the target variable and the reported error.
  • domain assumption Training on one EV (V1) generalizes to other EVs of the same model
    Section 4.4 assumes a single vehicle's degradation history is sufficient to predict 19 others; the paper does not quantify the spread of degradation patterns.
  • ad hoc to paper SVD denoising removes noise without removing degradation signal
    SVD is applied to each normalized HI with no explicit justification of the retained components (Section 2.3.1).

how reviews work

0 comments
Cite this review

Pith. "Pith review of State of health prediction of lithium-ion batteries for driving conditions based on full parameter domain sparrow search algorithm and dual-module bidirectional gated recurrent unit." pith.science (2026). https://pith.science/paper/4CCED5M4

@misc{pith2026250517405,
  author       = {Pith},
  title        = {Pith review of: State of health prediction of lithium-ion batteries for driving conditions based on full parameter domain sparrow search algorithm and dual-module bidirectional gated recurrent unit},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/4CCED5M4}},
  note         = {Machine review of arXiv:2505.17405}
}
read the original abstract

Aiming at the state of health (SOH) prediction of lithium-ion batteries (LiBs) for electric vehicles (EVs), this paper proposes a fusion model of a dual-module bidirectional gated recurrent unit (BiGRU) and sparrow search algorithm (SSA) with full parameter domain optimization. With the help of Spearman correlation analysis and ablation experiments, the indirect health indicator (HI) that can characterize the battery degradation is extracted first based on the incremental capacity (IC) curves of the Oxford battery dataset, which simulates the driving conditions. On this basis, the filtered one-dimensional HI is inputted into the dual-module BiGRU for learning the pre- and post-textual information of the input sequence and extracting the sequence features. In order to combine the different hyperparameters in the dual-module BiGRU, SSA is used to optimize the hyperparameters in the full parameter domain. The proposed SSA-BiGRU model combines the advantages and structures of SSA and BiGRU to achieve the highly accurate SOH prediction of LiBs. Studies based on the Oxford battery dataset have shown that the SSA-BiGRU model has higher accuracy, better robustness and generalization ability. Moreover, the proposed SSA-BiGRU model is tested on a real road-driven EV charging dataset and accurate SOH prediction are obtained.

Figures

Figures reproduced from arXiv: 2505.17405 by the authors.

Figure 1
Figure 1. Schematic diagram of feature extraction based on IC curve. 2.3. Dataset 2.3.1. Oxford battery dataset The Oxford dataset [29] contains eight Kokam pouch batteries rated at 740 mAh with graphite as the negative material as well as lithium cobaltate and lithium nickel cobaltate as the positive materials. The aging tests of these batteries can be categorized into capacity calibration tests and drive cycle tests. Specif… view at source ↗
Figure 2
Figure 2. Capacity degradation curve of Oxford battery dataset. The HI of LiBs comes from its own parameters obtained through monitoring and secondary analysis during operation, which can effectively characterize the degradation state of the battery in terms of data. ICA can reflect the internal changes of the battery to a certain extent, and the source of data is simpler and easier to obtain than the J. Wen: Preprint submitt… view at source ↗
Figure 3
Figure 3. SOH curves of Oxford battery dataset. capacity or internal resistance. Therefore, obtaining effective and feasible HI through ICA is very important for SOH prediction of LiBs.The Oxford battery dataset simulates the charging and discharging under real driving conditions, while in reality, the complexity and uncertainty of the conditions lead to slightly different discharging conditions of the batteries. Thus, in thi… view at source ↗
Figures from the paper (16 more)
Figure 4
Figure 4. Figure 4: Partial charge IC curve of battery Cell2. Using the HI extraction method in subsection 2.2 and our previous work [28], based on the IC curves extracted from the Oxford battery dataset, margin factor, area, peak, waveform factor, pulse factor and kurtosis in total 6 HIs…
Figure 5
Figure 5. Figure 5: Thermogram of SVD-HI vs. SOH Spearman correlation for battery Cell2. 2.3.2. Real road-driven EV charging dataset The real road-driven EV dataset used in this paper is derived from the battery pack charging data of 20 commercial EVs used by Deng et al. in [32], with a t…
Figure 6
Figure 6. Figure 6: Charging curves of V1. The raw charging data for the 20 EVs is shown in [PITH_FULL_IMAGE:figures/full_fig_p007_6.png]
Figure 7
Figure 7. Figure 7: Raw charging data of 20 EVs. 2.4. Evaluation indicators In this paper, three error evaluation metrics are used to quantitatively assess the output of the prediction model, namely root mean square error (RMSE), mean absolute error (MAE) and mean absolute percentage erro…
Figure 8
Figure 8. Figure 8: Monthly average and median charging of 20EVs. address the limitations of RNN by introducing a gating mechanism, which enables it to selectively remember key information. LSTM has three gates: forget gate, input gate, and output gate. Different from LSTM, GRU combines t…
Figure 9
Figure 9. Figure 9: Structure of GRU. J. Wen: Preprint submitted to Elsevier Page 9 of 25 [PITH_FULL_IMAGE:figures/full_fig_p009_9.png]
Figure 11
Figure 11. Figure 11: Solution process of SSA. J. Wen: Preprint submitted to Elsevier Page 11 of 25 [PITH_FULL_IMAGE:figures/full_fig_p011_11.png]
Figure 12
Figure 12. Figure 12: Schematic diagram of the network layer structure of the dual-module BiGRU model [PITH_FULL_IMAGE:figures/full_fig_p013_12.png]
Figure 13
Figure 13. Figure 13: SOH prediction flowchart based on SSA-BiGRU model. 4. SOH prediction experiment and result analysis based on SSA-BiGRU model In this section, the Oxford battery dataset and the real road-driven EV charging dataset are selected as the research objects, which are used t…
Figure 15
Figure 15. Figure 15: Error of model ablation for battery Cell2. According to [PITH_FULL_IMAGE:figures/full_fig_p016_15.png]
Figure 16
Figure 16. Figure 16: Experiment results of model comparison for different prediction starting points for battery Cell1. prediction over the previous two sets of experiments, while the SSA-BiGRU model achieves the best fitting effect and achieves almost the same prediction as the original …
Figure 17
Figure 17. Figure 17: SOH prediction results of different models for batteries Cell3, Cell7 and Cell8. Quantitative analysis. Taking the RMSE of four groups of batteries as an example, the following results can be obtained: (1) 5% prediction starting point: Cell8 reduces the most compared …
Figure 18
Figure 18. Figure 18: Prediction errors for different prediction starting points for batteries Cell3 and Cell7. the rated capacity of the battery, i.e., 145 Ah. Moreover, since there are multiple incomplete charging and discharging behaviors under actual driving conditions, if 145 Ah is us…
Figure 19
Figure 19. Figure 19: SOH of 20 EVs charging data by using median. 0 5 10 15 20 25 30 Month 0.85 0.9 0.95 1 SOH Vehicle 1 0 5 10 15 20 25 30 Month 0.85 0.9 0.95 1 SOH Vehicle 2 0 5 10 15 20 25 30 Month 0.85 0.9 0.95 1 SOH Vehicle 3 0 5 10 15 20 25 30 Month 0.86 0.88 0.9 0.92 0.94 0.96 0.98…
Figure 20
Figure 20. Figure 20: SOH prediction of 20 EVs when the prediction starting point is the second month. J. Wen: Preprint submitted to Elsevier Page 21 of 25 [PITH_FULL_IMAGE:figures/full_fig_p021_20.png]
Figure 21
Figure 21. Figure 21: Prediction results of different prediction starting points for 6 EVs. an increasing trend. For the different EVs, V15 outperforms the others on all evaluation indicators, indicating higher predictability and stability. On the other hand, V20 remains consistent with th…

Discussion (0). Sign in to comment.

Reference graph

Works this paper leans on

41 extracted references · 39 canonical work pages

  1. [1]

    Aglobal–localcontextembeddinglearningbasedsequence- free framework for state of health estimation of lithium-ion battery.Energy, 282:128306, 2023

    ZhengyiBao,JiahaoNie,HuipinLin,JiahaoJiang,ZhiweiHe,andMingyuGao. Aglobal–localcontextembeddinglearningbasedsequence- free framework for state of health estimation of lithium-ion battery.Energy, 282:128306, 2023

    work page 2023

  2. [2]

    Research directions for next-generation battery management solutions in automotive applications.Renewable and Sustainable Energy Reviews, 152:111695, 2021

    Xiaosong Hu, Zhongwei Deng, Xianke Lin, Yi Xie, and Remus Teodorescu. Research directions for next-generation battery management solutions in automotive applications.Renewable and Sustainable Energy Reviews, 152:111695, 2021

    work page 2021

  3. [3]

    Historical data-independent remaining useful life prediction method based on dual-input deep learning neural network.Journal of Energy Storage, 72:108427, 2023

    Junyi Xia, Qionglin Shi, Haomiao Li, Min Zhou, Wei Wang, Kangli Wang, and Kai Jiang. Historical data-independent remaining useful life prediction method based on dual-input deep learning neural network.Journal of Energy Storage, 72:108427, 2023

    work page 2023

  4. [4]

    Acomprehensivereviewofbatterymodeling and state estimation approaches for advanced battery management systems.Renewable and Sustainable Energy Reviews, 131:110015, 2020

    YujieWang,JiaqiangTian,ZhendongSun,LiWang,RuilongXu,MinceLi,andZonghaiChen. Acomprehensivereviewofbatterymodeling and state estimation approaches for advanced battery management systems.Renewable and Sustainable Energy Reviews, 131:110015, 2020

    work page 2020

  5. [5]

    Electrical energy storage for the grid: a battery of choices.Science, 334(6058):928– 935, 2011

    Bruce Dunn, Haresh Kamath, and Jean-Marie Tarascon. Electrical energy storage for the grid: a battery of choices.Science, 334(6058):928– 935, 2011

    work page 2011

  6. [6]

    A comparative investigation of aging effects on thermal runaway behavior of lithium-ion batteries.ETransportation, 2:100034, 2019

    Dongsheng Ren,Hungjen Hsu,Ruihe Li, XuningFeng, DongxuGuo, Xuebing Han,Languang Lu,Xiangming He, ShangGao, Junxian Hou, et al. A comparative investigation of aging effects on thermal runaway behavior of lithium-ion batteries.ETransportation, 2:100034, 2019

    work page 2019

  7. [7]

    Areviewonlithium-ionbattery ageing mechanisms and estimations for automotive applications.Journal of Power Sources, 241:680–689, 2013

    AnthonyBarré,BenjaminDeguilhem,SébastienGrolleau,MathiasGérard,FrédéricSuard,andDelphineRiu. Areviewonlithium-ionbattery ageing mechanisms and estimations for automotive applications.Journal of Power Sources, 241:680–689, 2013

    work page 2013

  8. [8]

    Healthprognosticsforlithium-ionbatteries:mechanisms,methods, and prospects

    YunhongChe,XiaosongHu,XiankeLin,JiaGuo,andRemusTeodorescu. Healthprognosticsforlithium-ionbatteries:mechanisms,methods, and prospects. Energy & Environmental Science, 16(2):338–371, 2023

    work page 2023

Show all 41 references
  1. [9]

    Energy, 262:125497, 2023

    ZhelinHuang,FanXu,andFangfangYang.Stateofhealthpredictionoflithium-ionbatteriesbasedonautoregressionwithexogenousvariables model. Energy, 262:125497, 2023

    2023

  2. [10]

    Battery lifetime prognostics.Joule, 4(2):310–346, 2020

    Xiaosong Hu, Le Xu, Xianke Lin, and Michael Pecht. Battery lifetime prognostics.Joule, 4(2):310–346, 2020

    2020

  3. [11]

    Stateofhealthpredictionoflithium-ion batteries based on machine learning: advances and perspectives.iScience, 24(11):103265, 2021

    XingShu,ShiquanShen,JiangweiShen,YuanjianZhang,GuangLi,ZhengChen,andYonggangLiu. Stateofhealthpredictionoflithium-ion batteries based on machine learning: advances and perspectives.iScience, 24(11):103265, 2021

    2021

  4. [12]

    A single particle model with chemical/mechanical degradation physics for lithium ion battery state of health (soh) estimation.Applied Energy, 212:1178–1190, 2018

    Jie Li, Kasim Adewuyi, Nima Lotfi, Robert G Landers, and Jonghyun Park. A single particle model with chemical/mechanical degradation physics for lithium ion battery state of health (soh) estimation.Applied Energy, 212:1178–1190, 2018

    2018

  5. [13]

    State-of-health estimation of lithium-ion batteries for electrified vehicles using a reduced-order electrochemical model.Journal of Energy Storage, 52:104684, 2022

    Seyedmehdi Hosseininasab, Changwei Lin, Stefan Pischinger, Michael Stapelbroek, and Giovanni Vagnoni. State-of-health estimation of lithium-ion batteries for electrified vehicles using a reduced-order electrochemical model.Journal of Energy Storage, 52:104684, 2022

    2022

  6. [14]

    DatongLiu,JianbaoZhou,HaitaoLiao,YuPeng,andXiyuanPeng. Ahealthindicatorextractionandoptimizationframeworkforlithium-ion battery degradation modeling and prognostics.IEEE Transactions on Systems, Man, and Cybernetics: Systems, 45(6):915–928, 2015

    2015

  7. [15]

    SynchronousstateofhealthestimationandremainingusefullifetimepredictionofLi-Ionbattery through optimized relevance vector machine framework.Energy, 251:123852, 2022

    ZhiqiangLyu,GengWang,andRenjingGao. SynchronousstateofhealthestimationandremainingusefullifetimepredictionofLi-Ionbattery through optimized relevance vector machine framework.Energy, 251:123852, 2022

    2022

  8. [16]

    A novel multistage support vector machine based approach for Li-ion battery remaining useful life estimation.Applied Energy, 159:285–297, 2015

    Meru A Patil, Piyush Tagade, Krishnan S Hariharan, Subramanya M Kolake, Taewon Song, Taejung Yeo, and Seokgwang Doo. A novel multistage support vector machine based approach for Li-ion battery remaining useful life estimation.Applied Energy, 159:285–297, 2015

    2015

  9. [17]

    Lithium-ion batteries health prognosis via differential thermal capacity with simulated annealing and support vector regression.Energy, 250:123829, 2022

    Mingqiang Lin, Chenhao Yan, Jinhao Meng, Wei Wang, and Ji Wu. Lithium-ion batteries health prognosis via differential thermal capacity with simulated annealing and support vector regression.Energy, 250:123829, 2022

    2022

  10. [18]

    State of health estimation of lithium-ion battery based on improved ant lion optimization and support vector regression.Journal of Energy Storage, 50:104215, 2022

    Qianglong Li, Dezhi Li, Kun Zhao, Licheng Wang, and Kai Wang. State of health estimation of lithium-ion battery based on improved ant lion optimization and support vector regression.Journal of Energy Storage, 50:104215, 2022

    2022

  11. [19]

    Jingwen Wei, Guangzhong Dong, and Zonghai Chen. Remaining useful life prediction and state of health diagnosis for lithium-ion batteries using particle filter and support vector regression.IEEE Transactions on Industrial Electronics, 65(7):5634–5643, 2017. J. Wen: Preprint sub...

    2017

  12. [20]

    Capacity degradation prediction of lithium-ion battery based on artificial bee colony and multi-kernel support vector regression.Journal of Energy Storage, 72:108160, 2023

    Kui Chen, Qiang Liao, Kai Liu, Yan Yang, Guoqiang Gao, and Guangning Wu. Capacity degradation prediction of lithium-ion battery based on artificial bee colony and multi-kernel support vector regression.Journal of Energy Storage, 72:108160, 2023

    2023

  13. [21]

    ShuzhiZhang,BaoyuZhai,XuGuo,KaikeWang,NianPeng,andXiongwenZhang. Synchronousestimationofstateofhealthandremaining useful lifetime for lithium-ion battery using the incremental capacity and artificial neural networks.Journal of Energy Storage, 26:100951, 2019

    2019

  14. [22]

    An indirect remaining useful life prognosis for Li-ion batteries based on health indicator and novel artificial neural network.Journal of Energy Storage, 52:104701, 2022

    Ting Tang and Huimei Yuan. An indirect remaining useful life prognosis for Li-ion batteries based on health indicator and novel artificial neural network.Journal of Energy Storage, 52:104701, 2022

    2022

  15. [23]

    Health and lifespan prediction considering degradation patterns of lithium-ion batteries based on transferable attention neural network.Energy, 279:128137, 2023

    Aihua Tang, Yihan Jiang, Yuwei Nie, Quanqing Yu, Weixiang Shen, and Michael G Pecht. Health and lifespan prediction considering degradation patterns of lithium-ion batteries based on transferable attention neural network.Energy, 279:128137, 2023

    2023

  16. [24]

    A machine-learning prediction method of lithium-ion battery life based on charge process for different applications.Applied Energy, 292:116897, 2021

    Yixin Yang. A machine-learning prediction method of lithium-ion battery life based on charge process for different applications.Applied Energy, 292:116897, 2021

    2021

  17. [25]

    Akram Eddahech, Olivier Briat, Nicolas Bertrand, Jean-Yves Delétage, and Jean-Michel Vinassa. Behavior and state-of-health monitoring of Li-ion batteries using impedance spectroscopy and recurrent neural networks.International Journal of Electrical Power & Energy Systems , 42(...

    2012

  18. [26]

    TianyiLuo,MingLiu,PengShi,GuangrenDuan,andXibinCao. Ahybriddatapreprocessing-basedhierarchicalattentionBiLSTMnetwork for remaining useful life prediction of spacecraft lithium-ion batteries.IEEE Transactions on Neural Networks and Learning Systems, 2023

    2023

  19. [27]

    Convolutional autoencoder- based SOH estimation of lithium-ion batteries using electrochemical impedance spectroscopy.Journal of Energy Storage, 60:106680, 2023

    Josue Obregon, Yu-Ri Han, Chang Won Ho, Devanadane Mouraliraman, Chang Woo Lee, and Jae-Yoon Jung. Convolutional autoencoder- based SOH estimation of lithium-ion batteries using electrochemical impedance spectroscopy.Journal of Energy Storage, 60:106680, 2023

    2023

  20. [28]

    Incremental capacity-based multi-feature fusion model for predicting state-of-health of lithium-ion batteries.arXiv preprint arXiv:2503.23858, 2025

    Chenyu Jia, Guangshu Xia, Yuanhao Shi, Jianfang Jia, Jie Wen, and Jianchao Zeng. Incremental capacity-based multi-feature fusion model for predicting state-of-health of lithium-ion batteries.arXiv preprint arXiv:2503.23858, 2025

    2025 arXiv

  21. [29]

    Aparametricopencircuitvoltagemodelforlithiumionbatteries

    ChristopheRBirkl,EuanMcTurk,MRRoberts,PGBruce,andDAHowey. Aparametricopencircuitvoltagemodelforlithiumionbatteries. Journal of The Electrochemical Society, 162(12):A2271, 2015

    2015

  22. [30]

    Diagnosis and prognosis of degradation in lithium-ion batteries.Department of Engineering Science, University of Oxford , 2017

    RB Christoph. Diagnosis and prognosis of degradation in lithium-ion batteries.Department of Engineering Science, University of Oxford , 2017

    2017

  23. [31]

    Ming Zhao and Xiaodong Jia. A novel strategy for signal denoising using reweighted SVD and its applications to weak fault feature enhancement of rotating machinery.Mechanical Systems and Signal Processing, 94:129–147, 2017

    2017

  24. [32]

    Prognostics of battery capacity based on charging data and data-driven methods for on-road vehicles.Applied Energy, 339:120954, 2023

    Zhongwei Deng, Le Xu, Hongao Liu, Xiaosong Hu, Zhixuan Duan, and Yu Xu. Prognostics of battery capacity based on charging data and data-driven methods for on-road vehicles.Applied Energy, 339:120954, 2023

    2023

  25. [33]

    Learningrepresentationsbyback-propagatingerrors

    DavidERumelhart,GeoffreyEHinton,andRonaldJWilliams. Learningrepresentationsbyback-propagatingerrors. Nature,323(6088):533– 536, 1986

    1986

  26. [34]

    Long short-term memory networks for accurate state-of-charge estimation of Li-ion batteries.IEEE Transactions on Industrial Electronics, 65(8):6730–6739, 2017

    Ephrem Chemali, Phillip J Kollmeyer, Matthias Preindl, Ryan Ahmed, and Ali Emadi. Long short-term memory networks for accurate state-of-charge estimation of Li-ion batteries.IEEE Transactions on Industrial Electronics, 65(8):6730–6739, 2017

    2017

  27. [35]

    Long short-term memory.Neural Computation, 9(8):1735–1780, 1997

    Sepp Hochreiter and Jürgen Schmidhuber. Long short-term memory.Neural Computation, 9(8):1735–1780, 1997

    1997

  28. [36]

    KyunghyunCho,BartVanMerriënboer,CaglarGulcehre,DzmitryBahdanau,FethiBougares,HolgerSchwenk,andYoshuaBengio.Learning phrase representations using RNN encoder-decoder for statistical machine translation.arXiv preprint arXiv:1406.1078, 2014

    2014 arXiv

  29. [37]

    A novel swarm intelligence optimization approach: sparrow search algorithm.Systems Science & Control Engineering, 8(1):22–34, 2020

    Jiankai Xue and Bo Shen. A novel swarm intelligence optimization approach: sparrow search algorithm.Systems Science & Control Engineering, 8(1):22–34, 2020

    2020

  30. [38]

    Improved whale optimization algorithm.Application Research of Computers, 37(12):3618–3621, 2020

    Zequan Wu and Yongmin Mu. Improved whale optimization algorithm.Application Research of Computers, 37(12):3618–3621, 2020

    2020

  31. [39]

    Squeeze-and-excitation networks.IEEE Transactions on Pattern Analysis and Machine Intelligence, 42(8):2011–2023, 2020

    Jie Hu, Li Shen, Samuel Albanie, Gang Sun, and Enhua Wu. Squeeze-and-excitation networks.IEEE Transactions on Pattern Analysis and Machine Intelligence, 42(8):2011–2023, 2020

    2011

  32. [40]

    Cloud computing beckons scientists.Nature, 509(7502):543–544, 2014

    Nadia Drake et al. Cloud computing beckons scientists.Nature, 509(7502):543–544, 2014

    2014

  33. [41]

    The internet of things comes to the lab.Nature, 542(7639):125–126, 2017

    Jeffrey M Perkel. The internet of things comes to the lab.Nature, 542(7639):125–126, 2017. J. Wen: Preprint submitted to Elsevier Page 25 of 25

    2017

Pith tools

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