Machine Learning-Based State-of-Charge Prediction for Electric Bus Fleet: A Critical Analysis

Round 1

Reviewer 1 Report

Comments and Suggestions for Authors

The main suggestions are listed below:

(1) It is suggested that abbreviations should not be used in the title, as this would cause a lot of confusion for readers and reviewers.

(2) The term "Remaining Useful Life (RUL)" was mentioned in the title, but it was not present in the abstract or introduction. Is this appropriate?

(3) In the introduction, you need to connect the state of the art to your paper goals. Please follow the literature review with a clear and concise state of the art analysis. This should clearly show the knowledge gaps identified and link them to your paper goals.

(4) In Line 17-20, “In this context, Electric Vehicles (EVs) emerge as one of the most widely adopted alternative, thanks to electric drives offering several benefits, if compared to traditional Internal Combustion Engines (ICEs) like energy efficiency and regeneration.” It would be better to reference some literatures. For example: https://doi.org/10.1016/j.applthermaleng.2025.126035”

(5) In Line 124-126, Line 272-278, Line 348-377, its numerical designation is the same as that of the main title. That is inappropriate.

(6) To enhance clarity, the curves in Figure 3 and Figure 8 should be distinguished using more than just color. Additionally, the clarity of Figure 3 and Figure 7, as well as the font sizes of the horizontal and vertical axes, needs to be adjusted.

(7) The author might consider incorporating the key quantitative findings into the abstract and conclusion to enhance the paper's clarity and impact.

(8) The table header of Table 1 should be placed at the top of the table.

Author Response

Please, refer to the attached file for point-by-point rewiews

Author Response File: Author Response.pdf

Reviewer 2 Report

Comments and Suggestions for Authors

The authors presented machine learning methods to estimate the battery SOC. Given the rapid progress of the AI and EV industry, the study can have a considerable impact. However, some issues must be resolved.
1. The title says "Estimation of battery RUL"; however, the paper mainly deals with SOC estimation, and the RUL is only implicitly estimated. The title should be more consistent with the main text.
2. The authors showed results from many ML techniques without any introduction. For the methods applied in the paper, the authors should introduce them in the methodology section.
 3. Figure 3 is confusing; the purpose of doing the average over temperatures is unclear since you include temperature as an input variable.
4. For the ML part, how the training, validation, and test data are distributed should be discussed.
5. It's unclear why the author arrived at the results from the bagged tree method in Fig. 4. It's suggested to show how the study progresses. For example, how does it compare with simple polynomial regression?
6. Fig 5 is very confusing, and it's not clear what the model names mean, not to mention models with the same names. The authors should introduce them thoroughly in the methodology section. For example, for neural networks, the authors should answer the following questions: what is the architecture of the network? How many layers are there? What is in each layer? What's the activation function used? etc. The provided information should be enough for people to reproduce the results.
7. In Fig.6, it's unclear why the training fails. Is it due to the limited amount of time or out of memory? If none of the neural networks worked, it's not necessary to present them.
8. For the daily datasets section. The results are trivial and probably can be removed safely.
9. In Fig.8, the authors should provide information on how the optimization is carried out. If that's part of the algorithm, then it's not meaningful to present unconverged results (first few iterations).

Author Response

Please, refer to attached file for review concerns

Author Response File: Author Response.pdf

Reviewer 3 Report

Comments and Suggestions for Authors

The authors present a kind of comparative study of model-based (PI observer) and machine learning techniques for estimating the State of Charge (SoC) of Li-ion batteries in electric buses, using both experimental (DST, FUDS) and real-world fleet datasets. The manuscript shows that ML models can achieve better accuracy with limited inputs through iterative data preprocessing, comparable to traditional observers. The approach is briefly extended to imply RUL estimation for fleet management, highlighting scalability for practical EV applications. The manuscript has some potential; however, it requires a thorough major revision before it can be considered for possible publication. The major issues are highlighted below:

1. The title emphasizes RUL, yet the core content is on machine learning-based SOC. Therefore, either expand the RUL estimation with dedicated methodology and validation sections, OR revise the title to reflect the primary focus on SOC.
2. The abstract and introduction sections need to be rewritten. The current abstract is highly generalized and does not provide any specific information about the content. Optimize the content using “specific issue of SOC estimation to be solved + introduction of your technique as a potential solution + main technical contributions + comparative advantage of your techniques + prospects (if any)”. 
3. The authors mainly compared their results with control observer-based methods (owing to their practicality, fair enough); however, the introduction does not include any specific details on observer-based battery SoC estimation. Please add sufficient literature on it (10.1109/TMECH.2024.3459644). Better divide the SOC estimation into four sections (Direct measurement approach, electrochemical model-based techniques, ECM-based method, and ML-based techniques), OR subdivide the model-based techniques into electrochemical and equivalent circuit models.
4. The mentioned literature on the PI observer is very outdated. Please include the latest development on the PI/PID observer for SOC estimation (10.1109/TMECH.2025.3565278)
5. While authors tried to address overfitting. It is suggested to better include specific metrics and explore advanced techniques to strengthen generalization claims.
6. It is highly suggested to compare your ML models not only with the PI observer but also with recent state-of-the-art methods (like physics-informed ML for SoC) on standard datasets to better contextualize performance.
7. As the measurement uncertainties are a critical issue in real BMS (10.1109/TPEL.2024.3386739), it is suggested to include the impact of measurement uncertainties on the accuracy of the proposed algorithm to reflect the robustness and suitability of the proposed approach for practical applications.

Author Response

Please, refer to the attached file for reviewers' comments 

Author Response File: Author Response.pdf

Round 2

Reviewer 1 Report

Comments and Suggestions for Authors The author needs to carefully address and respond to each of the proposed questions. In the response document, try to demonstrate how the relevant modifications were made, rather than simply stating that the changes were made according to the reviewers' suggestions. Additionally, in the revised manuscript, reference 3 is too outdated. The author needs to add the latest references as per the reviewers' recommendations. Moreover, it seems that references 4 and 5 are not formatted properly.  

Author Response

Please refer to the attached file for response to comments. Authors apologize for having omitted to emphasize the textual changes during the previous review round. Response file now includes textual variations. For Figure and Tables changes requested, those have been modified in the updated version of manuscript.

Author Response File: Author Response.pdf

Reviewer 2 Report

Comments and Suggestions for Authors

The authors presented machine learning methods to estimate the battery SOC. Given the rapid progress of the AI and EV industry, the study can have a considerable impact. The authors made progress to make the paper clearer. However, some improvements can be made.
1. On page 7, Figure 4. The authors should detail what each model means. For example, no one would understand what model "2.1" means.
2. On page 9, Figure 6. Since there are no results from neural networks, they can be removed from the figure, though mentioning them in the text is okay. 
3. On page 13, lines 370-374. These sentences are confusing and even incomplete. The author never mentioned LSTM-RNN before.

Author Response

Please, refer to the attached file for responses to Round 2 reviews.

Author Response File: Author Response.pdf

Reviewer 3 Report

Comments and Suggestions for Authors

Thank you for your efforts. no more comments. 

Author Response

Authors thank the reviewer for having expressed his useful comments aiming to improve the overall quality of the manuscript.

Round 3

Reviewer 1 Report

Comments and Suggestions for Authors

1. To enhance clarity, the curves in Figure 10  should be distinguished not only by color. While color distinction is visually effective for digital readers, have you considered the accessibility challenges for print-based audiences? In grayscale or printed formats, color-reliant distinctions may become ambiguous, potentially hindering reader engagement and long-term reference utility.
2. Additionally, in the revised manuscript, reference 3 is too outdated. The author needs to add the latest references as per the reviewers' recommendations. 

Author Response

Please, refer to the attached file

Author Response File: Author Response.pdf