A Methodology for Turbine-Level Possible Power Prediction and Uncertainty Estimations Using Farm-Wide Autoregressive Information on High-Frequency Data

Round 1

Reviewer 1 Report

Comments and Suggestions for Authors

The content of the work corresponds to the topic of the journal, but the work requires major revision before publication.

1. The design and text of the work must comply with the Instructions for Authors, including the abstract, conclusions and other sections. The abstract should be written in accordance with the Instructions for Authors. "The abstract should consist of one paragraph and correspond to the style of structured abstracts, but without headings: 1) Background: place the issue under consideration in a broad context and highlight the purpose of the study; 2) Methods: briefly describe the main methods or treatments used. Include any relevant pre-registration numbers, and the species and strains of any animals used; 3) Results: summarize the main findings of the paper; and 4) Conclusion: state the main conclusions or interpretations." The conclusion should be brief and contain some numerical values ​​obtained during the work.
2. Authors are encouraged to expand the background to include a more detailed analysis of the research, rather than simply listing the work of others. The aims and objectives of your research should flow logically from the review to highlight the relevance and necessity of your work.
3. I think the dataset is important knowledge and should be processed. How did you clean the data? What results did you get? How did you deal with outliers? The same goes for other procedures like labeling and normalization, which I haven't seen in your work. It would also be nice to back this up with numbers illustrating the results of such procedures.
4. Put your terminology in order. For example, MAE is generally accepted as mean absolute error. The work lacks other important metrics such as: RMSE , R², MAPE. Why?
5. The work does not present the results of comparison even with known simple methods.

Author Response

Comment 1:

The design and text of the work must comply with the Instructions for Authors, including the abstract, conclusions and other sections. The abstract should be written in accordance with the Instructions for Authors. "The abstract should consist of one paragraph and correspond to the style of structured abstracts, but without headings: 1) Background: place the issue under consideration in a broad context and highlight the purpose of the study; 2) Methods: briefly describe the main methods or treatments used. Include any relevant pre-registration numbers, and the species and strains of any animals used; 3) Results: summarize the main findings of the paper; and 4) Conclusion: state the main conclusions or interpretations." The conclusion should be brief and contain some numerical values ​​obtained during the work.

Response 1:

We appreciate this observation. The abstract was changed to create a better summary based on the instructions for authors. Also the Introduction was changed according to the Instructions for Authors.

Comment 2:

Authors are encouraged to expand the background to include a more detailed analysis of the research, rather than simply listing the work of others. The aims and objectives of your research should flow logically from the review to highlight the relevance and necessity of your work.

Response 2:

We thank you for this observation. A more detailed analysis of the research is added in the Introduction section. We compiled some more paragraphs on how the work is of importance and how it adds based on the literature reviewed. The objectives and aims are more thoroughly described on the introduction

Comment 3:

I think the dataset is important knowledge and should be processed. How did you clean the data? What results did you get? How did you deal with outliers? The same goes for other procedures like labeling and normalization, which I haven't seen in your work. It would also be nice to back this up with numbers illustrating the results of such procedures.

Response 3:

We thank the reviewer for this suggestion. In the updated manuscript under the Data section in methodology, it was added and cited how the methodology of annotation is thought. This is based on a methodology already published in the literature.

Comment 4:

Put your terminology in order. For example, MAE is generally accepted as mean absolute error. The work lacks other important metrics such as: RMSE , R², MAPE. Why?

Response 4:

Indeed, Mean Absolute Error is the metric used for prediction. We performed the experiments using the suggested metrics, and mention them in the results section in the updated manuscript. However, as the different metrics show similar results, we added them in the appendix for the interested reader, in order to maintain focus within the main text. R^2 was also added for the interested reader in the appendix.

Comment 5:

The work does not present the results of comparison even with known simple methods.

Response 5:

We thank you for this observation. In the updated manuscript, we clarified that our benchmark, or simple method, is the power curve provided by the manufacturer.

 

Changes added in responses are highlighted in Blue on the attachment

 

 

Author Response File: Author Response.pdf

Reviewer 2 Report

Comments and Suggestions for Authors

This paper gives prediction and uncertainty estimation to wind power using. The English writing is ok, while the logic and the contribution are to be improved. It is more like statistical analysis instead of actual prediction.

The following comments are to be responded by the authors,

1. In the title, “Spatio-Temporal Information” is fully supported by the content. For the spatial simulation to the wind farm, the layout, the landscape, weather condition, etc. are to be considered.
2. In the Abstract, “dynamic interactions” is usually applied for the system described by differential equations. It is not fully verified in the paper. It is not the coherence.
3. The prediction methods are not explained clearly. The probabilistic method provides the statistics impression instead of the wind power data at given time. The machine leaning method and its originality presented by the authors are to be enhanced.
4. The simulation analysis is quite poor to show the accuracy and advantage of this paper. It is more like curve fitting which is quite basic and easy. The framework of prediction is 600 hours, which is actually neither needed nor practical. Predicting the average power is easy, but of little engineering value.

Author Response

Comment 1:

In the title, “Spatio-Temporal Information” is fully supported by the content. For the spatial simulation to the wind farm, the layout, the landscape, weather condition, etc. are to be considered.

Response 1:

We appreciate this observation. As the model uses correlations that implicitly consider multiple turbines and trends over time, we consider this a spatio-temporal model. We clarified this property of the model in the introduction and methodology.

Comment 2:

In the Abstract, “dynamic interactions” is usually applied for the system described by differential equations. It is not fully verified in the paper. It is not the coherence.

Response 2:

We thank the reviewer for this comment, as it indeed could create confusion. Specifically, we considered ‘dynamic interactions’ as a property of the model, as it uses temporal information of how the system behaves. Specifically, the model uses past and immediate information from SCADA. We clarified this in the manuscript.

Comment 3:

The prediction methods are not explained clearly. The probabilistic method provides the statistics impression instead of the wind power data at given time. The machine leaning method and its originality presented by the authors are to be enhanced.

Response 3:

We appreciate the comment. Indeed, it is expected to provide a probabilistic method, machine learning methodologies were not entirely discussed. The prediction in itself is based on the small park-wide correction of the power curve. The probabilistic method proposed is compiling information from the entire park for the last 150 seconds, with addition to the power curve information. The method is assuming that by their association, expressed by the covariance. This constitutes the novelty of the method. The main takeaway that is intended to show is the capacity of 1) Calculate uncertainties based on farm-wide data. This is further clarified in the manuscript. 2) Show that inference is possible based on any fit of the active power.

Comment 4

The simulation analysis is quite poor to show the accuracy and advantage of this paper. It is more like curve fitting which is quite basic and easy. The framework of prediction is 600 hours, which is actually neither needed nor practical. Predicting the average power is easy, but of little engineering value.

Response 4:

We appreciate the observation. The prediction can be qualified as a now-casting. There are no curves fitted, the curves are already taken from the manufacturer, The idea is to add the probabilistic component and also have an analytical inclusion of the correlation between turbines. No curves were fitted for prediction, but the main focus is the calculation of uncertainties on real time. In addition to this, an initial naïve esitmation of the probability of under-performance is added.

 

Extracts changed are presented in Blue on supporting documents

Author Response File: Author Response.pdf

Round 2

Reviewer 1 Report

Comments and Suggestions for Authors

I must note that most of my comments have not been properly addressed or have only been partially implemented. I would like to describe the main disadvantages in detail once again:

Please follow the author guidelines!!!
Despite the stated changes, the abstract still does not meet the required structure (a clearly formulated problem statement, methods, results, and conclusions presented in one logically organized paragraph). The conclusion section lacks specific numerical data to support your conclusions. These sections require significant revision and should be brought into full compliance with the journal's requirements.

Background
The additional material in the introduction does not resolve the issue—the background remains superficial. Most of the cited works by other authors are listed without any explanation of their contributions, advantages, or shortcomings. It is unclear how the presented background logically leads to the formulation of the aims and objectives of your own study. The introduction requires revision.

Dataset
Your response did not resolve the key issue: detailed descriptions of procedures such as data cleaning, outlier handling, labeling, and normalization are missing. A mere reference to a previous publication instead of your own description is not acceptable. You need to clearly describe how you processed your dataset, including the numerical results of these procedures. At present, this section is severely underdeveloped and lacks sufficient justification.

Metrics 
The presentation of metrics (RMSE, MAPE, R²) is unsatisfactory. It is unclear why important metrics are moved to the appendix and not discussed in the main text. This gives the impression that these metrics are not considered significant, which should not be the case. You need to return the analysis of all key metrics to the main text and strictly adhere to accepted terminology.

Comparison with Simple Methods
Your explanation regarding the comparison with simple methods is incomplete. You must provide numerical results comparing your approach with the chosen benchmark directly in the main text of the article so that the advantages of your method can be objectively assessed.

I strongly recommend carefully proofreading the manuscript and correcting obvious flaws and mistakes, such as those identified, for example, in line 450.

Author Response

Response to Reviewer 1 Comments
1. Summary
Thank you very much for taking the time to review this manuscript. Please find the detailed responses below and the corresponding revisions in track changes in the re-submitted files. The changes performed for this draft are highlighted in blue.
2. Questions for General Evaluation	Reviewer’s Evaluation	Response and Revisions
Does the introduction provide sufficient background and include all relevant references?	Must be improved
Are all the cited references relevant to the research?	Not applicable
Is the research design appropriate?	Must be improved
Are the methods adequately described?	Must be improved
Are the results clearly presented?	Must be improved
Are the conclusions supported by the results?	Must be improved
3. Point-by-point response to Comments and Suggestions for Authors
Comments 1: Please follow the author guidelines!!! Despite the stated changes, the abstract still does not meet the required structure (a clearly formulated problem statement, methods, results, and conclusions presented in one logically organized paragraph). The conclusion section lacks specific numerical data to support your conclusions. These sections require significant revision and should be brought into full compliance with the journal's requirements.
Response 1: We would like to thank the reviewer for this observation. This has been changed in the Abstract more appropriately. The abstract now contains numerical results with the benchmark. This can be seen around line 18 to 21. Specific numerical data supporting the improvements were added on the Conclusions section. This is remarkable from line 580 to 585.
Comments 2: Background The additional material in the introduction does not resolve the issue—the background remains superficial. Most of the cited works by other authors are listed without any explanation of their contributions, advantages, or shortcomings. It is unclear how the presented background logically leads to the formulation of the aims and objectives of your own study. The introduction requires revision.
Response 2: We have, accordingly, revised the shortcomings in our Background section. We appreciate the observation. Changes are noticeable in lines 60 to 65, 71 to 72, 90 to 91, and in general landing the idea from line 101 to line 118.   Comment 3: Dataset Your response did not resolve the key issue: detailed descriptions of procedures such as data cleaning, outlier handling, labeling, and normalization are missing. A mere reference to a previous publication instead of your own description is not acceptable. You need to clearly describe how you processed your dataset, including the numerical results of these procedures. At present, this section is severely underdeveloped and lacks sufficient justification.   Response 3: We have accordingly revised these methodological aspects. The annotation methodology is more thoroughly explained between lines 171 and 194. The scaling of the data is also more thoroughly addressed between lines 200 to 213. The entirety of the 2.1 section has been changed accordingly.   Comment 4: Metrics  The presentation of metrics (RMSE, MAPE, R²) is unsatisfactory. It is unclear why important metrics are moved to the appendix and not discussed in the main text. This gives the impression that these metrics are not considered significant, which should not be the case. You need to return the analysis of all key metrics to the main text and strictly adhere to accepted terminology.   Response 4: We appreciate this comment since it has helped adding clarity. All the error metrics were added in the main text. Interpretations of them were also added. They were added from line 343 to line 351 as their definitions. Overall in section 3.1 there has been a thorough change of this. Further they were also added from line 372 to 404 with interpretations, numerical results and plots.   Comment 5: Comparison with Simple Methods Your explanation regarding the comparison with simple methods is incomplete. You must provide numerical results comparing your approach with the chosen benchmark directly in the main text of the article so that the advantages of your method can be objectively assessed.   Response 5: We have made clear how we are making marginal improvements on the power curve that the manufacturer has provided us. Comparisons have been added in results section and it. This is specially clear from line 380 to 404. Also the improvements are further discussed on line 521 and line 596, and how it adds to the current methods. This is addressed in. Section 3.1. Also a whole new section with a specific use case was added, in order to show possible use cases.

 

 

Author Response File: Author Response.pdf

Reviewer 2 Report

Comments and Suggestions for Authors

The response seems to orally justify instead of actually revision. Few comments are directly responded or improved in the paper. By this way, the models in the paper can not fully support the contribution or feasibility of the paper. The simulation results also do not show their effectiveness and merit compared with existing models. The expressions “Spatio-Temporal Information”, “dynamic interaction”, etc. exaggerate the research range of this paper, possibly misleading the future readers.

Author Response

Response to Reviewer 2 Comments
1. Summary
Thank you very much for taking the time to review this manuscript. Please find the detailed responses below and the corresponding revisions in track changes in the re-submitted files. The changes performed for this draft are highlighted in blue
2. Questions for General Evaluation	Reviewer’s Evaluation	Response and Revisions
Does the introduction provide sufficient background and include all relevant references?	Can be improved
Are all the cited references relevant to the research?	Not applicable
Is the research design appropriate?	Must be improved
Are the methods adequately described?	Must be improved
Are the results clearly presented?	Can be improved
Are the conclusions supported by the results?	Yes
3. Point-by-point response to Comments and Suggestions for Authors   Here we add the comments from the first revision   Comments 0: The response seems to orally justify instead of actually revision. Few comments are directly responded or improved in the paper. By this way, the models in the paper can not fully support the contribution or feasibility of the paper. The simulation results also do not show their effectiveness and merit compared with existing models. The expressions “Spatio-Temporal Information”, “dynamic interaction”, etc. exaggerate the research range of this paper, possibly misleading the future readers.   Response 0: We would like to thank the reviewer for this comment and the past ones. Since they are related to the first ones provided, these comments were re-addressed according to the observations from the last review. In order to not keep them vague we will use as basis observations from the last time.

Comments 1: In the title, “Spatio-Temporal Information” is fully supported by the content. For the spatial simulation to the wind farm, the layout, the landscape, the weather condition, etc. are to be considered.

 

Response 1: We thank the reviewer for the observation. We decided to use more appropriate wording to express the relationships in the input and output. This is addressed in lines 3-6. Also from lines 527 to 543. It was also taken out of the title and it was opted to use the more appropriate term, autoregressive.

 

Comments 2: In the Abstract, “dynamic interactions” is usually applied for the system described by differential equations. It is not fully verified in the paper. It is not the coherence.

 

Response 2: These terms, just like in the last observation were addressed. Can be seen from lines 476 to 480. The word dynamic or the phrase dynamic interaction is being avoided.

 

Comments 3: The prediction methods are not explained clearly. The probabilistic method provides the statistics impression instead of the wind power data at given time. The machine leaning method and its originality presented by the authors are to be enhanced.

 

Response 3: This has been addressed. There is inclusion of the actual purpose of the inclusion of statistical methods. The current work constitutes a way of White Box Machine Learning using power curves as prior function. This is further added for clarification in the Results section and Discussion has been clarified surrounding that. Especially from line 546 to 568

 

Comments 4: The simulation analysis is quite poor to show the accuracy and advantage of this paper. It is more like curve fitting which is quite basic and easy. The framework of prediction is 600 hours, which is actually neither needed nor practical. Predicting the average power is easy, but of little engineering value

 

Response 4: We appreciate this observation. We want to highlight that our study is not based on simulation. This study is based on a Wind Farm SCADA data. The advantage was highlighted, not only based on the predictive power which has an improve but also it shows the advantage of using the proposed uncertain estimation methodology. The data section is more thorough, explaining where the data comes from, which is not a simulation. We also clarify that this is a model for normal behavior. This is clarified more thoroughly in the Introduction.

 

Author Response File: Author Response.pdf

Round 3

Reviewer 1 Report

Comments and Suggestions for Authors

The work can be published in present form.

Author Response

Response to Reviewer 1 Comments
1. Summary
Thank you very much for taking the time to review this manuscript. Please find the detailed responses below and the corresponding revisions/corrections highlighted/in track changes in the re-submitted files.
2. Questions for General Evaluation	Reviewer’s Evaluation	Response and Revisions
Does the introduction provide sufficient background and include all relevant references?	Can be improved
Are all the cited references relevant to the research?	Can be improved
Is the research design appropriate?	Can be improved
Are the methods adequately described?	Can be improved/
Are the results clearly presented?	Can be improved
Are the conclusions supported by the results?	Can be improved
3. Point-by-point response to Comments and Suggestions for Authors
Comments 1:   The work can be published in present form.
Response 1: We appreciate the time invested by the reviewer on this manuscript.

 

 

 

 

 

 

 

 

 

 

Author Response File: Author Response.pdf

Reviewer 2 Report

Comments and Suggestions for Authors

The paper may be accepted now.

Originality of the paper is to be enhanced. Correspondence of the simulation results with the contributions is to be highlighted.

Author Response

Response to Reviewer 2 Comments
1. Summary
Thank you very much for taking the time to review this manuscript. Please find the detailed responses below and the corresponding revisions/corrections highlighted/in track changes in the re-submitted files.
2. Questions for General Evaluation	Reviewer’s Evaluation	Response and Revisions
Does the introduction provide sufficient background and include all relevant references?	Yes
Are all the cited references relevant to the research?
Is the research design appropriate?	Can be improved
Are the methods adequately described?	Can be improved/
Are the results clearly presented?	Can be improved
Are the conclusions supported by the results?	Yes
3. Point-by-point response to Comments and Suggestions for Authors
Comments 1: The paper may be accepted now. Originality of the paper is to be enhanced. Correspondence of the simulation results with the contributions is to be highlighted.
Response 1: We appreciate the time invested by the reviewer on this manuscript. With the originality and contributions, they have been highlighted in the manuscripts by adding lines in the Discussion section and Conclusion.

Author Response File: Author Response.pdf