S-Wave Velocity Forecasting Using Drill Cuttings and Deep Hybrid Neural Networks: A Case Study on a Tight Glutenite Reservoir in Mahu Sag, Junggar Basin
Round 1
Reviewer 1 Report
The topic is interesting—good contribution to the literature.
1. Literature summary should be enhanced a bit. For example, LSTM examples could be added (like: https://www.mdpi.com/1424-8220/22/12/4419).
2. Introduction should be sharpened. The gains of the work should be added with bullets concretely.
3. Methodly could be illustrated well. With the current form, it's not described in detail enough. Each step of the methods could be summarized with bullets.
4. The results of the work should be improved well to highlight the importance. The graphs and figures are enough, but the results should be explained more.
5. The abstract should be enhanced with the concrete comparison results of the proposed work.
6. The language of the paper is acceptable.
7. Although it is close to publication, overall improvements need for publication. The current version is not enough for publication.
Author Response
- Literature summary should be enhanced a bit. For example, LSTM examples could be added (like: https://www.mdpi.com/1424-8220/22/12/4419).
Answer: thanks for your suggestion, I have cited the article: https://www.mdpi.com/1424-8220/22/12/4419, as shown in Line 435.
- Introduction should be sharpened. The gains of the work should be added with bullets concretely.
Answer: Considering the literature on existing methods and also considering the fact that the we have published many similar methods, I sort out the current problems with bullets about the Vs prediction and show clearly the main contribution and significant novelty of our work., as shown in Line 64-83.
- Methodology could be illustrated well. With the current form, it's not described in detail enough. Each step of the methods could be summarized with bullets.
Answer: We appreciate your kind suggestions, we have a restructuration of the paper, a) Introduction b) Methodology c) Results d) discussion e) Conclusion. Data collection, data analysis, and model construction are systematically introduced, and the studied case is given to facilitate a better presentation of our approach.
- The results of the work should be improved well to highlight the importance. The graphs and figures are enough, but the results should be explained more.
Answer: We are grateful for the suggestion. As suggested by the reviewer, we have added more details to the figure description as shown in Line 337-356. In addition, we present the advantages and disadvantages of the model in the discussion section.
- The abstract should be enhanced with the concrete comparison results of the proposed work.
Answer: Thanks for your suggestions, we rewrite the abstract and add model input parameters, evaluation indices, and model performance comparison.
- The language of the paper is acceptable.
Answer: Thank you very much for your recognition of our manuscript. In order to avoid affecting readers due to language issues, we have checked typing errors throughout the whole paper.
- Although it is close to publication, overall improvements need for publication. The current version is not enough for publication.
Answer: Thank you for your careful reading of our manuscript. We hereby resubmit the revised manuscript and hope that all corrections are satisfactory.
Author Response File: 
Author Response.docx
Reviewer 2 Report
The authors have presented a research paper entitled "S-wave Velocity Forecasting Using Drill Cuttings and deep hybrid neural networks". This study proposed CNN-SGRU hybrid model. After a careful reading of the paper. Reviewer has some comments as follows:
1. The proposed method is a combination of recently published results. Considering the vast amount of literature on existing methods and also considering the fact that the authors have published many similar methods, the novelty especially compared to their previous work should be explained. Authors need to show clearly the main contribution and significant novelty of their work in highlights and research content.
2. The obtained results still have some limitations as shown Figure 14, as evidenced by the fact that the results are still not really optimal. What does the author comment on the results achieved? And how will the author improve in the future? This leads to my next question.
3. In the future, can this method be combined with the optimization algorithm to enhance the accuracy of the model? Such as:
ü A new metaheuristic algorithm: Shrimp and Goby association search algorithm and its application for damage identification in large-scale and complex structures. Advances in Engineering Software, 176, 103363.
ü A surrogate-assisted stochastic optimization inversion algorithm: Parameter identification of dams. Advanced Engineering Informatics, 55, 101853.
ü Base resistance of super-large and long piles in soft soil: performance of artificial neural network model and field implications. Acta Geotechnica, 1-21.
ü A new movement strategy of grey wolf optimizer for optimization problems and structural damage identification. Advances in Engineering Software, 173, 103276.
4. The abstract should be rewritten. Too many employed methods are described in detail but the novelty and contribution to the state-of-the-art is not addressed. Several sentences should belong in introduction section instead of Abstract.
5. Authors should modify the captions of all Figures. They look academically unprofessional, for example Figure.1.
6. The authors should check typing errors throughout the whole paper. English style should also be improved.
Author Response
- The proposed method is a combination of recently published results. Considering the vast amount of literature on existing methods and also considering the fact that the authors have published many similar methods, the novelty especially compared to their previous work should be explained. Authors need to show clearly the main contribution and significant novelty of their work in highlights and research content.
Answer: We appreciate your kind suggestions. Based on literature data, we sort out the current problems with bullets about the Vs prediction as shown in Line 64-83. The innovation lies in the model input, replacing log data with drill cuttings data. In addition, to consider the spatiotemporal relationships between data, a hybrid neural network model is used to learn the data relationships.
- The obtained results still have some limitations as shown Figure 14, as evidenced by the fact that the results are still not really optimal. What does the author comment on the results achieved? And how will the author improve in the future? This leads to my next question.
Answer: We deeply appreciate your suggestions. We explore how to improve model prediction accuracy in the discussion section and try to find new optimization algorithms, as shown in Line 374-382.
- In the future, can this method be combined with the optimization algorithm to enhance the accuracy of the model? Such as:
ü A new metaheuristic algorithm: Shrimp and Goby association search algorithm and its application for damage identification in large-scale and complex structures. Advances in Engineering Software, 176, 103363.
ü A surrogate-assisted stochastic optimization inversion algorithm: Parameter identification of dams. Advanced Engineering Informatics, 55, 101853.
ü Base resistance of super-large and long piles in soft soil: performance of artificial neural network model and field implications. Acta Geotechnica, 1-21.
ü A new movement strategy of grey wolf optimizer for optimization problems and structural damage identification. Advances in Engineering Software, 173, 103276.
Answer: cited as shown in Line 516-522.
- The abstract should be rewritten. Too many employed methods are described in detail but the novelty and contribution to the state-of-the-art is not addressed. Several sentences should belong in introduction section instead of Abstract.
Answer: Thanks for your suggestions, we rewrite the abstract and add model input parameters, evaluation metrics, and model performance comparison.
- Authors should modify the captions of all Figures. They look academically unprofessional, for example Figure.1.
Answer: We apologize for any confusion caused and appreciate the valuable suggestions. We have carefully checked the captions of all Figures.
- The authors should check typing errors throughout the whole paper. English style should also be improved.
Answer: We sincerely apologize for the inconvenience of reading caused by grammatical problems and typos. Accordingly, we have corrected the problems and all similar ones throughout the manuscript without altering the paper’s original meaning.
Author Response File: Author Response.docx
Reviewer 3 Report
S-wave Velocity Forecasting Using Drill Cuttings and deep hybrid neural networks
The present study discussed very important parameter (S-wave Velocity), using deep hybrid neural network coupling the neural network (CNN), stacked gated recurrent unit (SGRU) where the inputs to the model are drill cuttings features. The paper is well written, the methodology is well done and in the scope of the journal, I suggest major revision before acceptance.
Comments:
Title : I believe that it will be better authors to add the case study (study area) is good for the paper the apparition and visibility in the database.
Abstract :
Authors should add some detail in methodology for the drill cuttings features, it will be more evidence.
Authors should add the performances assessment metrics in the abstract.
Keywords
Add the study area to the keywords
Methodology:
Authors should add some reservoir descriptions... before drilling cutting parameters acquisition
Figure.1 Please check the title
Line 90
Authors said that "many scholars have proven that mineral compositions and pore characteristics are the key factors affecting Vs [35-37], ", authors used specific references on tight gas reservoirs ( we cannot generalize on all reservoirs type!!!, also reference 37 is not peer reviewed reference (workshop) we cannot based on!!!!, also, "the reservoir fluids have no effect on Vs[38]" I believe that the reservoir fluids have direct effect on Vs, we used the (Vp/Vs) ratio as seismic attribute to determine fluid!!! Be careful for this.
Figure, 2 and 3, authors should well highlight the scale and in the Fig2 (b) is totally missed and add legend for this picture. (black and blue!!)
Figure 8 9 10 authors should change the depth in the vertical axe in conventional form.
Figure 11 the Pearson correlation analysis is not sufficient authors should calculate the P-value with target Vs and the other parameters.
Line 264-265 Authors should describe the normalization and transformation methods used, and how the dataset was divided into the training set, validation set and test set and the percentage ?
Table1. title should check
Figure 12 in the figure 12 authors display only two step (training and validation) without test, and in the previous section authors said that dataset was divided into 3 steps!!!!
Figure 14: based on the figure 14 it clear that authors do not take a randomly in the separation of the database, they take the 70 % colored blue for training, 15% validation colored green and 15 test colored red, the separation for the database is not adequate, what your explanation? And the same remark figures 8 9 10 authors should change the depth in the vertical axe in conventional form.
Authors should add a table summarize all the performance metrics of the different ML algorithm used not only graphical visualization.
Conclusion. Authors should add results with performance metrics comparasion.
General comment
1. Authors should separate (results and discussion from the case study) a restructuration of the paper is more than requested
a) Introduction
b) Materials and methods
c) Results and discussion
d) Conclusion
2. Authors should precise the input variables used in the models
3. in the discussion section, add the merits and demerits of your work
Author Response
Comments:
Title: I believe that it will be better authors to add the case study (study area) is good for the paper the apparition and visibility in the database.
Answer: Thanks for your kind suggestion, we have added the study area.
Abstract:
Authors should add some detail in methodology for the drill cuttings features, it will be more evidence.
Authors should add the performances assessment metrics in the abstract.
Answer: Thanks for your suggestions, we rewrite the abstract and add model input parameters, evaluation metrics, and model performance comparison.
Keywords
Add the study area to the keywords
Answer: thanks for your suggestion, we have added the study area in the keyword section.
Methodology:
Authors should add some reservoir descriptions... before drilling cutting parameters acquisition
Answer: Since the research block is in the early stages of development, many details need to be kept confidential, so no further details of the block can be provided. However, we give a brief description of the block in section 3.1.
Figure.1 Please check the title
Answer: Thank you for your careful reading of our manuscript. We have changed the right title as shown in Line 105.
Line 90:Authors said that "many scholars have proven that mineral compositions and pore characteristics are the key factors affecting Vs [35-37], ", authors used specific references on tight gas reservoirs ( we cannot generalize on all reservoirs type!!!, also reference 37 is not peer reviewed reference (workshop) we cannot based on!!!!, also, "the reservoir fluids have no effect on Vs[38]" I believe that the reservoir fluids have direct effect on Vs, we used the (Vp/Vs) ratio as seismic attribute to determine fluid!!! Be careful for this.
Answer：Thanks for your kind suggestions.
There are few studies on the Vs and Vp of glutenite reservoirs, and no rock physics model for Vs prediction in glutenite reservoirs has been proposed. As shown in Fig.7, the ultrasonic velocities in glutenite samples have a large distribution range, however, the ultrasonic velocities in sandstone samples are around 4000 m/s, and the value range is more concentrated. Overall, there is some error in applying the model of sandstone to glutenite, but it also has some reference value.
As for the “the reservoir fluids have no effect on Vs[38]", we changed “the reservoir fluids have no effect on Vs[38]" into “the S-wave does not propagate in the reservoir fluids”, as shown in Line 108.
Source: Liu J , Ge H , Mou S , et al. Characterization of meso-structure of glutenite reservoirs by ultrasonic characteristics and the velocity heterogeneity[J]. Journal of Petroleum Science and Engineering, 2022, 208:109436-.
Figure, 2 and 3, authors should well highlight the scale and in the Fig2 (b) is totally missed and add legend for this picture. (black and blue!!)
Answer: Thanks for your suggestions, we have added the legend for Figure 2 and 3.
Figure 8 9 10 authors should change the depth in the vertical axe in conventional form.
Answer: Thanks for your suggestions. We have changed the depth in the vertical axe.
Figure 11 the Pearson correlation analysis is not sufficient authors should calculate the P-value with target Vs and the other parameters.
Answer: In this study, we want to establish the relationship between drill cuttings characteristics and Vs, so we did not carry out the correlation analysis between Vp and Vs.
Line 264-265 Authors should describe the normalization and transformation methods used, and how the dataset was divided into the training set, validation set and test set and the percentage ?
Answer: the normalization and transformation methods are Z-score normalization and input-output pairs transformation as shown in Line 230. The whole dataset is divided into the training set, validation set and test set in the ratio of 70%, 15% and 15% with sampling depth as shown in Line 239.
Table1. title should check
Answer: Thank you for your careful reading of our manuscript. We have changed the right title as shown in Line 361.
Figure 12 in the figure 12 authors display only two step (training and validation) without test, and in the previous section authors said that dataset was divided into 3 steps!!!!
Answer: Thank you very much for your concern. Figure 12 shows the variation of loss function with respect to epoch during the model training, including the training set and the validation set. The training set, validation set and test set are divided for different purposes. The test set is used to test the forecasting performance of a machine learning model on unseen data, so it cannot be involved in the model training and optimization, and does not appear in Figure 12. The validation set and the training set are used to show the convergence of loss function to avoid overfitting, so they are shown in Figure 12.
Figure 14: based on the figure 14 it clear that authors do not take a randomly in the separation of the database, they take the 70 % colored blue for training, 15% validation colored green and 15 test colored red, the separation for the database is not adequate, what your explanation? And the same remark figures 8 9 10 authors should change the depth in the vertical axe in conventional form.
Answer: Because of the spatial and temporal correlation between drill cuttings at different locations, random sampling cannot be used, and in our study, we used sequential sampling according to the sampling depth. In addition, we have changed the depth in the vertical axe for Figure 8, 9, and 10.
Authors should add a table summarize all the performance metrics of the different ML algorithm used not only graphical visualization.
Answer: Thanks for your suggestion, we have added the table as shown in Table 2.
Conclusion. Authors should add results with performance metrics comparison.
Answer: thanks for your suggestion, we added a comparison of different model performance indices.
General comment
- Authors should separate (results and discussion from the case study) a restructuration of the paper is more than requested
- a) Introduction
- b) Materials and methods
- c) Results and discussion
- d) Conclusion
Answer: We appreciate your kind suggestions, we have a restructuration of the paper, a) Introduction b) Methodology c) Results d) discussion e) Conclusion.
- Authors should precise the input variables used in the models
Answer: we have added the input variables in the abstract and also add in 3.1 Section as shown in Line270.
- in the discussion section, add the merits and demerits of your work
Answer: We are grateful for the suggestion. we present the advantages and disadvantages of the model in the discussion section.
Author Response File: Author Response.docx
Reviewer 4 Report
-Was the analysis also performed on other types of sediments?
-How could the relative error of the method be reduced?
-Have other implementations of the algorithm been tried?
-It would be appropriate to add a comparison with another work.
Author Response
Comments and Suggestions for Authors
-Was the analysis also performed on other types of sediments?
Answer: At present, we have only carried out drill cuttings analysis for glutenite reservoirs, but not for reservoirs with other different lithologies. This issue may affect the generalization ability of the proposed model, so we add the discussion section as shown in Line 362.
-How could the relative error of the method be reduced?
Answer: In the next study, we want to optimize the drill cuttings’ sampling interval and find new optimization algorithms.
-Have other implementations of the algorithm been tried?
Answer: We used a similar algorithm applied to fractured horizontal well production prediction and obtained better prediction results, and now, try to study the application of this algorithm in the field of rock physics.
-It would be appropriate to add a comparison with another work.
Answer: Thanks for your kind suggestions. We hope to have a larger volume of data to train and validate our proposed model, but the research is still in the early stage, we are carrying out drill cuttings analysis of different lithological reservoirs, and we hope we can propose a more perfect model later.
We hereby resubmit the revised manuscript and hope that all corrections are satisfactory. Please feel free to contact us with any questions and we look forward to your decision.
Author Response File: Author Response.docx
Round 2
Reviewer 1 Report
The authors updated the article well. I think its ready for publication.
Reviewer 2 Report
I have no further comments
Reviewer 3 Report
Authors replied seriously to all comments and suggestions, the ms had been highly improved in this current form I believe it acceptable for publication. only one comment; authors should add description sequential sampling used in the final version for more clarification.
