Prediction of Slope Safety Factor Based on Attention Mechanism-Enhanced CNN-GRU
Round 1
Reviewer 1 Report
Comments and Suggestions for AuthorsThis paper proposes a new method for predicting slope safety factors based on multiple techniques to improve the accuracy of predictions. This study may provide reference for the safety control of slopes. However, there is still room for improvement in the paper as follows:
(1) The abstract is too lengthy. The common writing style for abstracts is to describe your research background in a short paragraph, then clarify your main research content, briefly describe your "important and major" research results, and finally briefly explain the research significance. Of course, you can also choose not to follow this style, but the prerequisite is that you write it well. In the abstract section of this article, it is recommended to adjust the structure of the abstract (adjust the background to the first paragraph), streamline unnecessary content (such as: your statistical indicators do not need to be listed one by one), and delete some unnecessary content to streamline the abstract.
(2) Is there a strong logical relationship between slope engineering and "material demand" in line 37? Every sentence needs to be carefully considered, and at the same time, the tense of this sentence may also be problematic.
(3) Line 38, the increase in the number of slope engineering projects will not lead to an increase in the frequency of slope failure and instability, but rather the overall number of accidents. Similar issues should be avoided.
(4) Line 46, "Other methods" should be based on references or replaced with "et al.".
(5) In the introduction, many abbreviations are not explained when they first appear, which is non-standard and readers will not be able to understand their meanings. In addition, the "GRU" appearing on line 77 is only explained on line 81?
Overall, this is a valuable article that may provide reference for predicting the stability of slopes. But it still needs further improvement: the language section of the entire text needs to be modified, especially paying attention to tense and voice, as well as logic, to improve image clarity.
After completing the above modifications, I can agree to the publication of this paper.
Comments on the Quality of English Language(1) Lines 91-106 are too lengthy ,suggest necessary streamlining.
(2) Lines 114 “F12 is the,”, there should be an extra "," here, please make the correction.
(3) Lines 116, Data acquisition - noun? Analysis - simple present tense? "And processing - present continuous tense?", This should be unified and expressed in simple present tense or present continuous tense.
(4) Lines 127, “The safety factor” should be replaced with “Factor of Safety”
(5) The order of lines 205 and 241 is duplicated, please correct the image number (including the text).
(6) Lines 247 and 259, "Where," "Where:", please maintain consistent style.
Author Response
Comment 1：The abstract is too lengthy. The common writing style for abstracts is to describe your research background in a short paragraph, then clarify your main research content, briefly describe your "important and major" research results, and finally briefly explain the research significance. Of course, you can also choose not to follow this style, but the prerequisite is that you write it well. In the abstract section of this article, it is recommended to adjust the structure of the abstract (adjust the background to the first paragraph), streamline unnecessary content (such as: your statistical indicators do not need to be listed one by one), and delete some unnecessary content to streamline the abstract.
Response:Thanks for your comments and suggestions.We have provided separate structural and modeling process diagrams for the methods used.
Comment 2：Is there a strong logical relationship between slope engineering and "material demand" in line 37? Every sentence needs to be carefully considered, and at the same time, the tense of this sentence may also be problematic.
Response:Thanks for your comments and suggestions.We have provided separate structural and modeling process diagrams for the methods used.
Comment 3：Line 38, the increase in the number of slope engineering projects will not lead to an increase in the frequency of slope failure and instability, but rather the overall number of accidents. Similar issues should be avoided.
Response:Thanks for your comments and suggestions.We have provided separate structural and modeling process diagrams for the methods used.
Comment 4：Line 46, "Other methods" should be based on references or replaced with "et al.".
Response:Thanks for your comments and suggestions.We have revised the questions you raised. If you have any further questions, please let us know in a timely manner. Thank you again for your suggestions.
Comment 5：In the introduction, many abbreviations are not explained when they first appear, which is non-standard and readers will not be able to understand their meanings. In addition, the "GRU" appearing on line 77 is only explained on line 81?
Response:Thanks for your comments and suggestions.We have revised the questions you raised. If you have any further questions, please let us know in a timely manner. Thank you again for your suggestions.
Comment 6：Lines 91-106 are too lengthy ,suggest necessary streamlining.
Response:Thanks for your comments and suggestions.We have revised the question you raised.
Comment 7：Lines 114 “F12 is the,”, there should be an extra "," here, please make the correction.
Response:Thanks for your comments and suggestions.We have revised the question you raised.
Comment 8：Lines 116, Data acquisition - noun? Analysis - simple present tense? "And processing - present continuous tense?", This should be unified and expressed in simple present tense or present continuous tense.
Response:Thanks for your comments and suggestions.We have revised the questions you raised. If you have any further questions, please let us know in a timely manner. Thank you again for your suggestions.
Comment 9：Lines 127, “The safety factor” should be replaced with “Factor of Safety”
Response:Thanks for your comments and suggestions.We have revised the questions you raised. If you have any further questions, please let us know in a timely manner. Thank you again for your suggestions.
Comment 10：The order of lines 205 and 241 is duplicated, please correct the image number (including the text).
Response:Thanks for your comments and suggestions.We have revised the questions you raised. If you have any further questions, please let us know in a timely manner. Thank you again for your suggestions.
Comment 11：Lines 247 and 259, "Where," "Where:", please maintain consistent style.
Response:Thanks for your comments and suggestions.We have revised the question you raised.
Reviewer 2 Report
Comments and Suggestions for Authors1. Is there a classical slope stability model (limit / static equilibrium) adopted, if so, how it has been considered.
2. The abstract is too descriptive it needs to be simplified
3. Most of the figures are not clear, use high resolution images
4. Figure:1 can be simplified to a simple 2-D image clearly indicating the parameters.
5. About 200 (183 data sets) is quite a large number how the variation in rock type and geotechnical conditions other then the mention parameters have been addressed.
6. Units for the slope stability parameters are required to be mentioned in the start of the manuscript.
7. The seismic effects are not considered at this stage, can they be incorporated in future?
8. What are the recommendations for the design of slope stability infrastructure.
9. How field engineers will adopt the study results?
Comments on the Quality of English LanguageOverall the language is fine, however, the manuscript needs a through proofread.
Author Response
Comment 1：Is there a classical slope stability model (limit / static equilibrium) adopted, if so, how it has been considered.
Response:Thanks for your comments and suggestions.The data used in this article are all obtained through literature retrieval. Through tracing the source, it was found that different data sources use different models to calculate the safety factor, making it impossible to provide a reasonable explanation. The machine learning used in this article fits the input and output parameters, rather than using the classic slope stability model. If you have any further questions, please let us know in a timely manner. Thank you again for your suggestions.
Comment 2：The abstract is too descriptive it needs to be simplified
Response:Thanks for your comments and suggestions.We have revised the questions you raised. If you have any further questions, please let us know in a timely manner. Thank you again for your suggestions.
Comment 3：Most of the figures are not clear, use high resolution images
Response:Thanks for your comments and suggestions.We have revised the questions you raised. If you have any further questions, please let us know in a timely manner. Thank you again for your suggestions.
Comment 4：Figure:1 can be simplified to a simple 2-D image clearly indicating the parameters.
Response:Thanks for your comments and suggestions.We have revised the questions you raised. If you have any further questions, please let us know in a timely manner. Thank you again for your suggestions.
Comment 5：About 200 (183 data sets) is quite a large number how the variation in rock type and geotechnical conditions other then the mention parameters have been addressed.
Response:Thanks for your comments and suggestions.The questions you raised are very valuable. In practical engineering, there are many factors that affect slope stability. The data used in this study were collected using literature retrieval methods, making it difficult to collect comprehensive information on changes in slope rock types and geotechnical conditions. At the same time, it is difficult to convert qualitative factors into quantitative factors. In subsequent research, we will consider more influencing factors and refine them. For the geotechnical changes of different slopes, due to the robustness of the model, it is not possible to train only for a certain region or slope. In order to more easily predict the situation of different slopes, at present, only the geotechnical conditions of different slopes can be parameterized. However, in future research, it can be considered to improve it.If you have any further questions, please let us know in a timely manner. Thank you again for your suggestions.
Comment 6：Units for the slope stability parameters are required to be mentioned in the start of the manuscript.
Response:Thanks for your comments and suggestions.We have revised the questions you raised. If you have any further questions, please let us know in a timely manner. Thank you again for your suggestions.
Comment 7：The seismic effects are not considered at this stage, can they be incorporated in future?
Response:Thanks for your comments and suggestions.Your suggestions are very valuable.The actual engineering situation is always complex. At this stage, the impact of earthquakes is not considered, but in subsequent research, the influence of earthquakes on slope stability can be added as an input. By focusing on the most relevant features in the input data through attention mechanisms, the robustness of the model can be improved.If you have any further questions, please let us know in a timely manner. Thank you again for your suggestions.
Comment 8：What are the recommendations for the design of slope stability infrastructure.
Response:Thanks for your comments and suggestions.The model proposed in this article aims to provide reference suggestions for engineers to judge the stability of slopes, but it cannot provide constructive suggestions. The design of slope stability infrastructure for specific projects requires consideration of many factors, and currently the model considers only six factors, which cannot directly provide suggestions for design. However, in future research, more factors should be added to the model based on the site, so as to improve the model and provide good suggestions for slope stability infrastructure.If you have any further questions, please let us know in a timely manner. Thank you again for your suggestions.
Comment 9：How field engineers will adopt the study results?
Response:Thanks for your comments and suggestions.The model established can provide a reference for field engineers to assess slope stability. By inputting various parameters of the slope into the model, the predicted value of the slope safety factor can be obtained. Combined with traditional slope stability analysis methods (such as the principle of static equilibrium), it can improve the accuracy of field engineers' assessment of slope safety factors.If you have any further questions, please let us know in a timely manner. Thank you again for your suggestions.
Reviewer 3 Report
Comments and Suggestions for AuthorsThe starting point and idea of the article are very good, but there are big problems with the article.
Problems：
Question 1: There is no data or graph for X1-X7 and F1-F12 in lines 108-115 of the introduction.
Problem 2: The illustrations for the whole article are too mushy
Question 3: In chapter 3.3, several attention points are mentioned earlier, but this is used later, which is not clear, why not introduce se directly? Or, if so, why not experiment with other attention mechanisms?
Problem 4: Loss curve Try to use the loss of epoch to check whether the model reaches the fit, the randomness of each iterations is much larger than that of epoch, at least from the graph of the article, the loss can only indicate that the model is converging.
Problem 5: The hyperparameters of the model are not given
Problem 6: The article only evaluates and tests the model, and does not give a visual graph of the actual predictions.
Question 7: Figure 8. The abscissa of the picture for Error bar plots for different prediction models is not stated, and I don't know what it represents.
Author Response
Comment 1：There is no data or graph for X1-X7 and F1-F12 in lines 108-115 of the introduction.
Response:Thanks for your comments and suggestions.X1-X7 and F1-F12 refer to the parameters used in the references listed in Table 1 of the introduction, which are listed in the notes for the convenience of readers to better understand.If you have any further questions, please let us know in a timely manner. Thank you again for your suggestions.
Comment 2：The illustrations for the whole article are too mushy
Response:Thanks for your comments and suggestions.We have revised the questions you raised. If you have any further questions, please let us know in a timely manner. Thank you again for your suggestions.
Comment 3：In chapter 3.3, several attention points are mentioned earlier, but this is used later, which is not clear, why not introduce se directly? Or, if so, why not experiment with other attention mechanisms?
Response:Thanks for your comments and suggestions.We are aware of this error in our communication and have corrected it.If you have any further questions, please let us know in a timely manner. Thank you again for your suggestions.
Comment 4：Loss curve Try to use the loss of epoch to check whether the model reaches the fit, the randomness of each iterations is much larger than that of epoch, at least from the graph of the article, the loss can only indicate that the model is converging.
Response:Thanks for your comments and suggestions.We apologize in advance as we adjust the model's training strategy based on monitoring the loss variation and the performance of the validation set, aiming to avoid overfitting and enhance the model's generalization ability. We have not yet considered using the epoch loss to verify whether the model has achieved fitting. If possible, please provide a few references for us to learn from. We will take your suggestions into consideration in our subsequent research.
Comment 5：The hyperparameters of the model are not given
Response:Thanks for your comments and suggestions.We have already provided it in Table 2.If you have any further questions, please let us know in a timely manner. Thank you again for your suggestions.
Comment 6：The article only evaluates and tests the model, and does not give a visual graph of the actual predictions.
Response:Thanks for your comments and suggestions.We have placed the actual predicted visualization chart in the article as per your request.If you have any further questions, please let us know in a timely manner. Thank you again for your suggestions.
Comment 7：Figure 8. The abscissa of the picture for Error bar plots for different prediction models is not stated, and I don't know what it represents.
Response:Thanks for your comments and suggestions.We have recognized the mistake and explained it in the first paragraph of the 4.2 summary. If you have any further questions, please let us know in a timely manner. Thank you again for your suggestions.
Reviewer 4 Report
Comments and Suggestions for AuthorsDear authors,
Thank you very much for the interesting article submitted to the journal. I read your article with interest. The article is devoted to the current problem of ensuring slope stability based on the use of neural networks. At present, this is an extremely important geotechnical problem due to the increasing number of disasters. I have the following small remarks and comments on the article:
1. There is no comparison with the results of other researchers in the DISCUSSION section.
2. It is recommended to justify the parameters chosen by the authors to predict slope stability. Namely: H, 𝛼, 𝑐, 𝜑, 𝛾 and 𝑟𝑢. For example, this can be done using some regression dependence, where the response function will be slope stability.
3. It is recommended to show the applicability of the CNN-GRU-SE model using a specific example, based on monitoring a real object, and compare it with known analytical methods for calculating slope stability.
Author Response
Comment 1：Thank you very much for the interesting article submitted to the journal. I read your article with interest. The article is devoted to the current problem of ensuring slope stability based on the use of neural networks. At present, this is an extremely important geotechnical problem due to the increasing number of disasters. I have the following small remarks and comments on the article:
Response:Thanks for your comments and suggestions.Due to the use of different datasets by various researchers, the impact on the model varies, leading to differing results. Therefore, this paper does not compare its results with those of other researchers, but merely compares the widely used three regression models based on the dataset used in this paper.If you have any further questions, please let us know in a timely manner. Thank you again for your suggestions.
Comment 2：It is recommended to justify the parameters chosen by the authors to predict slope stability. Namely:H、c、、、、 . For example, this can be done using some regression dependence, where the response function will be slope stability.
Response:Thanks for your comments and suggestions.We have analyzed the variable contribution and importance of each model in the final part of Chapter 4.If you have any further questions, please let us know in a timely manner. Thank you again for your suggestions.
Comment 3：It is recommended to show the applicability of the CNN-GRU-SE model using a specific example, based on monitoring a real object, and compare it with known analytical methods for calculating slope stability.
Response:Thanks for your comments and suggestions.We are very sorry that we cannot compare our results with the known slope stability analysis methods. In traditional slope safety calculation formulas, the parameter of slope gravity is usually used, but in the data we collected, there are few documents providing this parameter, so we cannot compare them. However, in the follow-up work, we will closely link with the engineering project according to your suggestions, achieve close integration with actual detection, and enhance the credibility of the model.
Round 2
Reviewer 3 Report
Comments and Suggestions for AuthorsI am very happy to see that you have adopted some of my suggestions during the initial review, which has significantly improved the quality of the article. Based on your revision this time, I think the content of the article has been better optimized and improved.
