Research on Automatic Classification of Coal Mine Microseismic Events Based on Data Enhancement and FCN-LSTM Network
, Dexing Li 1,2, Gan Qin 2 and Hao Li 4,*Round 1
Reviewer 1 Report
Dear Guojun Shang, Li Li, Liping Zhang, Xiaofei Liu, Dexing Li, Gan Qin and Hao Li, authors of the manuscript “Research on Automatic Classification of Coal Mine Microseismic Events Based on Data Enhancement and FCN-LSTM Network“, your manuscript is initially written fine and fluently, although some variations in words (eg. shows should be replaced with yields, demonstrates, indicates, illustrates etc.) should be considered.
While known abbreviations may appear in the title, in the abstract they need to be mentioned fully prior abbreviations (FCN and LSTM).
In the introduction are fines studies of the matter cited, but several classic or modern of importance are missing, such as:
LIANG, B. Y., Shen, C., LENG, C. B., GUO, B. P., YANG, Y. C., & ZHENG, B. T. (2015). Development of microseismic monitoring for hydro-fracturing. Progress in Geophysics, 30(1), 401-410.
Van Der Baan, M., Eaton, D., & Dusseault, M. (2013, May). Microseismic monitoring developments in hydraulic fracture stimulation. In Isrm international conference for effective and sustainable hydraulic fracturing (pp. ISRM-ICHF). ISRM.
Chen, Y., Saad, O. M., Savvaidis, A., Chen, Y., & Fomel, S. (2022). 3D microseismic monitoring using machine learning. Journal of Geophysical Research: Solid Earth, 127(3), e2021JB023842.
In the abstract and in the purpose (last paragraph of introduction) as well as in the final part of the conclusions (twice) of the manuscript it has been mentioned that the obtained data may serve to disaster prevention and early warning, however, there is no indication how this shall be. The entire manuscript lacks of any further specification of this issue. Mentioning that the data may serve as a potential early warning but not indicating how, may distract readers about the possible importance in this matter. Therefore, either this part shall be eliminated, or a better explanation shall be provided.
It has been mentioned, that the data were provided from a coal mine, while no site has been mentioned at all (neither the time lapse of those data).
There are three types of waveforms presented, and in the proper manuscript, there are the weaknesses mentioned (unbalanced data amounts etc.), which is an excellent approach, but if so, why taking in consideration these waveforms as example for the further development of the study? In this respect, it would be recommendable to include the data of the table 1 into the text (eliminating the table) and also the listing of line 106-112, involving into a fluid text rather to a listing.
Later appears table1 again, but is table 2 (lines 181-182) and should be also be included in the text (recommendation to eliminate this table). The later table 2 (in reality 3) and table 3 (in reality table 4) should be also eliminated and the corresponding data should be included in the text of the manuscript. Table 4 (in reality table 5) is already mentioned in the text (initial part of page 9) and should be also eliminated.
Finally, the presentation of the enhanced form of the FCN-LSTM model is adequately presented and the conclusions (less the early warning issue) are correspondingly explained.
Therefore, the manuscript should be accepted after the minor recommendations previously expressed.
Minor English improvements should be realized, next to other minor issues as previously mentioned
Author Response
- Dear Guojun Shang, Li Li, Liping Zhang, Xiaofei Liu, Dexing Li, Gan Qin and Hao Li, authors of the manuscript “Research on Automatic Classification of Coal Mine Microseismic Events Based on Data Enhancement and FCN-LSTM Network“, your manuscript is initially written fine and fluently, although some variations in words (eg. shows should be replaced with yields, demonstrates, indicates, illustrates etc.) should be considered.
Reply: Thank you for the concern. We have made the corresponding replacement, see the article for details.
- While known abbreviations may appear in the title, in the abstract they need to be mentioned fully prior abbreviations (FCN and LSTM).
Reply: Thank you for the concern. We have completed the modifications in the abstract.
- In the introduction are fines studies of the matter cited, but several classic or modern of importance are missing, such as:
LIANG, B. Y., Shen, C., LENG, C. B., GUO, B. P., YANG, Y. C., & ZHENG, B. T. (2015). Development of microseismic monitoring for hydro-fracturing. Progress in Geophysics, 30(1), 401-410.
Van Der Baan, M., Eaton, D., & Dusseault, M. (2013, May). Microseismic monitoring developments in hydraulic fracture stimulation. In Isrm international conference for effective and sustainable hydraulic fracturing (pp. ISRM-ICHF). ISRM.
Chen, Y., Saad, O. M., Savvaidis, A., Chen, Y., & Fomel, S. (2022). 3D microseismic monitoring using machine learning. Journal of Geophysical Research: Solid Earth, 127(3), e2021JB023842.
Reply: Thank you for the concern. We have been adding references.
- In the abstract and in the purpose (last paragraph of introduction) as well as in the final part of the conclusions (twice) of the manuscript it has been mentioned that the obtained data may serve to disaster prevention and early warning, however, there is no indication how this shall be. The entire manuscript lacks of any further specification of this issue. Mentioning that the data may serve as a potential early warning but not indicating how, may distract readers about the possible importance in this matter. Therefore, either this part shall be eliminated, or a better explanation shall be provided.
Reply: Thank you for the concern. We have eliminated this part.
- It has been mentioned, that the data were provided from a coal mine, while no site has been mentioned at all (neither the time lapse of those data).
Reply: Thank you for the concern. We have added the corresponding expression in the papaer. See lines 90 to letter 92 for details.
- There are three types of waveforms presented, and in the proper manuscript, there are the weaknesses mentioned (unbalanced data amounts etc.), which is an excellent approach, but if so, why taking in consideration these waveforms as example for the further development of the study? In this respect, it would be recommendable to include the data of the table 1 into the text (eliminating the table) and also the listing of line 106-112, involving into a fluid text rather to a listing.
Reply: Thank you for the concern. It does not matter to choose any waveform as an example, which can show the effectiveness of the method in this paper. We have completed the reformatting of the table as suggested.
- Later appears table1 again, but is table 2 (lines 181-182) and should be also be included in the text (recommendation to eliminate this table). The later table 2 (in reality 3) and table 3 (in reality table 4) should be also eliminated and the corresponding data should be included in the text of the manuscript. Table 4 (in reality table 5) is already mentioned in the text (initial part of page 9) and should be also eliminated.
Reply: Thank you for the concern. We have completed the reformatting of the table as suggested.
Author Response File: 
Author Response.pdf
Reviewer 2 Report
This study presents the intelligent classification of microseismic events is of great significance for coal mine disaster detection and early warning. Through training, the accuracy of the model validation set is 93.41%, and the accuracy of the validation set is 91.18%. The trained FCN-LSTM model is highly available. However, I didn't get the idea, what do you classify: what are these 3 classes? Is it event as shaking? failure? acoustic emission? Which kind of disasters are they? Who will cut these signals from the whole registered seismogram?
Comments:
Figures titles upon the guidelines template: Figure 1. Here you see. Table 1. The number of samples of each kind in the training set after upsampling.
Line 74. What do you mean as Data Sparsification? what kind of methods. Why with the capital letter?
Line 118 0.2 ms (space between value and unit). Line 171 2 dB, etc.
Fig.6. Why (c) number of points is bigger than in (b)?
Line 201. warping rate is 2 and 0.5 units? counts/by slice? or it is ratio upon Table 2?
Please, explain the abbreviation FCN-LSTM
Fig. 8 convolution 2D. Relu? Softmax? Which Classification? Flatten? Batch Normalization equation? This diagram is non-understandable.
References:
[2] Tectonophysics 1998,
[27] Murphey, Y.; Guo, H.; Feldkamp, L.
Author Response
Reply: Thank you for the concern. The three classes of events are groundwater movement, continuous rupture of coal seam and tectonic activation induced microseismic events. It’s floor water inrush warning in the working face and they are segments that cut to a fixed degree according to the first break position by workers.
Comments:
- Figures titles upon the guidelines template: Figure 1. Here you see. Table 1.The number of samples of each kind in the training set after upsampling.
Reply: Thank you for the concern. The template of figures titles has been modified as suggested, please see the article for details.
- Line 74. What do you mean as Data Sparsification? what kind of methods. Why with the capital letter?
Reply: Thank you for the concern. The sparsification initial capitalization is a misspelling and has been corrected in the article.
- Line 118 0.2 ms (space between value and unit). Line 171 2 dB, etc.
Reply: Thank you for the concern. The errors have been modified as suggested, please see the article for details.
- 6. Why (c) number of points is bigger than in (b)?
Reply: Thank you for the concern. The window slicing method can intercept segments of different lengths and the slice lengths of (b) and (c) are 75% and 87.5% of the original signal, respectively. So, (c) number of points is bigger than in (b).
- Line 201. warping rate is 2 and 0.5 units? counts/by slice? or it is ratio upon Table 2?
Reply: Thank you for the concern. The warping rate is just a number without units.
- Please, explain the abbreviation FCN-LSTM
Reply: Thank you for the concern. The FCN-LSTM is the abbreviation for fully convolutional network and Long short-term memory network which represents a neural network with two network solutions in series.
- 8 convolution 2D. Relu? Softmax? Which Classification? Flatten? Batch Normalization equation? This diagram is non-understandable.
Reply: Thank you for the concern.
‘convolution 2D’ means create a 2D convolutional layer and converted the 1-D signal into a 2-D signal.
‘RelU’ which stands for Rectified Linear Unit, is a commonly used activation function in artificial neural networks.
The Softmax function, or normalized exponential function, is a generalization of the logistic function. It "compreszes" a k-dimensional vector z of any real number into another k-dimensional real vector σ(z) such that each element is in the range (0,1) and the sum of all elements is 1.
Classification is the classification function that determines the label of the input data.
Flatten means flattening the layer, that is, compressing the data of (height,width,channel) into a one-dimensional array of height × width × channel.
Batch Normalization forces the distribution of the input values in each neural network layer back to a standard normal distribution with zero mean and one variance.
- References:
[2] Tectonophysics 1998,
[27] Murphey, Y.; Guo, H.; Feldkamp, L.
Reply: Thank you for the concern. We have modified the two errors.
Author Response File: Author Response.pdf
Reviewer 3 Report
The paper illustrate an application of LSTM method for signal classification of microseismic data in coal mine production.
The manuscript is of general interest to the readers of the journal. It is well written and organized, however, in the reviewer's opinion, the paper will be suitable for publication after undergoing the suggested revisions:
The type of classification used was not fully described in the manuscript.
In order to make a judgment about the FCN-LSTM model, the type of classification used should be explained in detail. In particular, it should be specified whether a classification adopted in proven standards or documents was used, or whether classes or guidelines specific to the coal mine analyzed were used. In that case, the possibility of extending the classification to other coal mines should be illustrated, also explaining possible limitations.
Author Response
Thank you for the concern. The three classes of events are groundwater movement, continuous rupture of coal seam and tectonic activation induced microseismic events. It’s floor water inrush warning in the working face. The first kind of microseismic event is related to the underground water movement. The number and proportion of water inrush disasters have markedly increased during its occurrence. Attention should be provided to the number and proportion of these kinds of events during microseismic monitoring. If the classification is to be applied to other coal mines, it is necessary to improve the types of samples in the training set as much as possible according to the actual situation, and include as many sample types as possible to improve the classification accuracy.
Author Response File: Author Response.pdf
