Production Prediction Model of Tight Gas Well Based on Neural Network Driven by Decline Curve and Data
Round 1
Reviewer 1 Report
Comments and Suggestions for AuthorsOPENING
This review concerns to manuscript MDPI/PROCESSES No. 2931975, by M. Chen et al.
SUMMARY
This paper applies neural networks to predict the production decline in tight gas reservoirs. Arps, SPED, and MFF models are tested and compared against a "pure data-driven" approach.
The proposal is interesting regarding its attempt to compare classical methods with artificial intelligence. There are some positive aspects in showing how the model performs, but little is informed regarding data to validate the results.
It is notorious that some improvement is being reached, but it is not clear how MFF relates to the methodological body explained. From the text: "The neural network model driven by MFF and data is better than the other two neural network models driven by decline curve and data." Is it "model-driven", "data-driven", or what?
Moreover, the conclusion remarks are badly constructed, weak and basically replicate what was already said in the text with almost no further basis of discussion or argumentation.
The paper needs full check of writing and style. I point out several inconsistencies below.
---
. gas Wells -> gas wells
. actual field application -> actual field applications
. data driving -> data-driven
. still some limitations: which ones?
. can not -> cannot
. Colloquial/weird wording:
.. "pay more and more attention"
.. "borrowing the physics"
.. "anti-jamming" [?]
.. "adjustment of the data-driven model through punishment and encouragement in the process of neural network learning"
>> Punishment?? Did you really mean this?! Punishment and encouragement??
.. "J.J. Arps put forward"-> has proposed
.. "introduce machine learning" -> "introduced machine learning"
. Coherence problems and style.
.. In the single statement on Lines 25-27, "oil and gas" appear 4 times.
.. "Representative scholars are: (...)"; "(...) the above two scholars (...) ": writing style is unusual and approaches to archaism.
.. Production prediction and productivity prediction are different concepts. The former measures the volume; the latter the efficiency of recovery. Both concepts are used interchangeably in the text. Although related, they are not exactly the same thing.
.. "The production prediction methods established by scholars in the past have strongly promoted the development of tight gas well production prediction models". Are the authors meaning that prior experience of usage of prediction methods for conventional reservoirs suggests that they can be replicated to tight reservoirs? If so, make this clear.
. Incorrectness [?]
.. Dual medium theory? Would be that "dual porosity model"?
.. Define "BP" network. BackPropagation?
. Prefer indirect citations. Text is too overwhelmed in direct references to authors with no special reason. As a matter of fact, they seems to be regionally biased.
. Many parts of the are redundant.
PRINCIPLE OF DECLINE MODEL
. Text explanation between equations and mathematical description is poorly structured and organized. The text seems a bunch of patches glued to each other.
Example: "first assume x ... Then do the relation curve... real numbers". This is not a pseudocode.
3. PRODUCTION PREDICTION ...
. "Can assist the BP neural network (...)" [?]
. "The decline model is used to optimize the BP neural network and give full play to their respective strengths to achieve the complementarity of the two, so as to improve the accuracy of the model."
>> Bad construction. Who's "their"? "complementarity of the two"?! In what sense? Of whom?
."and replace the original single fitting loss function with the traditional model and the total data loss function, and the total loss function is:"
>> Improve. The total loss is the sum of two losses.
. In Eq. (15), the quadratic expressions of (14) are not maintained, since o_k = f(net_k) as read in (17). Also, the derivation from (13) to (28) does not add too much value in terms of explain how DATA and DC interact. "Decreasing model" or "decline model"?
. Eq. (30). \times symbol is weird.
. Eq (31): explain "rand".
. Fig. 1 adds nothing beyond the words already said.
4. CASE CALCULATION
. It is not explained how the data was processed concerning to make sense with the derivation presented. Which is the volume of information available. Just "horizontal wells in 3 tight gas reservoirs" is obscure.
. Figs. 2 and 3. ylabel must not contain any.
. "Obvious error". Wording... "dramatic/pronounced"? Error is obvious since the most elementary hypothesis, since it exists not only for some parts of the prediction show overshoots. Error is embedded in the loss by construction...
Comments on the Quality of English Language
Included in the first part of comments.
Author Response
Please see the attachment.
Author Response File: 
Author Response.pdf
Reviewer 2 Report
Comments and Suggestions for Authorsl Please check the equation numbers and number them correctly, they are very confusing in your manuscript!
l More detailed justification of the choice of model - why neural networks were chosen over other regression algorithms such as Random Forest, Adaboost, etc., which have been shown to be superior to neural network models in published work.
l What is your motivation for the weights used in the loss function (i.e., Equation 14)? Why are the Data and DC components both 0.5?
l When discussing ML models used in production forecasting, you can take a look at some recent successful applications of physics-informed ML models and add the following:
Li, X., Ma, X., Xiao, F., Xiao, C., Wang, F., & Zhang, S. (2022). A physics-constrained long-term production prediction method for multiple fractured wells using deep learning. Journal of Petroleum Science and Engineering, 217, 110844.
Wang, H., Wang, M., Chen, S., Hui, G., & Pang, Y. (2024). A Novel Governing Equation for Shale Gas Production Prediction via Physics-informed Neural Networks. Expert Systems with Applications, 123387.
Cornelio, J., Mohd Razak, S., Cho, Y., Liu, H. H., Vaidya, R., & Jafarpour, B. (2022). Residual learning to integrate neural network and physics-based models for improved production prediction in unconventional reservoirs. SPE Journal, 27(06), 3328-3350.
Author Response
Please see the attachment.
Author Response File: Author Response.pdf
Round 2
Reviewer 1 Report
Comments and Suggestions for AuthorsI recommend decision to be made by the editor.
Comments on the Quality of English LanguageI recommend observation to be made by the editor.
Author Response
Please see the attachment.
Author Response File: Author Response.pdf
