Pressure Drop Prediction of Crude Oil Pipeline Based on PSO-BP Neural Network
Round 1
Reviewer 1 Report
1. The quality of English is extremely poor. The manuscript has to be revised thoroughly. It should also go through professional proofreading.
2. The references are not cited properly in the body.
3. Figure 2 needs more detailed explanations. It is not clear why the density was neglected. Why didn't the authors consider viscosity for the analysis? Fluid viscosity is known to play a significant role in frictional pressure losses. Pipe diameter is another significant process parameter. It is not clear why it was neglected for the current study. A generalized model cannot be developed neglecting diameter and viscosity. The authors should clarify their selection of process variables.
4. The literature review part is not strong. Multiple kinds of research were conducted earlier in the field of predicting pressure losses in pipelines with ANN. The reasons for selecting the tested ANN algorithms should be rationalized based on the previous studies. Most of the references cited in the current manuscript are peripherally connected to the topic.
5. Numbers and titles of the figures should be corrected. A few figures were not numbered in the current manuscript.
6. Figure 6 and Figure 7 are exactly the same as Figure 9 and Figure 10. Either set of the figures should be removed.
7. Only two algorithms, BP and PSO-BP were used for the comparative analysis. The authors should clarify why they did not use other algorithms like DT and SVM.
8. The scope of the research is very limited. Only one pipeline was used for data collection. The proposed model is not likely to be applicable to other pipelines with different specifications operating under different conditions. The authors need to clarify how useful this model would be for the industry.
9. The friction factor correlations, which were developed based on rigorous theoretical analysis, are capable of predicting pressure losses in horizontal pipelines transporting Newtonian liquids with reasonable accuracies. The necessity of applying complex ANN algorithms in this field needs further clarification. It would be great if the authors could compare the predictions of the current ANN model with the similar predictions obtained using friction factor correlations, such as Blasius law and the Colebrook-White equation.
Author Response
Dear Editor and Reviewers:
Thanks very much for taking your time to review this manuscript. I really appreciate all your comments and suggestions! Please find my itemized responses in below and my revisions/corrections in the re-submitted files.
Thank you and best regards.
Yu Zhang
Response to Reviewer 1 Comments
Point 1: The quality of English is extremely poor. The manuscript has to be revised thoroughly. It should also go through professional proofreading.
Response 1: Thank you for your advice, we have made meticulous modifications to this manuscript, therefore, the readers may understand our work more clearly. The corrected details are listed as highlighted in the manuscript.
Point 2: The references are not cited properly in the body.
Response 2: I am sorry that the references have not been correctly cited in the paper, and I have modified them according to the requirements.
Point 3: Figure 2 needs more detailed explanations. It is not clear why the density was neglected. Why didn't the authors consider viscosity for the analysis? Fluid viscosity is known to play a significant role in frictional pressure losses. Pipe diameter is another significant process parameter. It is not clear why it was neglected for the current study. A generalized model cannot be developed neglecting diameter and viscosity. The authors should clarify their selection of process variables.
Response 3: I am sorry that this part was not clear in the original manuscript. I should have explained that Through the correlation analysis between density and pressure drop, the correlation coefficient between them is-0.1, and correlation is negatively correlated. Therefore, density is not considered when selecting the influencing factors. The pipeline studied in this paper is a fixed pipe, its length and diameter are fixed and will not change, so the diameter is not considered an influencing factor. For a fixed pipeline, the type of oil transported is single, and the viscosity is only affected by temperature. Therefore, the only temperature is considered in the analysis of influencing factors. I have revised the contents of this part. The corrected details are listed as highlighted in the manuscript.
Point 4: The literature review part is not strong. Multiple kinds of research were conducted earlier in the field of predicting pressure losses in pipelines with ANN. The reasons for selecting the tested ANN algorithms should be rationalized based on the previous studies. Most of the references cited in the current manuscript are peripherally connected to the topic.
Response 4: There are some references for previous studies on pressure drop, such as Li Shubin.Research on online simulation and operation optimization technology of long-distance hot oil pipeline. This study focuses on the optimization of the BP neural network, so many articles on neural network applications in related fields are cited. I'll pay attention to this problem when I write my next paper. thank you for your valuable comments.
Point 5: Numbers and titles of the figures should be corrected. A few figures were not numbered in the current manuscript.
Response 5: I'm sorry for this obvious mistake, and the paper has been revised as required.
Point 6: Figure 6 and Figure 7 are exactly the same as Figure 9 and Figure 10. Either set of the figures should be removed.
Response 6: Figures 6 and 7 changed to Figures 8 and 9, and Figures 8 and 9 changed to Figures 11 and 12. I am sorry that I did not think enough about this issue when writing the paper, and I have revised it according to the revision requirements.
Point 7: Only two algorithms, BP and PSO-BP were used for the comparative analysis. The authors should clarify why they did not use other algorithms like DT and SVM.
Response 7: According to previous research found that SVM is generally suitable for machine learning problems with small samples. It is difficult to train large-scale data sets, and there is no general solution for nonlinear problems. It is difficult to find suitable kernel functions. Since the number of data sets in this study is large and the data are nonlinearly correlated, this study mainly focuses on BP neural network. I am sorry that the consideration of this problem is not comprehensive enough. In the next study, I will describe the comparison results of other algorithms together.
Point 8: The scope of the research is very limited. Only one pipeline was used for data collection. The proposed model is not likely to be applicable to other pipelines with different specifications operating under different conditions. The authors need to clarify how useful this model would be for the industry.
Response 8: The test verifies that the PSO-BP pressure drop calculation model has high prediction accuracy for the pressure drop of fixed pipelines without considering the pipe length and diameter. At present, most of the long-distance crude oil pipelines in China are single-pipe transportation. This study predicts the pressure drop of such pipelines. In the next study will continue to learn to apply the model to more cases, thank you for your valuable comments.
Point 9: The friction factor correlations, which were developed based on rigorous theoretical analysis, are capable of predicting pressure losses in horizontal pipelines transporting Newtonian liquids with reasonable accuracies. The necessity of applying complex ANN algorithms in this field needs further clarification. It would be great if the authors could compare the predictions of the current ANN model with the similar predictions obtained using friction factor correlations, such as Blasius law and the Colebrook-White equation.
Response 9: I am sorry that this part was not clear in the original manuscript. I should have explained that for a certain pipeline, the pipe length and diameter are fixed, and the main factors affecting the pressure drop are friction coefficient λ and flow rate υ. The influence factors of λ and υ are analyzed respectively. It is found that the main influence factor of friction coefficient is the Reynolds number, which is related to flow rate, oil viscosity, and pipeline roughness. Among them, oil viscosity is correlated with oil transportation temperature and water content; for the fixed pipe, the pipe roughness changes little, so it is not used as an input parameter. For the flow velocity, due to the fixed length and diameter of the pipeline, the related variable is the volume flow of the pipeline. thank you for your valuable comments. In the next study, we will continue to learn how to compare the predictions of the current ANN model with the similar predictions obtained using friction factor correlations.
Author Response File: 
Author Response.docx
Reviewer 2 Report
There are some excellent agreements generated by the use of this model, however there need to be significant improvements made to the writing of the paper, literature of neural networks and machine learning and explanation of the model.
The use of English is required to be improved throughout. Be careful with units and capitalisation throughout.
In the introduction, referencing must be improved (using numeric referencing rather than just author names). Either in the introduction or the neural network discussion, need many more references to neural network literature.
Any acronyms must be defined the first time they are used.
Discussion of the neural network needs to be expanded and described more clearly. For example, there is no detailed discussion of the relu algorithm etc.
The particle swarm optimisation algorithm is not referenced. Has this been used for similar applications? How much of the algorithm is presented in this paper for the first time and how much is standard from literature?
Any discussion of the reason for not retaining data for validation? Has this been included within the test set or has no validation been conducted?
Results are very impressive in comparison to data. Please include comments on the range of validity for the model?
Author Response
Dear Editor and Reviewers,
Thanks very much for taking your time to review this manuscript. I really appreciate all your comments and suggestions! Please find my itemized responses in below and my revisions/corrections in the re-submitted files.
Thank you and best regards.
Yu Zhang
Response to Reviewer 1 Comments
Point 1: There are some excellent agreements generated by the use of this model, however, there need to be significant improvements made to the writing of the paper, literature of neural networks and machine learning, and explanation of the model.
Response 1: Thank you for your advice, We have made detailed revisions to the manuscript, including Abstract, Introduction, and improvement of the neural network model. In future writing, we will pay attention to these issues and make the manuscript more rigorous. The corrected details are listed as highlighted in the manuscript.
Point 2: The use of English is required to be improved throughout. Be careful with units and capitalization throughout.
Response 2: Thank you for your advice, we have made meticulous modifications to this manuscript, therefore, the readers may understand our work more clearly. The corrected details are listed as highlighted in the manuscript.
Point 3: In the introduction, referencing must be improved (using numeric referencing rather than just author names). Either in the introduction or the neural network discussion, need many more references to neural network literature.
Response 3: I am sorry that the references have not been correctly cited in the paper, and I have modified them according to the requirements.
Point 4: Any acronyms must be defined the first time they are used.
Response 4: I'm sorry for this obvious mistake, and the paper has been revised as required.
Point 5: Discussion of the neural network needs to be expanded and described more clearly. For example, there is no detailed discussion of the relu algorithm etc.
Response 5: I am sorry that this part was not clear in the original manuscript. The activation function is Relu, The convergence rate is faster than that of Sigmoid and tanh functions, and there is no gradient saturation, which alleviates the occurrence of overfitting problem. In the next study, I will describe in detail the influence of neural network selection of different optimizers and different activation functions on the prediction results. thank you for your valuable comments.
Point 6: The particle swarm optimization algorithm is not referenced. Has this been used for similar applications? How much of the algorithm is presented in this paper for the first time and how much is standard from literature?
Response 6: I am sorry that this part was not clear in the original manuscript. I should have explained that since the global search ability of the BP neural network is low and the local search ability is strong, on the contrary, the global search ability of the particle swarm algorithm is high and the local search ability is low, so the PSO-BP neural network mechanisms can be constructed. In the manuscript, we describe the calculation steps of the PSO algorithm and optimize the weights and thresholds of the BP neural network by the PSO algorithm. The position of each particle in the particle swarm represents the set of weights in the current iteration of BP neural network. The dimension of each particle is determined by the number of weights and thresholds that play a connecting role in the network. The output error of the neural network with a given training sample set is used as the fitness function of the neural network training problem. The fitness value represents the error of the neural network. The smaller the error, the better the performance of the particle in the search. The particle moves in the weight space to search so that the error of the network output layer is minimized. Changing the speed of the particle is to update the network weights to reduce the mean square error. thank you for your valuable comments.
Point 7: Any discussion of the reason for not retaining data for validation? Has this been included within the test set or has no validation been conducted?
Response 7: I am sorry that this part was not clear in the original manuscript. The manuscript selects the data of actual production operation in the past three years. In the model training, 70 % of the data set in the previous two years is selected as the training set and 30 % as the test set. On this basis, the third year data is used as the test set to verify. Prediction of pressure drop based on 3-year data. thank you for your valuable comments.
Point 8: Results are very impressive in comparison to data. Please include comments on the range of validity for the model?
Response 8: The test verifies that the PSO-BP pressure drop calculation model has high prediction accuracy for the pressure drop of fixed pipelines without considering the pipe length and diameter. At present, most of the long-distance crude oil pipelines in China are single-pipe transportation. This study predicts the pressure drop of such pipelines. In the next study will continue to learn to apply the model to more cases, thank you for your valuable comments.
Author Response File: Author Response.docx
Reviewer 3 Report
Small changes could be reconsidered.
1. Introduction
P.32 (BP needs to be explained. What does BP stand for?)
P. 35 (neural network needs to be explained)
P.50 (throughput-Explain this word what the autors meant )
P. 81 (the Darcy formula-should the authors mention this equation)
p. 91-95 (If possible, could the authors mention units?)
P. 104 (What does PSO-BP stands for )
P- 202 (mass flow (t/h)-rewrite to (tonne/hour))
Author Response
Dear Editor and Reviewers,
Thanks very much for taking your time to review this manuscript. I really appreciate all your comments and suggestions! Please find my itemized responses in the re-submitted files.
Thank you and best regards.
Yu Zhang
Author Response File: Author Response.docx
Round 2
Reviewer 1 Report
Thank you for revising the manuscript. Still, the current version of the manuscript does not answer the following questions:
1. Why is a databased MLA required for predicting pressure losses in the pipeline while a well-established physics-based methodology is already available?
2. How will a model that disregards essential process parameters, such as pipe diameter, fluid viscosity, and pipe roughness be useful?
Please incorporate the answers to these questions in your manuscript.
Author Response
Dear Editor and Reviewers,
Thanks very much for taking your time to review this manuscript. I really appreciate all your comments and suggestions! Please find my itemized responses in below and my revisions/corrections in the re-submitted files.
Thank you and best regards.
Yu Zhang
Author Response File: Author Response.docx
Reviewer 2 Report
Happy with the changes that have been made. Could still use professional proof-read for corrections to grammar.
Author Response
Dear Editor and Reviewers,
Thanks very much for taking your time to review this manuscript. I really appreciate all your comments and suggestions! Please find my itemized responses in below and my revisions/corrections in the re-submitted files.
Thank you and best regards.
Yu Zhang
Author Response File: Author Response.docx
Round 3
Reviewer 1 Report
Thank you for improving the manuscript. It would be great if you could compare the ML-based predictions with the results obtained using traditional models like Colebrook - Whyte or Blasius correlation.
Author Response
Dear Editor and Reviewers,
Thanks very much for taking your time to review this manuscript. I really appreciate all your comments and suggestions! Please find my itemized responses in below.
Thank you and best regards.
Yu Zhang
Author Response File: Author Response.docx
