Research on the Uplift Pressure Prediction of Concrete Dams Based on the CNN-GRU Model
Round 1
Reviewer 1 Report
1- Authors should provide more sufficient critical literature review to indicate the drawbacks in existed literature, then, well define the main stream of research direction, how did those previous studies perform? Employ which methodologies? Which problem still requires to be solved? Please clearly indicate (or prove) the proposed approach could solve the drawbacks of previous benchmark comparing approaches which have been widely employed in the literature. Then, please show the reasonableness why those literatures should be referred in the introduction section, finally, state clearly why the proposed approach is suitable to be employed. The introduction section is the most important section to evaluate the contribution to literature, without well introduction writing, it is very difficult to attract the focus of readers. The main contribution of the authors in comparison with previous works should discuss more profoundly and highlights in detail. The authors should explain clearly, what the contribution of this paper is.
2- Many grammar mistakes, improper word usages, and typos can be seen in this manuscript. Language editing service should be considered by authors.
3- It’s suggested that authors should utilize correlation analysis methods to discuss the selection of input factors
4- Please provide a brief description of VMD-SE and explain why this method is introduced for denoising in this manuscript
5- It is demonstrated that CNN-GRU achieves better prediction results than traditional machine learning methods and some deep learning models, but more comparisons with hybrid deep learning models (such as CNN-LSTM) on open-source datasets should be considered in order to prove the effectiveness of this proposed method
6- More experimental details should be introduced: location of the dam, its surrounding environment, process of hyper-parameter selecting, hyper-parameters settings of all the models, as well as hardware and software settings, so that researchers and engineers can follow the proposed method for practical use
Author Response
Please see attached document
Author Response File: 
Author Response.pdf
Reviewer 2 Report
Intelligent monitoring for predictive maintenance is an important area of research in order to limit failures, especially in critical infrastructure like dams as this paper chose to study.
The literature review and comparisons are accurate.
The authors propose a CNN-GRU architecture to modellize uplift pressure in dams. This design allows for extracting both spatial and temporal information from the data, which is especially important on nonlinear and high dimensional problems, which is the case here.
This is a solid work, showing how a CNN-RNN architecture can be implemented successfully to capture more information and features than statistical methods
We would have appreciated a brief description of how the surveillance is currently carried out and to see it compared to the method from this paper.
The denoising data process with VMD-SE needs to be explained concisely
The paper shows strong empirical results by comparing the performance to other benchmark methods. However, the study should be extended to all UPx’s, if possible. With the then acquired UPx data, the authors should then compute qualitative and quantitative results. This would contribute to the soundness of the method and would highlight the actual contribution of the paper.
- In lines 134-135, we disagree with the statement that pooling removes only irrelevant information. While it is useful for reducing dimensionality and removing redundancy, it does carry the risk of losing spatial information.
As for the standard of English in this paper, even though it is often of a high level, many details remain that absolutely need to be corrected. The authors need to proofread the paper on
1. There often too much info pushed into sentences, making them reader unfriendly;
2. Determiners are often missing when a singular is used and / or the singular is used where it should be a plural. We have listed some of these problems here, but there are too many to list them all. The writers need to go back and proofread their text on this problem
Some examples (page 1, line 12à the CNN model & the GRU model page 1 line 19 à for safe dam operations or a safe dam operation; line 26 à dam operations can be affected; line 40 à the prediction accuracy; page 2 à an artificial intelligence theory; page 2 line 48 -àan excellent performance; p2L52 à the SVM prediction accuracy; p2L73àthe LSTM network; p2L75-à a higher prediction accuracy, p5L167 àthe Rule activation function; p6L209 the piezometer and the vw piezometer; p10L275 a significantly better RMSE outcome;p10L292 The RMSE and the MAE)
3. The use of capitals (example page 2, line 48 àSupport
4. Problems with singular subjects & plural verbs (example page 2, line 48 Support Vector Machines = plural & has = singular)
5. Unsound choice of vocabulary (p2l61, due toà because of; p3L126 your model is a problem? ; p4L142 the input historical uplift pressure monitoring data set-à if you really wanted to have such a long list of adjectives, you could maybe say ‘the historical uplift input for the pressure monitoring data set; p12,L342 the forgetting gateà ‘the forget gate’
6. Grammatically unclear sentences (example: p3L116 In the dam online monitoring systems, ???? can accumulate --à there is no subject with the verb; p10L280-281, strange mix of verb tenses (starts & started)
Author Response
Please see attached document
Author Response File: Author Response.pdf
Round 2
Reviewer 1 Report
Authors addressed all the comments. However, there are still some mistakes and problems in this revision.
1. Line 72: Authors may not have a basic understanding of LSTM. Gradient explosion can not be fixed with the introduction of LSTM.
2. Line 95: Why is GRU more competent for insufficient datasets？ Please provide more references.
Author Response
Please see the attachment.
Author Response File: Author Response.pdf
