Study on Soil Parameter Evolution during Ultra-Large Caisson Sinking Based on Artificial Neural Network Back Analysis

Round 1

Reviewer 1 Report

In general, the direction of research chosen by the authors is interesting and relevant. However, the article still needs revision and improvement.

Some points that have to be improved.

How many records did the dataset for the ANN model contain?

How was the dataset divided into training, testing and validation sets?

Did the authors standardize the data, since some parameters in Table 2 differ by 6 orders of magnitude?

Why are 20 output neurons specified in the network structure? What variables did the authors consider as the outputs?

Why was only correlation used to estimate the model, and not the typical "error + correlation" approach?

I did not understand the results described in 5.2 Prediction Results Analysis.

In my opinion, classical forecasting using the ANN model has the following procedure: a previously trained, tested and validated ANN model is saved. After that, the model is loaded and new input data (x1 - x20) are fed into it. The model processes new data and outputs a predictive value.

Please add the discussion section. The "Discussion" section should contain a comparison of the results of this article and similar studies by other authors. The authors need to prove the superiority and scientific novelty of this study.

Author Response

Please see the attachment.

Author Response File: Author Response.pdf

Reviewer 2 Report

This paper applies artificial neural networks to the back analysis of soil parameters during the sinking process of open caisson, and has achieved good prediction results. This study has made a useful attempt for parameter inversion of caisson engineering, and obtained some innovative results. However, the following problems still exist.

(1) This paper simulates the dynamic sinking process under various static conditions. Suggest a detailed analysis of the differences between the two modeling methods.

(2) In the analysis of the results, this paper pointed out that the trend of changes in various soil parameters has obvious patterns, but did not describe them in detail.

(3) There are formatting issues with the text, such as some secondary titles not having uppercase letters.

(4) Suggest adjusting the image size to make the font display clear and beautiful.

(5) The spelling of coarse sand in Figure 8 is wrong, please correct it

(6) In Fig. 15, it is suggested to enlarge the size of points and lines in the small figures for easy reading.

(7) In the Introduction part, some related references are suggested added, e.g. https://doi.org/10.1016/j.soildyn.2022.107456& DOI: 10.1061/AJRUA6.0001251& Machine learning-based classification of rock discontinuity trace: SMOTE oversampling integrated with GBT ensemble learning.

(8) In Table 1, the heights of each soil layers are missing, please includes this information.

(9) For Fig.9, more information should be provided in this figure to show the details of the numerical model, e.g. the settings of the numerical boundary.

Some minor checking.

Author Response

Please see the attachment.

Author Response File: Author Response.pdf

Round 2

Reviewer 1 Report

I recommend publishing the article in present form