Forecasting Model of Silicon Content in Molten Iron Using Wavelet Decomposition and Artificial Neural Networks
, Klaus Fabian Côco 1, Flávio S. Vitorino Gomes 2 and José L. Félix Salles 1,*Round 1
Reviewer 1 Report
1. Some important papers below should be cited.
1) Prediction model of permeability index for blast furnace based on the improved multi-layer extreme learning machine and wavelet transform By: Su, Xiaoli; Zhang, Sen; Yin, Yixin; et al.
JOURNAL OF THE FRANKLIN INSTITUTE-ENGINEERING AND APPLIED MATHEMATICS Volume: 355 Issue: 4 Special Issue: SI Pages: 1663-1691 Published: MAR 2018 2) Process monitoring of iron-making process in a blast furnace with PCA-based methods By: Zhou, Bo; Ye, Hao; Zhang, Haifeng; et al. CONTROL ENGINEERING PRACTICE Volume: 47 Pages: 1-14 Published: FEB 2016 3. Evaluation and Prediction of Blast Furnace Status Based on Big Data Platform of Ironmaking and Data Mining Hongyang Li, Xiangping Bu, Xiaojie Liu, Xin Li, Hongwei Li, Fulong Liu, Qing Lyu ISIJ International2021 Volume 61 Issue 1 108-118 2. The input variables used in Table 1 are not independent. More precise selection of features is required. 3. What is the merit to publish this paper in Metals? I do not think that this paper convey scientific worthy information to the readers. Why does the suggested models give different prediction values? When the input data were measured? At the same time (6 hours before)? It does not look like. This paper needs comprehensive revision to be published.
Author Response
Please see the attachment
Author Response File: 
Author Response.pdf
Reviewer 2 Report
The authors proposes a new algorithm to predict the silicon content time series up to 8 hours ahead. The following issues are unclear to me.
- The silicon content last time or the nearest time should be considered.
- The prediction results should be quantitatively.
- 2160 sample may be not enough for the training.
Author Response
Please see the attachment
Author Response File: Author Response.pdf
Reviewer 3 Report
Dear authors,
I have some suggestions and questions. The Conclusion must be improved.
A high number of literary sources in the introductory chapter, chaotic arrangement of used sources, adjustment of the distribution of sources in the text.
Text editing - sources are in the order 18 - 19 - 18 - 18 in a short sequence in the paragraph.
341 We used the Toolboxes provided by [34] through http://www.iau.dtu.dk/research/control
342 /nnsysid.html to implement the neural networks and perform the pruning algorithm. -website directly in the text
How the time intensity of the model changes with decreasing value of percentage deviation?
How accurate a model can be if the operator defines that he wants the lowest possible error?
How long did it take for the NAR network to "learn" to recalculate the input parameters that are inserted into the blast furnace by the operator according to the current needs of the blast furnace?
How far does your model exceed the accuracy of the human factor of an "experienced" operator?
Given the production capacity and the volume of the blast furnace, how much costs will be saved by applying such a model?
You are writing about reducing emissions, does it have any perspective in the long run, given the current effort to remove blast furnaces from the steelmaking process? What percentage of reduction in CO2 emissions has occurred using this model?
What are the financial costs of implementing such a model in operating conditions compared to the amount of money saved?
What is the time required to apply such a model to a new blast furnace with different parameters? Have you tried such a model on several Blast Furnaces with different production capacity and volume? If so, what percentage of error occurred between the models used for the 1st and 2nd blast furnace?
Author Response
Please see the attachment
Author Response File: Author Response.pdf
Round 2
Reviewer 1 Report
1. Figure 1 is adapted from the book "Modern Blast Furnace Ironmaking: an Introduction" Did you receive a permission for this figure?
2. You have mentioned that the prediction was carried out 3 to 8 hours ahead. Then could you summarize the effect of the prediction time? Actually, the previous papers reports the prediction values for 3 hours ahead, because the chemical analysis results are supplied in 3 hours. If you mention that the prediction for 8 hours before is required, it means the analysis system is not so good.
3. What is the dominant factors affecting the Si content? The purpose of the prediction is to maintain the Si level in hot metal. This paper does not deliver enough information for this aspect. Furthermore, the predicted results do not show similar tendency. It means the controlled parameters do not affect the Si level as expected.
Author Response
"Please see the attachment."
Author Response File: Author Response.pdf
Reviewer 3 Report
Thank You for your response. The manuscript has now desired quality to be published.
Please, change in line 160 SiO2 to SiO2
Author Response
"Please see the attachment."
Author Response File: Author Response.pdf
