Real-Time Dynamic Carbon Content Prediction Model for Second Blowing Stage in BOF Based on CBR and LSTM
Round 1
Reviewer 1 Report
The research is a very interesting and new approach to solving the problem of controlling the converter process. The formally proposed scheme is very innovative and seems to be confirmed in practice with the results obtained.
However, the presented results lead to the discussion of the following doubt.
Fig. 11 shows the change in the error value as a function of the number of epochs. It should be noted that the error sets its value after several epochs and further training does not change it. It seems that the continuation of the training with subsequent similar heats does not bring anything more. Therefore, the result presented in Fig. 14 is incomprehensible. Why does increasing the m parameter value to 4 improve the quality of the model? I consider the clarification of this doubt to be of key importance for the presented research
Author Response
Reviewer 1:
（1）Fig. 11 shows the change in the error value as a function of the number of epochs. It should be noted that the error sets its value after several epochs and further training does not change it. It seems that the continuation of the training with subsequent similar heats does not bring anything more. Therefore, the result presented in Fig. 14 is incomprehensible. Why does increasing the m parameter value to 4 improve the quality of the model? I consider the clarification of this doubt to be of key importance for the presented research.
Response:
（1）The data used in LSTM model are from similar cases in history. Although the similarity between new case and similar cases is very high, there are still some differences in the process parameters (such as TSC[c] and the length of the time-series). The change law of carbon content between cases is also different. The model training error is no longer changed, which only means that the model can learn the change law of similar cases effectively, but when applied to new cases, it does not necessarily mean to get the best results. Take the data in the paper as an example: the TSC[c] of Heat 1 is 0.241% and the TSC[C] of similar cases (Heat 2, 3, 4, 5) are 0.207%, 0.226%, 0.299%, 0.303% respectively. The TSC[c] of Heat 1 is higher than Heat 2. When only using the data of Heat 2 to train the LSTM model. The change rule of carbon content between 0.241% and 0.207% is not trained. So the quality of model (m=1) is not better than the model (m=4). It should also be noted that the similarity of the retrieved similar cases decreases gradually. The parameter difference between new cases and similar cases also increases, resulting in the increase of the difference in the change law of carbon content. Therefore, increase of cases when m > 4 will lead to the decline of the prediction results of the model.
The above reasons lead to the results shown in Fig. 14。
Author Response File: 
Author Response.pdf
Reviewer 2 Report
- Dimensions and capacity of converter must be added to work (section 2).
- Difference between Authors model and CIM, CM and EM should be described in the work e.g. introduction chapter.
- Table 2 show argon flow. Therefore explanation on argon using during oxidation of hot metal is obligatory.
- What type of values Authors used in the table 3: mean, max, min or other.
- Figure 10 present very low quality, please improve.
- What means actual value in the figure 12. What time period represents time in the figure 12. In a standard way carbon deoxidation time period amounts 20 min.
- What means value in the figure 13.
- How were calculated % difference 16, 12 and 4 for compared models.
Author Response
Reviewer 2
（1）Dimensions and capacity of converter must be added to work (section 2).
（2）Difference between Authors model and CIM, CM and EM should be described in the work e.g. introduction chapter.
（3）Table 2 show argon flow. Therefore explanation on argon using during oxidation of hot metal is obligatory.
（4）What type of values Authors used in the table 3: mean, max, min or other.
（5）Figure 10 present very low quality, please improve.
（6）What means actual value in the figure 12. What time period represents time in the figure 12. In a standard way carbon deoxidation time period amounts 20 min.
（7）What means value in the figure 13.
（8）How were calculated % difference 16, 12 and 4 for compared models.
Response:
(1) Dimensions and capacity of converter must be added to work (section 2).
Response: We add the capacity of converter in the section 2.
Figure 1 shows the smelting process of converter steelmaking, the capacity of converter is 300t.
(2) Difference between Authors model and CIM, CM and EM should be described in the work e.g. introduction chapter.
Response: we add the description of CIM, CM and EM in section 3.3
3.3 Prediction Model Bsaed on off-gas analysis
（1）Carbon integral model
Based on the principle of mass balance, the carbon integral model first calculates the initial carbon amount of the molten steel according to the composition of steelmaking raw materials and then subtracts the amount of carbon overflowing from the offgas in the form of CO and CO2, the remaining part is the amount of carbon in the molten steel in the molten pool
According to the flow of off-gas and percentage content of CO and CO2 in the off-gas, the decarburization rate in the molten pool can be calculated by using the carbon balance in the converter smelting process:
|
(11) |
Where:
is decarburization rate of molten pool
is the flow of off-gas;
、is percentage content of CO and CO2 in the off-gas;
Then the instantaneous carbon content in molten pool can be expressed as:
|
(12) |
Where:
is the carbon content in molten pool at time t;
is carbon content in molten pool at initial conditions;
is weight of molten steel in molten pool;
(2) Exponential decay model
Exponential model is the most widely used decarburization characteristic model in the later stage of converter. It assumes that there is an exponential attenuation relationship between decarburization rate and carbon content in molten pool in the later stage of coverter blowing.
|
(13) |
Equation (14) is obtained from equation (13):
|
(14) |
Where: k1, k2 is the undetermined coefficient of the model.
(3) Cubic model
In addition to the exponential model, the curve used to describe the decarburization characteristics of the converter in the later stage also has a cubic equation:
|
(15) |
Where: b0,b1,b2,b3 is the undetermined coefficient of the model.
(3) Table 2 show argon flow. Therefore explanation on argon using during oxidation of hot metal is obligatory.
Response: we add the explanation on argonblowing.
The role of argon bowing is to stir the molten pool to achieve a uniform chemical compo-sition and temperature of the molten steel and accelerate the chemical reaction.
(4) What type of values Authors used in the table 3: mean, max, min or other.
Response: the type values used in table3 is actual value of process parameters, such X1， 4.13317 is the carbon content of hot metal of Heat 1.
(5) Figure 10 present very low quality, please improve.
Response：we improve the quality of Figure 10
|
（a） |
|
(b) |
|
(c) |
|
(d) |
Figure 10. Process parameters for the late blowing stage of similar cases (a) Heat 2; (b) Heat 3; (c) Heat 4; (5) Heat 5
(6) What means actual value in the figure 12. What time period represents time in the figure 12. In a standard way carbon deoxidation time period amounts 20 min.
Response: actual value means the actual carbon content of Heat1, This aim of this paper is to calculate the carbon content in the period between TSC and TSO. The time period is in the final stage of decarburization in Figure 2. This time period amounts 1~2min.
Figure 2. Variation in the decarburisation rate during the blowing process of converter steelmaking
(7) What means value in the figure 13.
Response: the prediction and actual value means the prediction and actual carbon content values at the end of blowing.
(8) How were calculated % difference 16, 12 and 4 for compared models.
Response: It’s my fault to calculate wrong. The right difference is 22%、14% and 6%.
Compared with the other three models, the proposed model improved the prediction accuracy in the range of [−0.02, 0.02] by 22%, 14% and 6%, respectively
Author Response File: Author Response.pdf
