Method for predicting steel output from an electric furnace

Abstract

This invention provides a method for predicting the amount of steel produced in an electric furnace by melting multiple types of scrap, thereby accurately predicting the amount of steel produced using a regression model, and ensuring that the desired number of ingots are secured without causing excessive excess molten metal or insufficient steel production. [Solution] For each steel type to be melted and manufactured, representative values ​​of the blending weights of the scrap used and the electric furnace factors are used as explanatory variables in the regression model, and the amount of steel produced is used as the objective variable of the regression model. The regression model is trained by providing it with training data from the most recent predetermined period, and a predicted value of the amount of steel produced during melting and manufacturing is obtained using the trained regression model.

Description

Technical Field

[0001] The present invention relates to a method for predicting the tapping weight of an electric furnace, and particularly to a method for predicting the tapping weight using a regression model.

Background Art

[0002] Steelmaking using an electric furnace is carried out by charging a plurality of types of scrap into the furnace and melting them. Conventionally, the predicted tapping weight is calculated by adding up the tapping specified weights of each scrap determined in advance for each steel type. As a result, there is a large error between the actual tapping weight. For this reason, there is a problem that the amount of surplus molten steel becomes excessive after obtaining the desired number of ingot materials, or conversely, the amount of surplus molten steel to be discarded becomes excessive after obtaining the desired number of ingot materials because the amount of molten steel poured into the mold is insufficient and the desired number cannot be obtained.

[0003] In Patent Document 1, there is shown a blending plan support method for estimating the component content in molten steel obtained by melting a plurality of types of scrap at an arbitrary blending ratio using a regression model.

Prior Art Documents

Patent Documents

[0004]

Patent Document 1

Summary of the Invention

Problems to be Solved by the Invention

[0005] Therefore, an object of the present invention is to provide a method for predicting the tapping weight of an electric furnace that can secure the desired number of ingot materials without causing excessive surplus molten steel or insufficient tapping weight by appropriately predicting the tapping weight when melting a plurality of types of scrap using a regression model.

Means for Solving the Problems

[0006] To achieve the above objective, in this invention, for each type of steel to be melted and manufactured, representative values ​​of the blending weight of the scrap used and each data of the electric furnace factors are used as explanatory variables in a regression model, and the amount of steel produced is used as the objective variable of the regression model. The regression model is trained by providing the most recent predetermined training data, and a predicted value of the amount of steel produced during melting and manufacturing is obtained using the trained regression model.

[0007] Furthermore, it is preferable that the electric furnace factor is the total amount of O2 used, and at least one of the amounts of SiC used, carbon injection used, CaO used, and breeze used during one operation. It is also preferable to select the median value as the representative value of the electric furnace factor.

[0008] Furthermore, it is preferable to use the total weight obtained by adding up the weights of each ingredient as an explanatory variable in the regression model. [Effects of the Invention]

[0009] According to the electric furnace steel tapping volume prediction method of the present invention, when steel is obtained by melting multiple types of scrap, the amount of steel tapped can be accurately predicted by using the weight of each type of scrap in the mixture and representative values ​​of each data point for the electric furnace factors as explanatory variables, and further, if necessary, using the total weight obtained by adding the above weights as an additional explanatory variable. As a result, the desired number of ingots can be obtained without causing excessive excess molten metal or insufficient steel tapping volume. [Brief explanation of the drawing]

[0010] [Figure 1] This figure shows an example of operational data for the steel types that were melted and manufactured. [Figure 2] This graph shows the relationship between the most recent number of months since training data acquisition and the RMSE of the multiple regression model. [Modes for carrying out the invention]

[0011] The embodiments described below are merely examples, and various design improvements made by those skilled in the art without departing from the spirit of the present invention are also included within the scope of the present invention.

[0012] In this embodiment, as shown in Figure 1, a predetermined number of operational data (1,973 in one example) were obtained for the steel grades that were melted and manufactured. The contents of each operational data include the blending weight (kg / ch: where ch represents one operation) of 13 types of scrap a to m, plus 5 types of electric furnace factors, and the planned steel output amount ys (kg / ch) calculated by adding up the predetermined steel output amounts for each scrap grade as defined in the conventional method, as well as the actual steel output amount y (pre-LF molten metal amount) (kg / ch).

[0013] In this embodiment, the five electric furnace factors are the total amount of O2 used (Nm3), the amount of SiC used (kg / ch), the amount of carbon injection (AF_C-INJ) used (kg / ch), the amount of CaO used (kg / ch), and the amount of breeze used (kg / ch). Note that instead of using all five electric furnace factors as shown in this embodiment, at least one of these factors may be used.

[0014] Then, a predetermined number of the most recent early period (for example, the first two months within the last four months) from the above operational data was used as training data for the multiple regression model shown in equation (1) below. The respective blend weights of scrap a to m were substituted into the explanatory variables x1 to x13 of the multiple regression model, as well as the usage amounts of each electric furnace factor, which were substituted into the explanatory variables x14 to x18. Furthermore, the total weight obtained by adding the respective blend weights of scrap a to m was substituted into the explanatory variable x19. The actual amount of steel produced was then substituted into the dependent variable y to calculate the bias α0 and regression coefficients α1 to α19. y=α0+Σαi·xi(i=1~19) … (1)

[0015] Since the above electric furnace factor data can only be obtained after melting operations, we decided to use representative values ​​from past operational data for each type of steel being melted and manufactured. Possible representative values ​​include the mean, median, and maximum, but in this embodiment, we used the median.

[0016] Subsequently, a predetermined number (n) of the most recent slow periods (e.g., the latter two months within the most recent four months) of the above operating data was used as test data for the multiple regression model shown in the above formula (1), and the predicted tapping volume y* was calculated. Then, the root mean square error (RMSE) of the actual tapping volume y and the predicted tapping volume y*, shown in the following formula (2), which is an evaluation index of the multiple regression model, was calculated to be 2183 (kg / ch).

[0017] JPEG2026056071000001.jpg17152

[0018] In contrast, the RMSE of the conventional predicted tapping quantity ys with respect to the actual tapping volume y is 4289 (kg / ch), and the prediction accuracy of the tapping volume has been improved by approximately 50%. As a result, the desired number of ingot materials can be reliably ensured without causing excessive molten metal or insufficient tapping volume.

[0019] By the way, regarding the training data, even if an excessive number is secured by looking back at past operating data, the RMSE may deteriorate instead. This is shown in Figure 2.

[0020] Figure 2 shows the number of months of the most recent acquisition of training data on the horizontal axis and the RMSE of the multiple regression model on the vertical axis. In Figure 2, lines a, b, and c are for different steel grades. According to this, the RMSE is minimized when the number of months of the most recent acquisition of training data is 2 months, 4 months, and 5 months respectively. When the number of months of the most recent acquisition of training data is more than this, the RMSE rapidly or gradually increases and the prediction accuracy deteriorates. Generally, it is good to give the operating data within the most recent 2 to 4 months as training data to the multiple regression model.

[0021] In the above embodiment, the total weight obtained by adding the respective blending weights of scraps a to m was used as the explanatory variable x19. However, when sufficient prediction accuracy can be obtained, it is not particularly necessary to use the total weight as the explanatory variable. Also, in the above embodiment, a multiple regression model was used as the regression model, but other regression models such as the LightGBM model can also be used.

Claims

1. A method for predicting the amount of steel produced in an electric furnace, characterized in that, for each type of steel to be melted and manufactured, representative values ​​of the blending weight of the scrap used and each data of the electric furnace factors are used as explanatory variables in a regression model, the amount of steel produced is used as the dependent variable of the regression model, the regression model is trained by providing the most recent predetermined training data to the regression model, and a predicted value of the amount of steel produced during melting is obtained using the trained regression model.

2. The method for predicting the amount of steel tapped from an electric furnace according to claim 1, wherein the electric furnace factors are the total amount of O2 used and at least one of the amount of SiC used, the amount of carbon injection used, the amount of CaO used, and the amount of breeze used during one operation.

3. A method for predicting the amount of steel produced in an electric furnace, according to claim 1 or 2, wherein the total weight obtained by adding up the weights of each of the aforementioned ingredients is further used as an explanatory variable in the regression model.

Images