Method for predicting tapping amount, device for predicting tapping amount, method for operating electric furnace, and method for creating machine learning model

Abstract

To appropriately predict a tapping amount of steel from an electric furnace.SOLUTION: Raw material information relating to a raw material charged into an electric furnace and including raw material charge amount data representing a total charge amount of the raw material, and operation information relating to a condition under which the electric furnace is operated and including input power data representing power to be input into the electric furnace are input into a tapping amount prediction model that outputs a tapping amount from the electric furnace based on the raw material information and the operation information, and the tapping amount from the electric furnace is predicted. Further, the electric furnace is operated according to the operation condition in which the steel tapping quantity having smaller difference with the charging quantity of the raw material charged into the electric furnace, is predicted.SELECTED DRAWING: Figure 3

Description

[Technical Field]

[0001] The present disclosure relates to a method for predicting the amount of steel tapped in an electric furnace (hereinafter simply referred to as an "electric furnace"), such as an arc-type electric furnace, a steel tapping amount prediction device, an electric furnace operation method, and a machine learning model creation method. [Background technology]

[0002] Electric furnaces produce molten iron by generating heat using the electricity supplied to the furnace, which melts raw materials (e.g., iron-based scrap) charged into the furnace. It is important to not supply more electricity than necessary to the furnace in order to reduce the operating costs of the electric furnace. However, if the electricity supplied to the furnace is insufficient, some raw materials may remain unmelted. This unmelted material not only reduces the amount of molten iron that should be produced from the raw materials, but can also cause problems such as affecting the raw material composition in the next operation of the electric furnace, so there is a need to reduce it.

[0003] Patent Document 1 describes a method for predicting the impurity concentration of molten iron refined in an electric furnace facility using an impurity concentration prediction model. [Prior art documents] [Patent documents]

[0004] [Patent Document 1] Patent No. 7207624 Summary of the Invention [Problem to be solved by the invention]

[0005] It is possible to reduce the amount of residual slag by creating a residual slag amount prediction model that predicts the amount of residual slag based on raw material information about the raw materials charged into the electric furnace and operational information about the conditions under which the electric furnace is operated, and predicting the amount of residual slag using the residual slag amount prediction model. However, since it is difficult to directly measure the amount of residual slag, creating a residual slag amount prediction model is also difficult.

[0006] However, the amount of unmelted molten metal can be indirectly estimated by subtracting the tapping amount from the mass of raw materials charged into the electric furnace. Therefore, an object of the present disclosure is to provide a method for predicting the tapping amount of steel that can appropriately predict the tapping amount of steel from an electric furnace. [Means for solving the problem]

[0007] The gist of the present disclosure is as follows.

[0008] (1) A method for predicting the amount of steel tapped from an electric furnace, comprising: inputting raw material information, including raw material charge amount data indicating the total amount of raw materials charged into the electric furnace, and operational information, including input power data indicating the power input to the electric furnace, regarding the conditions under which the electric furnace is operated, into a tapping amount prediction model that outputs the amount of steel tapped from the electric furnace based on the raw material information and the operational information, and predicting the amount of steel tapped from the electric furnace.

[0009] (2) The method for predicting the amount of steel tapping described in (1) above, wherein the model for predicting the amount of steel tapping is a machine learning model trained in advance, with the raw material information and the operational information as explanatory variables and the amount of steel tapping as a target variable.

[0010] (3) A steel tapping amount prediction device that inputs raw material information, including raw material charge amount data indicating the total amount of raw materials charged into an electric furnace, and operational information, including input power data indicating the power input to the electric furnace, regarding the conditions under which the electric furnace is operated, into a steel tapping amount prediction model that outputs the steel tapping amount from the electric furnace based on the raw material information and the operational information, and predicts the steel tapping amount from the electric furnace.

[0011] (4) predicting the amount of steel tapped from the electric furnace by inputting raw material information about the raw materials charged into the electric furnace and operational information about a first operational condition under which the electric furnace is operated into a steel tapping amount prediction model that outputs the amount of steel tapped from the electric furnace based on raw material information including raw material charge amount data indicating the total amount of the raw materials charged and operational information including input power data indicating the power input to the electric furnace under the operation of the electric furnace; predicting the amount of steel tapped from the electric furnace by inputting the raw material information about the raw materials to be charged into the electric furnace and the operational information about second operational conditions under which the electric furnace is operated, which second operational conditions are different from the first operational conditions, into the steel tapping amount prediction model; operating the electric furnace according to the operating conditions that predict a steel tapping amount that is smaller in difference from the amount of raw materials charged into the electric furnace, out of the first operating conditions and the second operating conditions; How to operate an electric furnace.

[0012] (5) A method for creating a machine learning model, in which a machine learning model is used in which, with regard to raw materials charged into an electric furnace, raw material information including raw material charge amount data representing the total amount of the raw materials charged, and operational information regarding the conditions under which the electric furnace is operated, including input power data representing the power input to the electric furnace, are used as explanatory variables, and the amount of steel tapped from the electric furnace is used as a target variable, and the raw material information, the operational information, and the amount of steel tapped from past operations of the electric furnace are used as training data.

[0013] (6) A method for creating a machine learning model according to (5) above, wherein the machine learning model is trained using data obtained by performing at least one of standardization, selection of items to be used as the explanatory variables, elimination of outliers, and reduction of data in which the objective variable corresponds to a predetermined range around the median, on data relating to past operations of the electric furnace, as training data. [Effects of the Invention]

[0014] According to the method for predicting the amount of steel tapped according to the present disclosure, the amount of steel tapped from an electric furnace can be appropriately predicted. [Brief explanation of the drawings]

[0015] [Figure 1] FIG. 1 is a schematic diagram showing an outline of operation in an electric furnace. [Figure 2] FIG. 1 is a schematic diagram showing the general configuration of a steel tapping amount prediction device. [Figure 3] FIG. 1 is a schematic diagram showing an outline of a steel tapping rate prediction model. [Figure 4] 1 is a table showing the prediction accuracy of the steel tapping amount prediction model. [Figure 5] Graph (a) is a first graph showing the prediction accuracy of the steel tapping amount prediction model, graph (b) is a second graph showing the prediction accuracy of the steel tapping amount prediction model, and graph (c) is a third graph showing the prediction accuracy of the steel tapping amount prediction model. [Figure 6] 1 is a table showing prediction results using a steel tapping rate prediction model. [Figure 7] 1 is a schematic diagram showing an overview of an electric furnace operation system. [Figure 8] 1 is a process flowchart of a method for operating an electric furnace. DETAILED DESCRIPTION OF THE INVENTION

[0016] Hereinafter, a steel tapping amount prediction device and a steel tapping amount prediction method that appropriately predict the steel tapping amount from an electric furnace will be described in detail with reference to the drawings. However, it should be understood that the present disclosure is not limited to the drawings or the embodiments described below.

[0017] FIG. 1 is a schematic diagram showing an overview of the operation of an electric furnace. When an electrode EL is inserted into an electric furnace EF into which raw material M has been charged and power is applied, an arc discharge occurs between the electrode EL and the raw material M, and the raw material M is melted by the heat of the arc. Molten iron LS, which is mainly composed of iron, is extracted from the melted raw material M from the electric furnace EF (hereinafter, the mass of the extracted molten iron LS is referred to as the tapping amount). The period from the charging of raw material M to the extraction of the molten iron LS is also referred to as charging. In charging, the portion of the melted raw material M other than the molten iron LS (slag) may be extracted from the electric furnace EF after the molten iron LS.

[0018] The components of raw material M that are not melted in the electric furnace EF (unmelted portion UD) remain in the electric furnace EF. It is difficult to directly measure the mass of unmelted portion UD. However, since a portion of raw material M charged into the electric furnace EF is extracted as molten iron LS and another portion remains unmelted as unmelted portion UD, the mass of unmelted portion UD can be roughly calculated by subtracting the amount of tapped steel from the mass of raw material M.

[0019] FIG. 2 is a schematic diagram showing the general configuration of the steel tapping amount prediction device 1.

[0020] The steel tapping amount prediction device 1 is a computer including an input / output interface 11, a memory 12, and a processor 13.

[0021] The input / output interface 11 has an interface circuit for receiving data to be processed by the steel tapping amount prediction device 1 or for outputting data processed by the steel tapping amount prediction device 1. The input / output interface 11 includes, for example, a communication interface circuit for connecting the steel tapping amount prediction device 1 to a communication network, or a peripheral device interface circuit for connecting the steel tapping amount prediction device 1 to various peripheral devices such as a keyboard, a mouse, and a display. The input / output interface 11 receives operation input from a terminal device via the communication interface circuit, or from input devices such as a keyboard and a mouse via the peripheral device interface circuit. The input / output interface 11 also outputs data representing the predicted steel tapping amount to the terminal device via the communication interface circuit, or to a display via the peripheral device interface circuit.

[0022] The memory 12 includes, for example, at least one of a semiconductor memory, a magnetic disk device, and an optical disk device. The memory 12 stores data, computer programs, etc. used for processing by the processor 13. The data stored in the memory 12 includes parameters representing a steel tapping amount prediction model.

[0023] The memory 12 stores computer programs such as a driver program, an operating system program, and a computer program for predicting the amount of steel tapped. The computer programs may be provided in a form recorded on a computer-readable portable recording medium such as a semiconductor memory, a magnetic recording medium, or an optical recording medium.

[0024] The processor 13 includes one or more processors and their peripheral circuits. The processor 13 is a processing circuit that comprehensively controls the overall operation of the steel tapping amount prediction device 1, and is, for example, a CPU (Central Processing Unit). The processor 13 controls the operation of the input / output interface 11, etc., so that various processes of the steel tapping amount prediction device 1 are executed by appropriate means based on computer programs, etc. stored in the memory 12. The processor 13 executes processes based on the various programs stored in the memory 12. The processor 13 can also execute multiple various programs in parallel.

[0025] Figure 3 is a schematic diagram showing an overview of the tapping amount prediction model. The tapping amount prediction model PM outputs the tapping amount SW based on raw material information MI and operational information OI. The raw material information MI includes raw material charge amount data that indicates the total amount of raw materials M charged into the electric furnace EF. The operational information OI includes input power data that indicates the power input to the electric furnace EF.

[0026] The tapping amount prediction model PM is a machine learning model trained in advance, with the raw material information MI and the operational information OI as explanatory variables and the tapping amount SW (tons) as a target variable. Alternatively, the tapping amount prediction model PM may be a statistical model generated by performing statistical processing such as multiple regression analysis on the actual data of the raw material information MI, the operational information OI, and the tapping amount SW.

[0027] The raw material information MI is information relating to the raw material M charged into the electric furnace EF, and includes, as an item, for example, the raw material charge amount indicating the mass of the raw material M charged into the electric furnace EF before the start of melting.

[0028] The operational information OI is information about the conditions under which the electric furnace EF is operated, and includes, for example, an input power indicating the average value of the power input to the electric furnace EF per unit time from the start to the end of melting.

[0029] The tapping amount prediction model PM, which is a machine learning model, is trained using training data in which raw material information MI and operational information OI collected from past operational records are used as explanatory variables and the tapping amount SW is used as a target variable. The training data is, for example, data in a table format in which items corresponding to the raw material information MI, operational information OI, and SW are arranged for the number of charges. In this embodiment, the training data is data for an electric furnace EF with a furnace capacity (maximum molten steel amount) of 73 tons.

[0030] The items included in the training data have different units. By standardizing the data included in the training data for each item, the prediction accuracy of the steel tapping volume prediction model PM can be improved. For example, each piece of data can be standardized and converted to a value between -1 and 1 using the following formula (1).

[0031]

number

[0032] where x i is each data of the i-th item, and μ i and σ i are the mean and standard deviation of each data item in the i-th item, respectively, and x i ′ is the standardized data for the i-th item.

[0033] The items included as explanatory variables in the training data may have a low correlation with the amount of steel tapped. Items with a low correlation with the amount of steel tapped not only do not contribute to the prediction of the amount of steel tapped, but may also cause a decrease in the accuracy of the prediction. By calculating the degree of contribution (contribution rate) of each item included as an explanatory variable in the training data and removing items with a low contribution rate from the training data, it is possible to appropriately select items to be included as explanatory variables in the training data.

[0034] The contribution of the items included as explanatory variables in the training data to the steel tapping volume prediction can be determined, for example, by Lasso analysis, which is based on linear regression (multiple regression). In Lasso analysis, the following equation (2) is solved for the coefficient β.

[0035]

number

[0036] Here, N is the number of data included in the training data, y is the tapping volume, X is an explanatory variable, and λ is an adjustment parameter.

[0037] Since the coefficient β tends to be 0 for explanatory variables that do not contribute to the prediction of the tapping volume, the contribution of each explanatory variable can be determined by whether the coefficient β is 0 or not.

[0038] Various known machine learning models can be applied to the steel tapping volume prediction model PM, such as nonlinear multiple regression, decision tree, random forest, support vector machine, Gaussian process regression, Lasso, Bayesian neural network, and LightGBM. Below, we explain an example in which LightGBM is used as the steel tapping volume prediction model PM.

[0039] The coefficient of determination R is used as a measure of the prediction accuracy of the steel tapping volume prediction model PM. 2 , root mean square error (RMSE), and mean absolute error (MAE) can be used. Each index is defined by the following equations (3)-(5).

[0040]

number

[0041]

number

[0042]

number

[0043] Here, N is the number of data included in the training data, and y i obs and y i pre are the measured and predicted values ​​of the tapping rate for the i-th data, respectively, and y ave is the average value of the measured tapping volume. The units of the root mean square error (RMSE) and the mean absolute error (MAE) are both ton, the same as the tapping volume.

[0044] FIG. 4 is a table showing the prediction accuracy of the tapping amount prediction model PM. The column "Before Data Cleansing" in Table TP shows the prediction accuracy of the tapping amount prediction model PM trained using training data before data cleaning, which will be described later. FIG. 5(a) is a first graph showing the prediction accuracy of the tapping amount prediction model, in which the horizontal axis represents the actual measured value per furnace capacity and the vertical axis represents the predicted value per furnace capacity. As shown in these graphs, the tapping amount prediction model PM can predict the tapping amount SW with a predetermined accuracy.

[0045] It would be more desirable if the prediction accuracy of the tapping amount prediction model PM could be further improved (for example, the error would be within 2% (1 ton) of the median for the electric furnace EF, whose median tapping amount SW is 57 tons). The prediction accuracy of the tapping amount prediction model PM can be further improved by excluding data that is not suitable for highly accurate prediction of the tapping amount SW from the training data (also known as "data cleansing").

[0046] Training data may contain outliers, data whose values ​​are significantly different from the rest. Although it is possible to infer the cause of outliers, such as measurement error, it is difficult to identify the specific cause, and this can significantly reduce prediction accuracy. Therefore, by excluding outliers from the training data, the prediction accuracy of the steel tapping volume prediction model PM can be further improved.

[0047] Cook's distance, which indicates the degree of influence that individual data has on the prediction, can be used as an index for detecting outliers from data. Data with a Cook's distance greater than a threshold has a large influence on the prediction and may be an outlier. Cook's distance is defined by the following equation (6):

[0048]

number

[0049] Here, *j is the jth predicted value, *j(i) is the jth predicted value based on data excluding the i-th data, and p is the total number of items (in this paragraph, "*" represents the character "y" with a "^" above it).

[0050] In this embodiment, for data of approximately 6000 charges, the threshold value was set to 4 / N, and data whose Cook distance calculated according to formula (6) exceeded the threshold value was deleted, and the operation of deleting the top 1% of data with the largest Cook distance was repeated 29 times (data cleansing A). As a result, the number of data was reduced to approximately 4000 charges.

[0051] The column "Data Cleansing A" in Table TP in Figure 4 shows the prediction accuracy of the tapping amount prediction model PM trained using the training data after data cleansing A. Figure 5(b) is a second graph showing the prediction accuracy of the tapping amount prediction model, in which the horizontal axis represents the actual measured value per furnace capacity and the vertical axis represents the predicted value per furnace capacity. As shown in these graphs, by training the tapping amount prediction model PM using the training data that has been subjected to data cleansing A, it is possible to predict the tapping amount SW with even higher accuracy.

[0052] In the operation of an electric furnace (EF), the tapping amount SW for each charge is often kept at a roughly constant value. As a result, the training data contains a large amount of data for the tapping amount SW that is close to a statistically representative value (e.g., the median). A tapping amount prediction model PM trained using such training data may overfit data for which the tapping amount SW is close to the statistically representative value. As a result, while such a tapping amount prediction model can make highly accurate predictions when the tapping amount SW is close to the statistically representative value, its prediction accuracy may decrease when the tapping amount SW deviates from this value.

[0053] By reducing the data in the vicinity of the statistical representative value, it is possible to avoid a decrease in prediction accuracy when the tapping amount SW deviates from the vicinity of the statistical representative value. In the tapping amount prediction model PM of this embodiment, the data that fell within a range twice the prediction error (±2 tons) obtained based on the median value of the tapping amount SW was reduced by half from the data after data cleansing A (data cleansing A+B). As a result of data cleansing A+B, the number of data points for approximately 4,000 charges after data cleansing A was reduced to approximately 2,250 charges.

[0054] The column "Data Cleansing A+B" in Table TP in Figure 4 shows the prediction accuracy of the tapping amount prediction model PM trained using the training data after data cleansing A+B. Figure 5(c) is a third graph showing the prediction accuracy of the tapping amount prediction model, where the horizontal axis represents the actual measured value per furnace capacity and the vertical axis represents the predicted value per furnace capacity. As shown in these graphs, the tapping amount prediction model PM can predict the tapping amount SW with even higher accuracy by training using training data that has been subjected to data cleansing A+B.

[0055] By training the machine learning model in this way, the prediction accuracy of the tapping amount prediction model PM can be improved. Figure 6 is a table showing the prediction results of the tapping amount prediction model PM. As shown in Table TR, the tapping amount prediction model PM predicts the tapping amount SW for each charge with high accuracy, with an absolute error of less than 1 ton.

[0056] The learning method for the machine learning model of this embodiment can also be applied to a prediction model that uses items other than the tapping amount in the operation of an electric furnace EF as objective variables. Examples of objective variables for a prediction model to which the learning method for the machine learning model can be applied include the temperature of the molten steel being taken out (tapping temperature) and the proportion of the charged Si that adheres to the molten steel.

[0057] Next, a method for operating an electric furnace EF using the method for predicting the steel tapping rate according to this embodiment will be described. Fig. 7 is a schematic diagram showing an overview of an electric furnace operation system 100, and Fig. 8 is a processing flowchart of the method for operating an electric furnace EF.

[0058] The electric furnace operation system 100 includes a tapping rate prediction device 1, an operation control device 2, and an electric furnace EF. The operation control device 2 is a computer that controls the operation of the electric furnace EF based on the tapping rate predicted by the tapping rate prediction device 1. The tapping rate prediction device 1 and the operation control device 2 are communicatively connected via a communication network NW.

[0059] The steel tapping amount prediction device 1 predicts the steel tapping amount SW (first predicted steel tapping amount) from the electric furnace EF by inputting raw material information MI regarding the raw materials to be charged into the electric furnace EF and operational information OI regarding the first operational conditions under which the electric furnace EF is operated into the steel tapping amount prediction model PM (step S11).

[0060] Next, the steel tapping amount prediction device 1 predicts the steel tapping amount SW (second predicted steel tapping amount) from the electric furnace EF by inputting raw material information MI regarding the raw materials to be charged into the electric furnace EF and operational information OI regarding second operational conditions under which the electric furnace EF is operated, which are different from the first operational conditions under which the electric furnace EF is operated, into the steel tapping amount prediction model PM (step S12).

[0061] The operation control device 2 operates the electric furnace EF according to the operating condition, out of the first operating conditions and the second operating conditions, that predicts a steel output SW that is smaller in difference from the amount of raw materials charged to the electric furnace EF (step S13), and terminates the electric furnace operation process.

[0062] The operating conditions under which the predicted tapping rate SW is smaller than the difference between the amount of raw materials charged into the electric furnace EF can also be said to be operating conditions under which a larger tapping rate SW is predicted. That is, in step S13, the operation control device 2 operates the electric furnace EF according to the first operating conditions when the first predicted tapping rate is larger than the second predicted tapping rate, and according to the second operating conditions when the opposite is true.

[0063] The tapping amount prediction device 1 may predict the tapping amount SW for more than two operating conditions. In this case, the operation control device 2 identifies the tapping amount SW that has the smaller difference from the amount of raw materials charged to the electric furnace EF from the predicted tapping amount SW, and operates the electric furnace EF according to the operating conditions corresponding to the identified tapping amount. The tapping amount prediction device 1 may also predict the tapping amount SW for different raw material information OI.

[0064] It should be understood that those skilled in the art can make various changes, substitutions, and alterations thereto without departing from the spirit and scope of the present disclosure. [Explanation of symbols]

[0065] 1. Steel tapping volume prediction device 2 Operation control device PM tapping volume prediction model EF electric furnace

Claims

1. 1. A method for predicting the amount of steel tapped from an electric furnace, comprising: inputting raw material information, regarding raw materials to be charged into an electric furnace, including raw material charge amount data indicating a total amount of the raw materials charged, and operational information, regarding conditions under which the electric furnace is operated, including input power data indicating the power input to the electric furnace, into a tapping amount prediction model that outputs the amount of steel tapped from the electric furnace based on the raw material information and the operational information, and predicting the amount of steel tapped from the electric furnace.

2. 2. The method for predicting the amount of steel tapping according to claim 1, wherein the model for predicting the amount of steel tapping is a machine learning model trained in advance, with the raw material information and the operation information as explanatory variables and the amount of steel tapping as a target variable.

3. A steel tapping amount prediction device that inputs raw material information, including raw material charge amount data that indicates a total amount of raw materials charged into an electric furnace, and operational information, including input power data that indicates the power input to the electric furnace, regarding the conditions under which the electric furnace is operated, into a steel tapping amount prediction model that outputs the steel tapping amount from the electric furnace based on the raw material information and the operational information, and predicts the steel tapping amount from the electric furnace.

4. predicting the amount of steel tapped from the electric furnace by inputting raw material information about the raw materials charged into the electric furnace and operational information about first operational conditions under which the electric furnace is operated into a tapping amount prediction model that outputs the amount of steel tapped from the electric furnace based on raw material information, the raw material information including raw material charge amount data indicating a total amount of the raw materials charged, and operational information, the operating conditions under which the electric furnace is operated, including input power data indicating the power input to the electric furnace; predicting the amount of steel tapped from the electric furnace by inputting the raw material information about the raw materials to be charged into the electric furnace and the operational information about second operational conditions under which the electric furnace is operated, which second operational conditions are different from the first operational conditions, into the steel tapping amount prediction model; operating the electric furnace according to one of the first and second operating conditions that predicts a steel tapping amount that is smaller in difference from the amount of raw materials charged into the electric furnace; How to operate an electric furnace.

5. A method for creating a machine learning model, in which a machine learning model is used in which raw material information, including raw material charge amount data indicating the total amount of raw materials charged into an electric furnace, and operational information, including input power data indicating the power input to the electric furnace, regarding the conditions under which the electric furnace is operated, are used as explanatory variables, and the amount of steel tapped from the electric furnace is used as a target variable, and the raw material information, the operational information, and the amount of steel tapped from past operations of the electric furnace are used as training data.

6. 6. The method for creating a machine learning model according to claim 5, wherein the machine learning model is trained using data obtained by performing at least one of standardization, selection of items to be used as the explanatory variables, elimination of outliers, and reduction of data in which the objective variable corresponds to a predetermined range near a median, on data related to past operations of the electric furnace, as the training data.

Images