JP7401752B2 - Model construction device, prediction device, model construction method, prediction method, and computer program - Google Patents

Description

The present invention provides a model construction device and a model construction method for constructing a predictive model for predicting the S concentration in hot metal at the end of the KR process in which desulfurization is performed in the hot metal pretreatment step, and a KR process using the constructed predictive model. The present invention relates to a prediction device, a prediction method, and a computer program for predicting the S concentration in hot metal at the time of completion.

In the KR process, which performs desulfurization treatment in the hot metal pretreatment process, the immersion depth of the impeller and the amount of flux used (quicklime and Al ash) are usually adjusted based on the judgment of the operator, and desulfurization is carried out by mechanical stirring for about 15 minutes. be done. The KR process is greatly influenced by disturbances, and variations occur in the S concentration in the hot metal at the end of the KR process (hereinafter also referred to as "post-treatment S concentration"). If desulfurization is insufficient in the KR process, desulfurization using Mg is performed as an additional treatment to adjust the target S concentration, but Mg is expensive, and additional treatment reduces production. Efficiency will decrease. For this reason, it is required to be able to predict the S concentration after treatment and also to determine whether desulfurization failure has occurred. There is a need for a predictive model that predicts S concentration.

In the KR process, hot metal is desulfurized using, for example, a mechanical stirring type desulfurization apparatus (KR) 50 as shown in FIG. For example, flux (desulfurization agent: quicklime (CaO), Al ash) is added to the hot metal 5 in the KR ladle 51, and the impeller 53 is immersed and rotated by the drive device 55. In desulfurization by such a mechanical stirring method, the desulfurization reaction is promoted by strongly stirring the hot metal 5 and flux using the impeller 53.

Here, since the S present in the hot metal 5 is a factor that reduces the quality of the steel plate manufactured, it is important to reduce the S concentration after treatment by the KR process to the target S concentration when delivering it to the converter. be. On the other hand, if flux is added excessively in order to ensure that the S concentration after treatment is equal to or lower than the target S concentration in the KR process, manufacturing costs will increase. Therefore, when setting the operating conditions for the KR process, it is desirable to be able to accurately predict the S concentration after treatment.

For example, Patent Document 1 discloses a method that can accurately guide processing agent input conditions in a hot metal pretreatment operation, taking into account practical variations. In Patent Document 1, processing agent input conditions are determined by repeatedly inputting various operating conditions including treatment agent input conditions into a neural network, and evaluation is performed by adding practical variations to determine the appropriate treatment agent. Find input conditions. As a result, the accuracy of predicting the component concentration after treatment can be improved regardless of whether the correspondence characteristic between the processing agent injection conditions and the component concentration after treatment exhibits linearity or nonlinearity.

Japanese Patent Application Publication No. 8-269518

“Improvement of hot metal desulfurization technology”, Sumitomo Metals Vol. 45-3. P. 52-58, April 1993

However, the method of Patent Document 1 assumes only the case where hot metal preliminary treatment is performed in a converter, and does not assume the KR process.

The KR process has not been sufficiently studied so far because the process phenomena are complex and there are few detection ends. In particular, regarding the mixed slag that flows into the hot metal ladle when being discharged from the torpedo car to the hot metal ladle, if the amount of mixed slag increases, the desulfurization rate in the KR process (= (S concentration before treatment - S concentration after treatment) / The S concentration before treatment x 100 [%]) tends to decrease, and although this is presumed to affect the desulfurization rate in the KR process, no detailed study has been made.

Furthermore, if the S concentration after treatment can be predicted before implementing the KR process, it will be possible to judge the need to increase the flux in the current KR process or the need to implement additional treatment based on the predicted S concentration after treatment. It becomes possible. As a result, insufficient desulfurization in the KR process can be avoided and the amount of desulfurization agent used can be reduced, leading to improved productivity and reduced manufacturing costs.

Therefore, the present invention has been made in view of the above-mentioned problems, and the purpose of the present invention is to: Provides a model construction device capable of constructing a predictive model for predicting the S concentration in hot metal at the end of the KR process (i.e., post-treatment S concentration), and a device for predicting the post-treatment S concentration using the same. It's about doing.

In order to solve the above problems, according to one aspect of the present invention, a prediction model for predicting the S concentration in hot metal at the end of the KR process is developed before the start of the KR process in which desulfurization is performed in the hot metal pretreatment process. A model building device that constructs a learning model from past operation performance data in the KR process and mixed slag information regarding mixed slag that flows into the hot metal ladle when hot metal is discharged from the torpedo car to the hot metal ladle in past operations. A model construction device is provided that includes a model construction section that constructs a predictive model using the model construction section.

The mixed slag information may include at least one of slag mixed amount information indicating the amount of slag mixed into the hot metal ladle based on the operator's evaluation, and slag property information indicating the slag properties based on the operator's evaluation.

Further, the mixed slag information may include category information obtained by classifying images when hot metal is discharged from the torpedo car to the hot metal pot based on preset classification criteria.

Alternatively, the mixed slag information is based on at least one of slag mixed amount information indicating the amount of slag mixed into the hot metal ladle and slag property information indicating the slag properties from an image at the time of discharging hot metal from the torpedo car to the hot metal ladle. It may also include extracted image feature amounts.

The learning model may be a multiple regression model or a generalized linear model.

The present invention also provides a prediction device for predicting the S concentration in hot metal at the end of the KR process before the start of the KR process using the prediction model built by the above-mentioned model building device, A prediction device is provided that includes a model prediction calculation unit that predicts the S concentration in hot metal at the end of the KR process based on operational performance data and mixed slag information regarding mixed slag in the most recent operation.

The prediction device may include a determination unit that determines whether or not additional processing for promoting desulfurization in the current KR process is necessary based on the prediction result by the model prediction calculation unit.

In order to solve the problem, another aspect of the present invention provides a prediction device for predicting the S concentration in hot metal at the end of the KR process before the start of the KR process in which desulfurization is performed in the hot metal pretreatment process. There is a prediction target data acquisition unit that acquires operation performance data in the KR process in the most recent operation, and mixed slag information regarding mixed slag flowing into the hot metal ladle when discharging hot metal from the torpedo car to the hot metal ladle in the most recent operation. Based on the operation performance data and mixed slag information acquired by the prediction target data acquisition unit, KR A prediction device is provided that includes a model prediction calculation unit that predicts the S concentration in hot metal at the end of a process.

Furthermore, in order to solve the problem, according to another aspect of the present invention, before the start of the KR process in which desulfurization is performed in the hot metal pretreatment step, a prediction method for predicting the S concentration in hot metal at the end of the KR process is provided. A model construction method for building a model, which learns from past operational performance data in the KR process and mixed slag information regarding mixed slag that flows into the hot metal ladle when hot metal is discharged from the torpedo car to the hot metal ladle in past operations. A model building method is provided that includes building a predictive model using the model.

Moreover, in order to solve the problem, according to another aspect of the present invention, a prediction method for predicting the S concentration in hot metal at the end of the KR process, before the start of the KR process in which desulfurization is performed in the hot metal pretreatment process, is provided. There is a prediction target data acquisition step of acquiring operational performance data in the KR process in the most recent operation and mixed slag information regarding mixed slag flowing into the hot metal ladle when discharging hot metal from the torpedo car to the hot metal ladle in the most recent operation. Based on the operation performance data and mixed slag information acquired in the prediction target data acquisition step, using a prediction model built using a learning model from past operation performance data and mixed slag information in past operations, A prediction method is provided, including a model prediction calculation step of predicting the S concentration in hot metal at the end of the KR process.

Furthermore, in order to solve the problem, according to another aspect of the present invention, a computer is used to predict the S concentration in hot metal at the end of the KR process, before starting the KR process in which desulfurization is performed in the hot metal pretreatment process. A computer program for functioning as a model construction device for constructing a predictive model for the KR process, the computer program includes past operation performance data in the KR process, and information on discharging hot metal from a torpedo car to a hot metal pot in past operations. A computer program is provided that functions as a model construction unit that uses a learning model to construct a predictive model from mixed slag information regarding mixed slag flowing into a hot metal ladle.

Furthermore, in order to solve the problem, according to another aspect of the present invention, a computer is used to predict the S concentration in hot metal at the end of the KR process before starting the KR process in which desulfurization is performed in the hot metal pretreatment process. A computer program for functioning as a prediction device, the computer program includes operational performance data in the KR process in the most recent operation, and flow into the hot metal ladle at the time of discharging hot metal from the torpedo car to the hot metal ladle in the most recent operation. The prediction target data acquisition unit acquires mixed slag information regarding mixed slag, and the prediction target data acquisition unit uses a prediction model constructed using a learning model from past operation performance data and mixed slag information in past operations. A computer program is provided that functions as a model prediction calculation unit that predicts the S concentration in hot metal at the end of the KR process based on the acquired operational performance data and mixed slag information.

As explained above, according to the present invention, a prediction is made to predict the S concentration in hot metal at the end of the KR process before the KR process, based on information regarding mixed slag that has a high influence on the desulfurization rate in the KR process. Models can be built.

It is a schematic diagram showing an example of a mechanical stirring type desulfurization device. This is an example of an image (discharging image) when hot metal is discharged from a torpedo car to a hot metal pot. FIG. 1 is a functional block diagram showing the functional configuration of a KR process evaluation device according to an embodiment of the present invention. It is a schematic diagram which shows an example of arrangement|positioning of the imaging device which acquires the payout image and the inside image of a pot. It is a flowchart which shows the KR process evaluation method based on the same embodiment. It is a hot metal area chart showing the change in the area (number of pixels) of the hot metal area in the hot metal discharging image, and shows an example of a case where there is little slag mixed into the hot metal. It is a hot metal area chart showing the change in the area (number of pixels) of the hot metal area in the hot metal discharging image, and shows another example of a case where there is little slag mixed into the hot metal. It is a hot metal region area chart showing changes in the area (pixel number) of the hot metal region in the hot metal discharging image, and shows an example of a case where there is a rather large amount of slag mixed into the hot metal. It is a hot metal area chart showing the change in the area (number of pixels) of the hot metal area in the hot metal discharging image, and shows another example of a case where there is a little more slag mixed into the hot metal. FIG. 3 is an explanatory diagram showing an example of a connected image used to obtain mixed slag information. It is an explanatory view showing an example of an image inside a pot. FIG. 2 is an explanatory diagram illustrating image feature amounts based on a histogram of RGB components. FIG. 2 is an explanatory diagram illustrating image feature amounts related to bright areas in an in-pot image. FIG. 2 is a block diagram illustrating an example of the hardware configuration of an information processing device that functions as a KR process evaluation device according to the same embodiment. It is a graph showing the results of Examples.

DESCRIPTION OF THE PREFERRED EMBODIMENTS Preferred embodiments of the present invention will be described in detail below with reference to the accompanying drawings. Note that, in this specification and the drawings, components having substantially the same functional configurations are designated by the same reference numerals and redundant explanation will be omitted.

<1. Effect of mixed slag on desulfurization>
In the present embodiment, a prediction model for predicting the S concentration in hot metal at the end of the KR process in which desulfurization is performed in the hot metal pretreatment process is applied to the mixed slag that flows into the hot metal ladle when the hot metal is discharged from the torpedo car to the hot metal ladle. Constructed with consideration of mixed slag information. First, the influence of mixed slag on desulfurization in the KR process will be explained.

Factors that influence desulfurization in the KR process include, for example, mixed slag, stirring capacity of a mechanically stirred desulfurization device, hot metal temperature, and the like. Specifically, with regard to mixed slag, the desulfurization efficiency varies depending on the amount and property (basicity) of mixed slag at the time of discharging hot metal from the torpedo car to the hot metal pot. Regarding the stirring ability, variations occur due to the immersion depth of the impeller and the sticking of metal to the impeller and hot metal pot. Regarding hot metal temperature, variations in the arrival time of hot metal from the blast furnace to the steelmaking plant cause variations in hot metal temperature, which affects the desulfurization reaction.

As described above, various factors that influence desulfurization can be considered in the KR process, but as described above, due to the complexity of the process phenomena and the small number of detection points, sufficient studies have not been carried out to date. In particular, as shown in FIG. 2, regarding the mixed slag 3a that flows into the hot metal ladle when the hot metal 3 is discharged from the torpedo car to the hot metal ladle, the desulfurization rate in the KR process decreases as the amount of mixed slag increases. This is thought to affect the desulfurization rate in the KR process.

For example, when the relationship between the amount of mixed slag and the properties of the slag was investigated based on the operational results for a certain period, the results shown in Tables 1 and 2 below were obtained. Tables 1 and 2 show the results of qualitative judgments of the amount of slag mixed in and the properties of slag by an operator visually based on images of hot metal 3 being discharged from a torpedo car to a hot metal ladle as shown in Figure 2 for 336 data. It shows. For example, the amount of slag mixed in is determined in three categories (large, medium, and small) based on the ratio of mixed slag 3a, which appears dark, to hot metal 3, which appears bright in the image of FIG. Further, for example, the slag properties are determined in three categories (large, medium, and small) based on the flow of the hot metal 3 and mixed slag 3a in the image of FIG. Table 1 shows the results of organizing the number of data regarding the amount of slag mixed in and slag properties judged by the operator, and Table 2 shows the results organized by category regarding the amount of slag mixed in and slag properties shown in Table 1. The average value of the desulfurization rate of the data is shown.

From Table 2, it can be seen that when the overall average value is 1 and this is used as a standard, the desulfurization rate in the KR process tends to decrease when there is a large amount of mixed slag and when the viscosity of the slag increases. .

Based on these studies, the inventor of the present application developed a prediction model for predicting the S concentration in hot metal at the end of the KR process, using information about mixed slag, which is estimated to have a high influence on desulfurization in the KR process, as an explanatory variable. We came up with the idea of improving the accuracy of the prediction model by constructing a . Hereinafter, a model construction device for constructing a prediction model for predicting the S concentration in hot metal at the end of the KR process in which desulfurization is performed in the hot metal pretreatment process, and the constructed prediction model, according to an embodiment of the present invention. The outline of the prediction device using the following will be explained in detail.

<2. KR process evaluation device>
First, the functional configuration of the KR process evaluation apparatus 100 according to this embodiment will be described based on FIGS. 3 and 4. FIG. 3 is a functional block diagram showing the functional configuration of the KR process evaluation device 100 according to this embodiment. FIG. 4 is a schematic diagram illustrating an example of the arrangement of an imaging device 47 that acquires a dispensing image and an in-pot image (hereinafter also collectively referred to as "captured images"). The KR process evaluation device 100 includes a model construction device 110 that constructs a prediction model for predicting the S concentration in hot metal (i.e., the S concentration after treatment) at the end of the KR process, and a model construction device 110 that constructs a prediction model for predicting the S concentration in hot metal at the end of the KR process (that is, the S concentration after treatment), and the S concentration after treatment using the prediction model. and a prediction device 120 that predicts. In FIG. 3, the KR process evaluation device 100 is equipped with a model construction device 110 and a prediction device 120, but the present invention is not limited to this example, and each may be configured as an independent device.

[2-1. Model construction device】
The model construction device 110 includes a model construction data acquisition section 111 and a model construction section 113.

The model construction data acquisition unit 111 acquires past operational performance data in the KR process used in constructing the prediction model, and mixed slag related to mixed slag that flows into the hot metal ladle when hot metal is discharged from the torpedo car to the hot metal ladle in past operations. Get information. The model construction data acquisition section 111 includes a data input section 111a and an image data processing section 111b.

The data input unit 111a acquires past operation performance data and receives input of mixed slag information from an external device connected to the model construction device 110.

Past operation performance data is acquired from the operation performance data storage unit 200 in which such data is stored. The operation performance data storage unit 200 includes, for example, the amount of hot metal when performing the KR process, the hot metal ladle used, the number of impellers, the number of KR treatments, the height of the hot water level, the impeller immersion depth, the stirring speed, the stirring time, and the stirring time. Torque, stirring power, amount of CaO used, amount of Al dross used, Si concentration at the start of the KR process, Mn concentration at the start of the KR process, P concentration at the start of the KR process, Ti concentration at the start of the KR process, S concentration at the start of the KR process, KR The final temperature after the process, the final C concentration after the KR process, the tare, the target S concentration, etc. are stored for each KR process performed. The data input unit 111a acquires a predetermined number of past operation performance data used for constructing a prediction model from the operation performance data storage unit 200.

Further, the data input unit 111a receives input information from the external input device 500. The input device 500 is a device for an operator to input information, and includes, for example, a mouse, a keyboard, a touch panel, a button, a switch, a lever, and the like. In this embodiment, slag mixing amount information indicating the amount of slag mixed into the hot metal ladle based on the operator's evaluation and slag property information indicating the slag property based on the operator's evaluation are used as mixed slag information to construct the prediction model. . The operator inputs this information into the model construction device 110 using the input device 500.

Here, the operator can confirm the amount of slag mixed in by checking captured images such as a discharging image of the situation in which the hot metal is discharged from the torpedo car to the hot metal ladle and an inside image of the inside of the hot metal ladle 40 on the display device 400. and quantitatively evaluate the slag properties. The captured image is acquired by an imaging device 300 installed in a processing plant where hot metal is paid out. The imaging device 300 may be installed in the pulp pit, for example, like the imaging device 47 shown in FIG. At this time, the imaging device 47 may be composed of a plurality of cameras. For example, in the example of FIG. 4, the first camera 47a is installed at a position where the torpedo car 20A is tilted so that it can image the flow of hot metal when the hot metal 5 pours into the hot metal ladle 40 and is discharged. Similarly, the second camera 47b is installed at a position where the torpedo car 20B is tilted so that it can image the flow of hot metal when the hot metal 5 pours into the hot metal ladle 40 and is discharged. The first camera 47a and the second camera 47b capture a payout image of the area Q as shown in FIG. 2, for example.

Further, as the imaging device 47, a third camera 47c that images the molten metal 5 inside the hot metal ladle 40 may be installed. The third camera 47c is installed at a position where it can image the hot metal 5 discharged into the hot metal ladle 40 from the upper opening of the hot metal ladle 40. The third camera 47c images the hot metal 5 discharged into the hot metal ladle 40 and obtains an image inside the ladle.

Returning to the explanation of FIG. 3, the operator uses the input device 500 to input slag mixing amount information indicating the amount of slag mixed into the hot metal ladle and slag property information showing the slag properties based on the captured image acquired in this way. and input it to the data input section 111a. Further, the operator may, for example, classify captured images based on preset classification criteria and generate classification category information as mixed slag information. The classification category information is, for example, information obtained by classifying operational performance data corresponding to a captured image into a plurality of types based on the results of visually determining the amount and properties of mixed slag. Specifically, the amount of mixed slag is visually classified into three categories (large amount, medium amount, and small amount), and the viscosity of mixed slag is visually classified into three categories (high viscosity, medium viscosity, and low viscosity). A process is performed to classify the results into categories (lower or lower).

The data input unit 111a outputs the acquired past operation performance data and mixed slag information to the model construction unit 113.

The image data processing unit 111b analyzes the captured image acquired by the imaging device 300, and acquires image feature amounts as mixed slag information. The image feature amount is extracted based on, for example, slag mixing amount information or slag property information that is set in advance for feature amount extraction. Note that details of the processing by the image data processing section 111b will be described later. The image data processing unit 111b outputs the acquired mixed slag information to the model construction unit 113.

The model construction unit 113 constructs a prediction model for predicting the post-processing S concentration of the KR process using a learning model based on the past operation performance data and mixed slag information acquired by the model construction data acquisition unit 111. The model construction unit 113 may use, for example, a multiple regression model or a general linearization model as the learning model. The model construction unit 113 may also construct a predictive model using a machine learning method such as Lasso (Least Absolute Shrinkage and Selection Operator), RandomForest, or GBDT (Gradient Boosting Decision Tree) as a learning model. Details of the predictive model construction process by the model construction unit 113 will be described later. The model construction unit 113 outputs the constructed prediction model to the prediction device 120.

[2-2. Prediction device]
The prediction device 120 includes a prediction target data acquisition section 121, a model prediction calculation section 123, and a determination section 125.

The prediction target data acquisition unit 121 acquires operation performance data in the KR process in the most recent operation, and mixed slag information regarding mixed slag flowing into the hot metal ladle when hot metal is discharged from the torpedo car to the hot metal ladle in the most recent operation. . The latest operational performance data and mixed slag information acquired by the prediction target data acquisition unit 121 are information regarding the hot metal that will undergo the KR process from now on. Note that if there is no operational performance data, operation setting values may be used. The prediction target data acquisition section 121 includes a data input section 121a and an image data processing section 121b. The data input unit 121a and the image data processing unit 121b are the data input unit 111a and the image data processing unit of the model construction data acquisition unit 111 of the model construction device 110, except that the information to which the data to be handled is input is the latest data. This is the same process as 111b. Therefore, a detailed explanation of the data input section 121a and the image data processing section 121b will be omitted. The prediction target data acquisition unit 121 outputs the acquired latest operation performance data and mixed slag information to the model prediction calculation unit 123.

The model prediction calculation unit 123 uses the prediction model constructed by the model construction device 110 to determine the S concentration in hot metal at the end of the KR process based on the operation performance data and mixed slag information acquired by the prediction target data acquisition unit 121. Predict. By inputting the operational performance data and mixed slag information acquired by the prediction target data acquisition unit 121 into the prediction model, a predicted value of the S concentration in hot metal at the end of the KR process is calculated. The model prediction calculation unit 123 outputs the calculated predicted value of the S concentration in the hot metal at the end of the KR process to the determination unit 125.

The determination unit 125 determines whether or not additional processing is necessary to promote desulfurization in the current KR process, based on the prediction result by the model prediction calculation unit 123. The determination unit 125 compares the predicted value of the S concentration in the hot metal at the end of the KR process with the target S concentration, and determines whether additional processing is necessary. If the determination unit 125 determines that the predicted value of the S concentration in hot metal at the end of the KR process does not satisfy the target S concentration, the determination unit 125 may increase the amount of desulfurization agent used in the KR process or perform desulfurization using Mg, for example. The operator may be notified via an output device (not shown) that additional processing is required. At this time, the determination unit 125 may also notify the additional amount of the desulfurizing agent, etc., based on past operational results or preset additional treatment implementation requirements.

The configuration of the KR process evaluation apparatus 100 according to the present embodiment has been described above.

<3. KR process evaluation method>
Next, a KR process evaluation method using the KR process evaluation apparatus 100 according to this embodiment will be described based on FIGS. 5 to 13. FIG. 5 is a flowchart showing the KR process evaluation method according to this embodiment. 7 to 13 are explanatory diagrams for explaining mixed slag information. As shown in FIG. 5, the KR process evaluation by the KR process evaluation device 100 is performed by a model construction process by the model construction device 110 that is performed offline and a prediction process by the prediction device 120 that is performed online. Each process will be explained in detail below.

[3-1. Model construction process]
First, model construction processing by the model construction device 110 will be described. In the model construction process, a prediction model for predicting the post-processing S concentration of the KR process is constructed based on past operational performance data and past mixed slag information. The model construction process is performed offline, and the prediction by the prediction device 120 is performed in advance before the prediction is performed.

(S100: Acquisition of past operation performance data)
As shown in FIG. 5, the model construction data acquisition unit 111 of the model construction device 110 first acquires a predetermined number of past operation performance data from the operation performance data storage unit 200 using the data input unit 111a (S100). The past operation performance data to be acquired may be included within a predetermined period. As mentioned above, the operational performance data includes, for example, the amount of hot metal when performing the KR process, the hot metal pot used, the number of impellers, the number of KR treatments, the height of the hot water level, the immersion depth of the impeller, the speed of each plate, and each half. Time, stirring torque, stirring power, amount of CaO used, amount of Al dross used, Si concentration at the start of the KR process, Mn concentration at the start of the KR process, P concentration at the start of the KR process, Ti concentration at the start of the KR process, S at the start of the KR process Concentration, final temperature after KR process, final C concentration after KR process, tare, target S concentration, etc.

(S110: Acquisition of mixed slag information based on operator evaluation)
The data input unit 111a also inputs information such as slag mixing amount information indicating the amount of slag mixed into the hot metal ladle based on the operator's evaluation and slag property information indicating the slag property based on the operator's evaluation, which is input from the external input device 500. The mixed slag information is acquired (S110). The mixed slag information based on the operator's evaluation is based on the operator's qualitative evaluation, as described above. In step S110, for example, the amount of slag mixed in is determined based on the brightness of each pixel of the captured image, based on the ratio of mixed slag that appears dark to hot metal that appears bright. Further, for example, the slag properties are determined empirically from the flow of hot metal and mixed slag in the captured image.

Based on the captured image thus obtained, the operator uses the input device 500 to input slag mixing amount information indicating the amount of slag mixed into the hot metal ladle and slag property information showing the slag properties to the data input unit 111a. . Further, the operator may, for example, classify captured images based on preset classification criteria and generate classification category information as mixed slag information. The classification category information is, for example, information obtained by classifying operational performance data corresponding to a captured image into a plurality of types based on the results of visually determining the amount and properties of mixed slag. Specifically, the amount of mixed slag is visually classified into three categories (large amount, medium amount, and small amount), and the viscosity of mixed slag is visually classified into three categories (high viscosity, medium viscosity, and low viscosity). A process is performed to classify the results into categories (lower or lower). The evaluation by such an operator can be said to be the evaluation of the entire discharge of hot metal.

(S120: Image feature acquisition)
On the other hand, the model construction data acquisition unit 111 uses the image data processing unit 111b to extract image feature amounts from the captured image as mixed slag information (S120). The image feature amount is extracted based on, for example, slag mixing amount information or slag property information that is set in advance for feature amount extraction. The image feature amount acquired by the image data processing unit 111b is quantitatively evaluated, such as the amount of slag mixed in, and is unlikely to have individual variations. It is also possible to capture partial changes in the captured image.

In step S120, in order to treat the captured image of hot metal acquired by the imaging device 300 as a model variable of the prediction model, the captured image is analyzed by the image data processing unit 111b and image feature amounts are acquired. Examples of images that can be used to obtain model variables include a discharging image captured from the torpedo car (area Q in FIG. 4), an in-pot image captured of the hot metal in the hot metal pot, and the like. The process of acquiring image feature amounts from these images will be described in more detail below. Further, in step S120, in addition to acquiring the image feature amount, information in which the payout state is patterned may be generated from the captured image. Such information can also be used as model variables.

(a. Image processing of payout image)
In the image processing of the dispensing images, it is conceivable to calculate the number of pixels in the region corresponding to the molten metal for each dispensing image acquired at each time as an evaluation of the dispensing situation. The area corresponding to the molten metal has high brightness, as shown in FIG. 2, for example, and appears as a bright part in the dispensing image. Therefore, the area corresponding to the hot metal in the dispensing image can be specified by performing threshold determination on the R component, G component, and B component. Similarly, regarding the mixed slag, the area corresponding to the mixed slag has low brightness, as shown in FIG. 2, for example, and appears as a dark part in the payout image. Therefore, it is also possible to specify the area corresponding to the mixed slag in the dispensing image by determining the threshold value for the R component, G component, and B component.

In this way, by specifying the brightness or HSV color space values (Hue, Saturation, Value) for each pixel in the dispensing image, it is possible to determine the state of the hot metal at the time of dispensing as shown below. Information that enables evaluation (ie, information that can be considered as operational variables) can be obtained.

For example, by using a plurality of dispensing images during the period from the start to the end of dispensing hot metal, it is possible to grasp the amount of slag mixed in from the change in brightness throughout the dispensing.

6 to 9 show examples of hot metal region area charts showing changes in the area (number of pixels) of the hot metal region in the hot metal discharging image. FIGS. 6 and 7 show cases in which there is little slag mixed into the hot metal, and FIGS. 8 and 9 show cases in which there is a slightly large amount of slag mixed into the hot metal. In each of FIGS. 6 to 9, the horizontal axis represents the frame number. (i.e., time), and the vertical axis represents the number of pixels in the hot metal region. Therefore, it can be considered that the time period (frame) in which the number of pixels in the hot metal region is larger, the more hot metal is discharged and the less slag is mixed in.

For example, as shown in FIGS. 6 and 7, it can be seen that in cases where there is little slag mixed into the hot metal, the number of pixels in the hot metal region, that is, the amount of hot metal discharged tends to decrease over time. Further, in the hot metal region area chart of FIG. 6, the number of pixels in the hot metal region rapidly decreases after 500 frames. It is thought that large chunks of slag were mixed in these areas. Also in FIG. 7, slag discharge is large in areas where the number of pixels in the hot metal region is decreased, and slag discharge is decreased in areas where the number of pixels in the hot metal area is increased. In this way, the amount of slag mixed in can be determined from the change in the number of pixels in the hot metal region. Furthermore, it is also possible to grasp the amount of hot metal discharged into the hot metal ladle in one discharge from the total number of pixels in the hot metal area area chart. Also, from the slope of the hot metal area chart, it is possible to understand changes in the amount of hot metal discharged.

Further, as shown in FIGS. 8 and 9, for example, in a case where a relatively large amount of slag is mixed into the hot metal, the shape is different from the hot metal region area chart shown in FIGS. 6 and 7. For example, as shown in FIG. 8, the discharge of hot metal is small in the first half of the discharging period, and the discharge of hot metal increases in the second half. Further, as shown in FIG. 9, for example, although there is almost no change in the amount of hot metal discharged in the first half of the discharging period, it may sharply decrease in the second half. Furthermore, in both cases of FIGS. 8 and 9, the number of pixels in the hot metal region is reduced compared to FIGS. 6 and 7.

In this way, by using the hot metal region area chart, it is possible to evaluate the state of slag contamination by looking at the area change trend of areas with high brightness. Furthermore, it is also possible to quantify the amount of slag mixed in from the area of the chart.

Alternatively, a specific area may be cut out from a plurality of dispensing images (frames) captured by the same imaging device (first camera 47a or second camera 47b in FIG. 4) between the start and end of dispensing of hot metal. By connecting these images in chronological order and creating one connected image, it becomes possible to understand the sequence of molten metal dispensing. For example, as shown in FIG. 10, the image data processing unit 111b cuts out a region F of the payout image and obtains cutout images f1, f2, f3, . . . from each frame. Then, when the cutout images f1, f2, f3, . . . are connected in chronological order, a connected image as shown in the lower part of FIG. 10 is generated. From the connected images, it is possible to understand the situation in which the hot metal 3 flows. Moreover, in the connected image of FIG. 10, it can be seen that a lump of mixed slag 3a was discharged near the end of dispensing. In this way, by focusing on the range in which the mixed slag 3a appears in the connected image, it is also possible to create variables related to the mixed slag.

(b. Image processing of images inside the pot)
In actual operation, the operator visually observes the hot metal discharged into the hot metal ladle that receives the hot metal discharged from the torpedo car, and evaluates the amount of slag mixed in and the slag properties. From this, it is considered effective to create and use operating variables based on an image of the interior of the hot metal ladle captured by an imaging device (the third camera 47c in FIG. 4). For example, if we compare the image of the inside of the pot immediately after discharging shown in the upper part of FIG. 11 and the image of the inside of the pot immediately before the hot metal ladle is transported to the KR process treatment plant after discharging, shown in the lower part of FIG. 11, , the brightness is different, and it can be seen that the temperature of the hot metal decreases as time passes. The degree of cooling of the hot metal also differs depending on the amount of slag mixed in. Note that in the in-pot image as well, regions corresponding to hot metal and regions corresponding to mixed slag can be identified by brightness and HSV color space values, similar to the dispensing image.

As the image feature amount of the in-pot image, for example, a statistical amount obtained from a histogram of each of the RGB components may be used for the in-pot image. Specifically, as shown in Figure 12, a histogram representing the brightness distribution of the R component, G component, and B component is created from the image inside the pot, and the average value and standard deviation of the brightness obtained from the histogram are calculated as statistics. Find it as. Such statistics can be used as mixed slag information. At this time, changes in the mixed slag in the hot metal ladle are taken into account by calculating statistics for the images of the inside of the pot immediately after discharging and the images of the inside of the pot just before conveyance shown in Figure 12, and using them as mixed slag information. be able to. For example, the change in slag mixed in the hot metal ladle may be expressed by the difference between the averages of the histograms obtained for each of the images inside the ladle.

Furthermore, the size of a bright area in the pan image immediately after being paid out from the torpedo car may be used as the image feature amount of the pan pan image. The areas that appear bright are areas where the temperature of the hot metal is high, and it can be estimated that the more high-temperature areas there are, the less slag is mixed in. The bright area of the in-pot image can be specified using a predetermined brightness threshold set in advance. The size of the area may be expressed by, for example, the number of pixels, total area, maximum area, etc. For example, as shown in FIG. 13, the upper left pan image to be analyzed is decomposed into R, G, and B components as shown in the upper right, and then binarized as shown in the lower left. do. Thereby, it is possible to extract a bright region representing a high temperature portion that is desired to be specified. When a bright area is identified by the binarization process, the image data processing unit 111b may count the number of pixels in the area as shown in the lower right corner and calculate the total area of the bright area.

As described above, the image feature amount may be obtained from the captured image obtained by the imaging device and used as the mixed slag information.

(S130: Prediction model construction)
Returning to the explanation of FIG. 5, when past operation performance data is acquired in step S100 and mixed slag information is acquired in steps S110 and S120, the model construction unit 113 uses the past operation performance data and mixed slag information to , a prediction model for predicting the S concentration after the KR process is constructed using the learning model (S130). As the learning model, for example, a multiple regression model, a general linearization model, or a machine learning method can be used. For example, when using a multiple regression model, the S concentration at the start of the KR process (assumed to be t=0) and the stirring time T (assumed to be the end time of the KR process) are used for the following equation (0). The S concentration in the hot metal after the KR process (t=T [min]) can be expressed by the following formula (A).

Regarding the general linearized model, for example, in the desulfurization physical equation expressed by the following formula (0) (see Non-Patent Document 1), the S concentration after treatment is determined using the overall rate coefficient Ks and the desulfurization reached S concentration Se as parameters. expressed.

In other words, since the desulfurization physical formula is expressed as an exp function for overall rate coefficient x time, the link function is a log function, and since it is a continuous value estimation of the S concentration after treatment, the probability distribution is a gamma distribution, the following formula is obtained. It is expressed as (1) and (2).

Note that (-Ks·T) in equation (A) corresponds to log(p), and this part is expressed in a linear form. In the formula, x1, ..., xn are n explanatory variables, β0, ..., βn are parameters, and the parameters β0, ..., βn are obtained by maximum likelihood estimation.

The model construction unit 113 outputs the constructed prediction model expressed, for example, by equation (A) to the prediction device 120.

[3-2. Prediction processing]
Next, prediction processing by the prediction device 120 will be explained. In the model construction process, the post-treatment S concentration of the KR process is predicted using the prediction model constructed in the model construction process based on the most recent operational performance data and the most recent mixed slag information. Such prediction processing is performed online prior to implementation of the KR process. Therefore, the prediction result of the prediction process can be reflected in the KR process to be executed this time.

(S200: Acquisition of latest operational performance data)
As shown in FIG. 5, the prediction device 120 first acquires the latest operation performance data using the data input unit 121a of the prediction target data acquisition unit 121 (S200). Recent operational performance data includes composition data for blast furnaces, etc. For variables for which actual data does not exist, operation setting values may be used.

(S210: Image feature acquisition)
Next, the prediction target data acquisition unit 121 uses the image data processing unit 121b to extract image feature amounts from the captured image acquired during the most recent discharge of hot metal as mixed slag information (S210). The image feature amount may be obtained by the same process as step S120.

(S220: Calculation of predicted value of S concentration in hot metal after completion of KR process)
Then, the model prediction calculation unit 123 uses the prediction model built by the model construction device 110 in step S130 to calculate the time when the KR process ends based on the operation performance data and mixed slag information acquired in steps S200 and S210. The S concentration in hot metal is predicted (S220). By inputting the operational performance data and mixed slag information acquired by the prediction target data acquisition unit 121 into the prediction model, a predicted value of the S concentration in hot metal at the end of the KR process is calculated.

(S230: Determining whether additional processing is necessary)
After that, the determination unit 125 determines whether additional processing for promoting desulfurization in the current KR process is necessary based on the prediction result calculated in step S220 (S230). The determination unit 125 compares the predicted value of the S concentration in the hot metal at the end of the KR process with the target S concentration, and determines whether additional processing is necessary. If it is determined that the predicted value of the S concentration in the hot metal at the end of the KR process does not satisfy the target S concentration, the determination unit 125 may, for example, increase the desulfurization agent used in the KR process or perform desulfurization using Mg. The operator may be notified via an output device (not shown) that additional processing is required. At this time, the determination unit 125 may also notify the additional amount of the desulfurizing agent, etc., based on past operational results or preset additional treatment implementation requirements.

The KR process evaluation method using the KR process evaluation apparatus 100 according to the present embodiment has been described above.

<4. Hardware configuration>
Based on FIG. 14, the hardware configuration of the KR process evaluation device 100 according to this embodiment will be described. FIG. 14 is a block diagram showing an example of the hardware configuration of an information processing device 900 that functions as the KR process evaluation device 100 according to the present embodiment. Note that the KR process evaluation apparatus 100 may be constructed using separate information processing apparatuses 900, or may be constructed so that they function together using one information processing apparatus 900.

Information processing device 900 includes a CPU 901, ROM 903, and RAM 905. The information processing device 900 also includes a bus 907, an input I/F 909, an output I/F 911, a storage device 913, a drive 915, a connection port 917, and a communication device 919.

The CPU 901 functions as an arithmetic processing device and a control device. The CPU 901 controls all or part of the operations within the information processing apparatus 900 according to various programs recorded in the ROM 903, RAM 905, storage device 913, or removable recording medium 925. The ROM 903 stores programs used by the CPU 901, calculation parameters, and the like. The RAM 905 temporarily stores programs used by the CPU 901 or parameters that change as appropriate during program execution. These are interconnected by a bus 907 constituted by an internal bus such as a CPU bus.

The bus 907 is connected to an external bus such as a PCI (Peripheral Component Interconnect/Interface) bus via a bridge.

The input I/F 909 is an interface that receives input from an input device 921 that is an operating means operated by a user, such as a mouse, keyboard, touch panel, button, switch, lever, or the like. The input I/F 909 is configured as, for example, an input control circuit that generates an input signal based on information input by the user using the input device 921 and outputs it to the CPU 901. The input device 921 may be, for example, a remote control device using infrared rays or other radio waves, or an external device 927 such as a PDA that is compatible with the operation of the information processing device 900. A user of the information processing device 900 can operate the input device 921 to input various data to the information processing device 900 and instruct processing operations.

The output I/F 911 is an interface that outputs input information to an output device 923 that can visually or audibly notify the user. The output device 923 may be, for example, a display device such as a CRT display device, a liquid crystal display device, a plasma display device, an EL display device, a lamp, or the like. Alternatively, the output device 923 may be an audio output device such as a speaker or headphones, a printer, a mobile communication terminal, a facsimile, or the like. The output I/F 911 instructs the output device 923 to output processing results obtained from various processing executed by the information processing device 900, for example. Specifically, the output I/F 911 instructs the display device to display the processing result by the information processing device 900 in text or as an image. Further, the output I/F 911 instructs the audio output device to convert an audio signal such as audio data that has received a playback instruction into an analog signal and output the analog signal.

The storage device 913 is one of the storage units of the information processing device 900, and is a device for storing data. The storage device 913 is configured of, for example, a magnetic storage device such as an HDD (Hard Disk Drive), a semiconductor storage device, an optical storage device, a magneto-optical storage device, or the like. The storage device 913 stores programs executed by the CPU 901, various data generated by executing the programs, various data acquired from the outside, and the like.

The drive 915 is a reader/writer for recording media, and is built into or externally attached to the information processing apparatus 900. The drive 915 reads information recorded on the attached removable recording medium 925 and outputs it to the RAM 905. The drive 915 can also write information to a removable recording medium 925 attached thereto. The removable recording medium 925 is, for example, a magnetic disk, an optical disk, a magneto-optical disk, or a semiconductor memory. Specifically, the removable recording medium 925 includes CD media, DVD media, Blu-ray (registered trademark) media, CompactFlash (registered trademark) (CF), flash memory, and SD memory card (Secure Digital memory card). etc. may be used. Furthermore, the removable recording medium 925 may be, for example, an IC card (Integrated Circuit card) equipped with a non-contact IC chip, an electronic device, or the like.

The connection port 917 is a port for directly connecting a device to the information processing device 900. The connection port 917 is, for example, a USB (Universal Serial Bus) port, an IEEE1394 port, a SCSI (Small Computer System Interface) port, an RS-232C port, or the like. The information processing apparatus 900 can directly acquire various data from the external device 927 connected to the connection port 917 or provide various data to the external device 927.

The communication device 919 is, for example, a communication interface configured with a communication device for connecting to the communication network 929. The communication device 919 is, for example, a communication card for wired or wireless LAN (Local Area Network), Bluetooth (registered trademark), or WUSB (Wireless USB). Further, the communication device 919 may be a router for optical communication, a router for ADSL (Asymmetric Digital Subscriber Line), a modem for various communications, or the like. The communication device 919 can transmit and receive signals and the like to and from the Internet and other communication devices, for example, in accordance with a predetermined protocol such as TCP/IP. Further, a communication network 929 connected to the communication device 919 is configured by a wired or wirelessly connected network. For example, the communication network 929 is the Internet, a home LAN, infrared communication, radio wave communication, or satellite communication.

An example of the hardware configuration of the information processing device 900 has been described above. Each of the above-mentioned components may be constructed using general-purpose members, or may be constructed using hardware specialized for the function of each component. The hardware configuration of the information processing device 900 can be changed as appropriate depending on the technical level at which this embodiment is implemented.

In order to verify the effects of the present invention, the performance of the predictive model was determined by multiple regression analysis (stepwise method (AIC)) using the operational variables and mixed slag information (operator evaluation value, image feature amount) shown in Table 3 below. It was investigated. In the study, the explanatory variables shown in Table 3 below were set as explanatory variables that can be used when constructing a prediction model. In this study, operational performance data obtained when implementing the KR process was used as an operational variable. Here, 20 pieces of operational performance data were used, including, for example, the amount of hot metal, the number of impellers, the component concentration of hot metal before and after the KR process, etc. Then, the explanatory variables used in constructing the prediction model are changed as shown in cases A to F shown in Table 4 below, and for each case A to F, the post-treatment S concentration after the KR process is estimated by the prediction model. The S concentration after actual treatment was investigated. The results are shown in FIG.

Referring to FIG. 15, it can be seen that more accurate prediction models are constructed for Cases B to F, which are examples, than Case A, which is a comparative example. Note that the evaluation results in this study cannot be said only when using the operational variables and mixed slag information (operator evaluation, image feature amount) shown in Table 3. Operational variables and mixed slag information (even if the number of data for operator evaluation and image feature amount is different from Table 3, by adding at least one of the operator evaluation or image feature amount to the explanatory variables in addition to the operation variables, The accuracy of the prediction model can be increased.

Although preferred embodiments of the present invention have been described above in detail with reference to the accompanying drawings, the present invention is not limited to such examples. It is obvious that a person with ordinary knowledge in the technical field to which the present invention pertains can come up with various changes or modifications within the scope of the technical idea stated in the claims. It is understood that these also naturally fall within the technical scope of the present invention.

3, 5 Hot metal 3a Mixed slag 5 Hot metal 10 Blast furnace 20, 20A, 20B Torpedo car (TPC)
30 TPC treatment plant 40 Hot metal pot 45 Observation window 47 Imaging device 47a First camera 47b Second camera 47c Third camera 50 Mechanical stirring type desulfurization equipment (KR)
51 KR pot 53 Impeller 55 Drive device 100 KR process evaluation device 110 Model construction device 111 Model construction data acquisition section 111a Data input section 111b Image data processing section 113 Model construction section 120 Prediction device 121 Prediction target data acquisition section 121a Data input section 121b Image data processing unit 123 Model prediction calculation unit 125 Judgment unit 200 Operation performance data storage unit 200 Operation performance data storage 300 Imaging device 400 Display device 500 Input device

Claims (7)

A model construction device that constructs a predictive model for predicting the S concentration in hot metal at the end of the KR process before the start of the KR process, which is a process of desulfurization in the hot metal pretreatment process,
Past operation performance data in the KR process and past mixed slag based on captured images that are images of the situation in which hot metal is discharged from the torpedo car to the hot metal ladle in the past operations or images of the inside of the hot metal ladle. a model construction data acquisition unit that acquires information;
a model construction unit that constructs the prediction model based on the past operation performance data and the past mixed slag information ;
Equipped with
The mixed slag information includes slag mixed amount information indicating the amount of slag mixed into the hot metal ladle, slag property information indicating the viscosity of the slag, and slag property information indicating the viscosity of the slag, which was obtained by qualitatively evaluating the captured image by an operator. at least one of image features quantitatively extracted from the captured image;
The prediction model is constructed using a multiple regression model, a general linearization model, or a machine learning method, and uses the operation performance data and the mixed slag information as explanatory variables to estimate the S concentration in hot metal at the end of the KR process. A model construction device that is a model for The mixed slag information includes category information that is classified based on preset classification criteria by an operator visually judging the amount and properties of mixed slag from an image taken when hot metal is discharged from a torpedo car to a hot metal pot. , The model construction device according to claim 1 . A prediction device that predicts the S concentration in hot metal at the end of the KR process before the start of the KR process, which is a process of desulfurization in the hot metal pretreatment process,
Past operation performance data in the KR process and past mixed slag based on captured images that are images of the situation in which hot metal is discharged from the torpedo car to the hot metal ladle in the past operations or images of the inside of the hot metal ladle. a model construction data acquisition unit that acquires information;
a model construction unit that constructs a prediction model for predicting the S concentration in hot metal at the end of the KR process based on the past operation performance data and the past mixed slag information;
a prediction target data acquisition unit that acquires the latest operational performance data regarding the hot metal that will undergo the KR process from now on , and the latest mixed slag information regarding the hot metal that will implement the KR process from now on ;
Using the prediction model built by the model construction unit , predicting the S concentration in hot metal at the end of the KR process based on the most recent operational performance data and the most recent mixed slag information acquired by the prediction target data acquisition unit . A model prediction calculation section,
Equipped with
The mixed slag information includes slag mixed amount information indicating the amount of slag mixed into the hot metal ladle, slag property information indicating the viscosity of the slag, and slag property information indicating the viscosity of the slag, which was obtained by qualitatively evaluating the captured image by an operator. at least one of image features quantitatively extracted from the captured image;
The prediction model is constructed using a multiple regression model, a general linearization model, or a machine learning method, and uses the operation performance data and the mixed slag information as explanatory variables to estimate the S concentration in hot metal at the end of the KR process. A prediction device that is a model for A model construction method for constructing a predictive model for predicting the S concentration in hot metal at the end of the KR process before the start of the KR process, which is a process of desulfurization in the hot metal pretreatment process, comprising:
Past operation performance data in the KR process and past mixed slag based on captured images that are images of the situation in which hot metal is discharged from the torpedo car to the hot metal ladle in the past operations or images of the inside of the hot metal ladle. a model building data acquisition step for acquiring the information;
a model construction step of constructing the prediction model based on the past operation performance data and the past mixed slag information ;
including;
The mixed slag information includes slag mixed amount information indicating the amount of slag mixed into the hot metal ladle, slag property information indicating the viscosity of the slag, and slag property information indicating the viscosity of the slag, which was obtained by qualitatively evaluating the captured image by an operator. at least one of image features quantitatively extracted from the captured image;
The prediction model is constructed using a multiple regression model, a general linearization model, or a machine learning method, and uses the operation performance data and the mixed slag information as explanatory variables to estimate the S concentration in hot metal at the end of the KR process. model construction method. A prediction method for predicting the S concentration in hot metal at the end of the KR process before the start of the KR process, which is a process of desulfurization in the hot metal pretreatment process, comprising:
Past operation performance data in the KR process and past mixed slag based on captured images that are images of the situation in which hot metal is discharged from the torpedo car to the hot metal ladle in the past operations or images of the inside of the hot metal ladle. a model building data acquisition step for acquiring the information;
a model construction step of constructing a prediction model for predicting the S concentration in hot metal at the end of the KR process, based on the past operation performance data and the past mixed slag information;
a prediction target data acquisition step of acquiring the latest operational performance data regarding the hot metal that will undergo the KR process from now on, and the latest mixed slag information regarding the hot metal that will implement the KR process from now on ;
Using the prediction model built in the model construction step , predict the S concentration in the hot metal at the end of the KR process based on the most recent operational performance data and the most recent mixed slag information acquired in the prediction target data acquisition step . a model prediction calculation step;
including;
The mixed slag information includes slag mixed amount information indicating the amount of slag mixed into the hot metal ladle, slag property information indicating the viscosity of the slag, and slag property information indicating the viscosity of the slag, which was obtained by qualitatively evaluating the captured image by an operator. at least one of image features quantitatively extracted from the captured image;
The prediction model is constructed using a multiple regression model, a general linearization model, or a machine learning method, and uses the operation performance data and the mixed slag information as explanatory variables to estimate the S concentration in hot metal at the end of the KR process. A prediction method that is a model for A computer for operating a computer as a model construction device for constructing a predictive model for predicting the S concentration in hot metal at the end of the KR process, before the start of the KR process, which is a process of desulfurization in a hot metal pretreatment process. A program,
The computer program causes the computer to
Past operation performance data in the KR process and past mixed slag based on captured images that are images of the situation in which hot metal is discharged from the torpedo car to the hot metal ladle in the past operations or images of the inside of the hot metal ladle. a model construction data acquisition unit that acquires information;
Functioning as a model construction unit that constructs the prediction model based on the past operation performance data and the past mixed slag information ,
The mixed slag information includes slag mixed amount information indicating the amount of slag mixed into the hot metal ladle, slag property information indicating the viscosity of the slag, and slag property information indicating the viscosity of the slag, which was obtained by qualitatively evaluating the captured image by an operator. at least one of image features quantitatively extracted from the captured image;
The prediction model is constructed using a multiple regression model, a general linearization model, or a machine learning method, and uses the operation performance data and the mixed slag information as explanatory variables to estimate the S concentration in hot metal at the end of the KR process. A computer program that is a model for A computer program for causing a computer to function as a prediction device for predicting the S concentration in hot metal at the end of the KR process before the start of the KR process, which is a process of desulfurization in a hot metal pretreatment process, comprising:
The computer program causes the computer to
Past operation performance data in the KR process and past mixed slag based on captured images that are images of the situation in which hot metal is discharged from the torpedo car to the hot metal ladle in the past operations or images of the inside of the hot metal ladle. a model construction data acquisition unit that acquires information;
a model construction unit that constructs a prediction model for predicting the S concentration in hot metal at the end of the KR process based on the past operation performance data and the past mixed slag information;
a prediction target data acquisition unit that acquires the latest operational performance data regarding the hot metal that will undergo the KR process from now on , and the latest mixed slag information regarding the hot metal that will implement the KR process from now on ;
function as
The mixed slag information includes slag mixed amount information indicating the amount of slag mixed into the hot metal ladle, slag property information indicating the viscosity of the slag, and slag property information indicating the viscosity of the slag, which was obtained by qualitatively evaluating the captured image by an operator. at least one of image features quantitatively extracted from the captured image;
The prediction model is constructed using a multiple regression model, a general linearization model, or a machine learning method, and uses the operation performance data and the mixed slag information as explanatory variables to estimate the S concentration in hot metal at the end of the KR process. A computer program that is a model for

Images