The methods and systems for guiding the coke ratio action of a blast furnace, the control devices for a blast furnace, the coke ratio action guidance program for a blast furnace, the information output devices, the operation methods of a blast furnace, and the methods for manufacturing molten iron.

Abstract

The coke ratio action guidance method for a blast furnace of the present invention includes: a model learning processing step, which takes image data obtained by image processing of historical data including observations of ventilation, tapping speed and molten iron temperature in the blast furnace process as input data, and takes historical data including labels representing the operation direction of coke ratio assigned by the operator within a specified time range of the operation timing of coke ratio as output data for machine learning, thereby generating a machine learning model that takes image data during the guidance prompt period as input variables and action prediction values ​​representing the operation direction of coke ratio during the guidance prompt period as output variables; an action prediction value calculation step, which calculates the action prediction value during the guidance prompt period by inputting image data during the guidance prompt period into the machine learning model generated in the model learning processing step; and an action guidance prompt step, which prompts the operation action of coke ratio based on the action prediction value calculated in the action prediction value calculation step.

Description

[Technical Field]

[0001] This invention relates to a method for guiding the coke ratio action of a blast furnace, a system for guiding the coke ratio action of a blast furnace, a blast furnace control device, a program for guiding the coke ratio action of a blast furnace, an information output device, a method for operating a blast furnace, and a method for manufacturing molten iron. [Previous Technology]

[0002] In a blast furnace (smelting furnace) in the ironmaking industry, iron ore and coke, used as raw materials, are charged from the top of the furnace. The iron ore is reduced and melted inside the furnace, and molten iron and slag are discharged from the bottom. In the blast furnace process, it is crucial to maintain the temperature of the molten iron, which becomes the finished product, within a specified range. Furthermore, since the operation takes place with solid raw materials inside the furnace, it is necessary to ensure proper ventilation within the furnace to allow the raw materials to descend steadily. Consequently, it is essential to maintain the molten iron production rate required for the next step, steelmaking.

[0003] In recent years, in order to reduce CO2 emissions and manufacturing costs, it is required to reduce the ratio of the weight of iron in the iron ore charged from the top of the furnace to the weight of coke, i.e., the coke ratio. The coke ratio is defined by the weight of coke required to produce 1 ton of molten iron. In addition, coke acts as a reducing agent and a spacer to ensure aeration in the furnace. Furthermore, the production rate of molten iron is determined by the product of the rate at which the oxygen and steam supplied from the tuyeres consume the coke and the weight ratio of iron in the coke layer and the iron ore layer at the height of the tuyeres.

[0004] That is, the operation of the coke ratio affects the molten iron temperature, the aeration within the furnace, and the molten iron production rate. Therefore, the operation of the coke ratio is crucial for the stable operation of the blast furnace in order to properly control these control variables. The coke ratio is particularly important for controlling the aeration within the furnace. On the other hand, the operation of the coke ratio is highly dependent on the operator's knowledge or experience. Therefore, a new method is needed to predict the overall coke ratio operation amount without being influenced by the operator's knowledge or experience.

[0005] Against this background, Patent Document 1 proposes a method that uses machine learning to construct a model that mimics the operational actions of a skilled blast furnace operator, and uses the constructed model to control the molten iron temperature. Specifically, the method described in Patent Document 1 establishes a model that determines the operational actions used to control the molten iron temperature of the blast furnace near a target temperature through three phases: rising, falling, and observation. Furthermore, the method described in Patent Document 1 controls the molten iron temperature of the blast furnace based on operational data obtained by inputting trend data of factors considered when determining the operator's operational actions into the model. [Prior Art Documents] [Patent Documents]

[0006] Patent Document 1: Japanese Patent Application Publication No. 2021-18569 [Summary of the Invention]

[0007] [Problem to be Solved by the Invention] However, the method described in Patent Document 1 is difficult to guide the operation that precedes the operator's consideration of the operation. In particular, the operation that increases the coke ratio is mostly performed after an operational failure such as a sudden deterioration in the ventilation of the furnace or a sharp drop in the molten iron temperature. Therefore, from the viewpoint of preventing operational failures from occurring, the timing of the operator's operation to increase the coke ratio may not be appropriate. Therefore, it is necessary not only to imitate the timing of the operator's operation to increase the coke ratio, but also to detect the signs of operational failures in advance and guide the operation to increase the coke ratio ahead of the operator. However, the method described in Patent Document 1 cannot guide such operation of increasing the coke ratio.

[0008] Furthermore, generally speaking, the operation of the coke ratio is at risk of becoming an excessive operation, and therefore it is usually performed after the descent time of the raw materials in the blast furnace has been separated. However, the method described in Patent Document 1 does not take into account this descent time of the raw materials when performing machine learning. Therefore, according to the method described in Patent Document 1, the learning data used in the machine learning is unbalanced data with a small number of data points related to the operation of the coke ratio, and it is possible to construct a model that is difficult to predict the appropriate timing of the operation of the coke ratio.

[0009] This invention was made to solve the aforementioned problems, and its object is to provide a blast furnace coke ratio operation guidance method, a blast furnace coke ratio operation guidance system, a blast furnace coke ratio operation guidance program, and an information output device that can prompt the optimal coke ratio operation at the optimal time. Another object of this invention is to provide a blast furnace control device and a blast furnace operation method that can stably operate a blast furnace. Furthermore, yet another object of this invention is to provide a method for manufacturing molten iron that can produce molten iron with a good yield. [Means for Solving the Problems]

[0010] The coke ratio action guidance method for a blast furnace of the present invention includes: a model learning processing step, wherein image data obtained by image processing of historical data including observations of ventilation, tapping speed and molten iron temperature in the blast furnace process is used as input data, and historical data including labels representing the operation direction of the coke ratio assigned by the operator within a specified time range of the operation timing of the coke ratio is used as output data for machine learning, thereby generating a machine learning model that takes the image data during the guidance prompt period as input variables and the action prediction value representing the operation direction of the coke ratio during the guidance prompt period as output variables; an action prediction value calculation step, wherein the action prediction value during the guidance prompt period is calculated by inputting the image data during the guidance prompt period into the machine learning model generated in the model learning processing step; and an action guidance prompt step, wherein the operation action guidance of the coke ratio is prompted based on the action prediction value calculated in the action prediction value calculation step.

[0011] The historical data of the tag can be generated in the following manner: a tag indicating a decrease in the coke ratio is assigned within an α-hour range before and after the time when the operator lowers the coke ratio, and a tag indicating an increase in the coke ratio is assigned within a β-hour range before the time when the operator increases the coke ratio, with the time parameter β set to be greater than the time parameter α.

[0012] The value of the time parameter β may vary depending on the deviation of the molten iron temperature and tapping speed from the target value after a specified time following the operator's operation of the coke ratio.

[0013] It is possible that, during the period from when the operator operates on the coke ratio until a specified time has elapsed, if the coke ratio is operated in the opposite direction to the previous coke ratio operation, the previous coke ratio operation is considered an erroneous operation and the historical data related to the operation is excluded from the learning data used during machine learning.

[0014] If, after a specified time has elapsed since the operator operated the coke ratio, the molten iron temperature or tapping speed deviates from the target range, or the ventilation exceeds the management limit, the operation of the coke ratio is considered an incorrect operation and the historical data related to the operation is excluded from the learning data used during machine learning.

[0015] The coke ratio action guidance system for a blast furnace of the present invention may include: a model learning processing unit, which takes image data obtained by image processing of historical data including observations of ventilation, tapping speed and molten iron temperature in the blast furnace process as input data, and takes historical data including labels representing the operation direction of the coke ratio assigned by the operator within a specified time range for the timing of the operation of the coke ratio as output data for machine learning, thereby generating a machine learning model that takes the image data during the guidance prompt period as input variables and the action prediction value representing the operation direction of the coke ratio during the guidance prompt period as output variables; a coke ratio action prediction unit, which calculates the action prediction value during the guidance prompt period by inputting the image data during the guidance prompt period into the machine learning model generated by the model learning processing unit; and an action guidance prompt unit, which prompts the operation action of the coke ratio based on the action prediction value calculated by the coke ratio action prediction unit.

[0016] The action guidance prompting unit may, based on the action prediction value, prompt an operation action that increases the coke ratio, an operation action that maintains the coke ratio, or an operation action that decreases the coke ratio as a guide for the operation action of the coke ratio.

[0017] The action guidance prompting unit may prompt an operation to increase the coke ratio when the action prediction value exceeds the upper limit value, prompt an operation to decrease the coke ratio when the action prediction value is lower than the lower limit value, and prompt an operation to maintain the coke ratio when the action prediction value is below the upper limit value and above the lower limit value.

[0018] The action guidance prompt unit may include a setting component for the upper limit value and the lower limit value.

[0019] The coke ratio action guidance system of the blast furnace may include an output unit, which outputs the action prediction value calculated by the coke ratio action prediction unit.

[0020] The blast furnace control device of the present invention includes a component that controls the blast furnace based on the predicted action value output from the output unit of the blast furnace coke ratio action guidance system of the present invention.

[0021] The coke ratio action guidance program for the blast furnace of the present invention enables a computer to function as the following components: a model learning processing unit, which takes as input image data historical data including observations of ventilation, tapping speed and molten iron temperature in the blast furnace process, as image data, and as output historical data including labels representing the operation direction of the coke ratio assigned by the operator within a specified time range for the timing of the operation of the coke ratio, and performs machine learning to generate a machine learning model that takes the image data during the guidance prompt period as input variables and the action prediction value representing the operation direction of the coke ratio during the guidance prompt period as output variables; a coke ratio action prediction unit, which calculates the action prediction value during the guidance prompt period by inputting the image data during the guidance prompt period into the machine learning model generated by the model learning processing unit; and an action guidance prompt unit, which provides guidance on the operation action of the coke ratio based on the action prediction value calculated by the coke ratio action prediction unit.

[0022] The information output device of the present invention constitutes the coke ratio action guidance system of the blast furnace of the present invention. The information output device includes the action guidance prompting unit, which prompts the operation action of coke ratio based on the action prediction value output from the coke ratio action guidance system of the blast furnace. When the action prediction value exceeds the upper limit value, it prompts the operation action of increasing the coke ratio. When the action prediction value is lower than the lower limit value, it prompts the operation action of decreasing the coke ratio. When the action prediction value is lower than the upper limit value and higher than the lower limit value, it prompts the operation action of maintaining the coke ratio.

[0023] The action guidance prompt unit may include a setting component for the upper limit value and the lower limit value.

[0024] The blast furnace operation method of the present invention includes the following steps: operating the coke ratio based on the guidance of the operation action of the coke ratio prompted by the coke ratio operation guidance method of the blast furnace of the present invention.

[0025] The method for manufacturing molten iron according to the present invention includes the following steps: operating the coke ratio based on the operation of the coke ratio prompted by the coke ratio operation guidance method of the blast furnace according to the present invention, thereby manufacturing molten iron. [Effects of the Invention]

[0026] By means of the coke ratio action guidance method, coke ratio action guidance system, coke ratio action guidance program, and information output device of the present invention, the optimal coke ratio operation action can be prompted at the optimal time. Furthermore, by means of the blast furnace control device and blast furnace operation method of the present invention, the blast furnace can be operated efficiently and stably. Additionally, by means of the molten iron manufacturing method of the present invention, molten iron can be manufactured efficiently and with a good yield.

Implementation Method

[0028] Hereinafter, the structure and operation of a coke ratio guiding device for a blast furnace, which is an embodiment of the present invention, will be described with reference to the drawings.

[0029] [Structure] First, the structure of the coke ratio action guide device for a blast furnace, which is an embodiment of the present invention, will be described with reference to FIG1.

[0030] Figure 1 is a block diagram showing the structure of a coke ratio action guidance device for a blast furnace as an embodiment of the present invention. As shown in Figure 1, the coke ratio action guidance device 1 for a blast furnace as an embodiment of the present invention (hereinafter referred to as guidance device 1) includes an information processing device such as a workstation or personal computer. The guidance device 1 functions as a model learning processing unit 11, a coke ratio action prediction unit 12, and an action guidance prompting unit 13 by executing computer programs through the arithmetic processing unit inside the information processing device such as a central processing unit (CPU). The functions of each unit will be described later.

[0031] In addition, the blast furnace operation database (blast furnace operation DB) 2, which stores the blast furnace operation data, is connected to the guide device 1 in a form that allows data to be read. In this embodiment, the blast furnace operation DB 2 stores historical data of time series of observations including the ventilation index (ventilation degree) in the blast furnace, the tapping speed and the molten iron temperature, as well as historical operation data of time series based on the coke ratio of the operator. As the ventilation index in the blast furnace, the ventilation resistance index ΔP / V, represented by the following formula (1), can be exemplified. In formula (1), BP represents the blast pressure [Pa], TP represents the furnace top pressure [Pa], and BGV represents the Bosch gas volume [m3 (standard state) / min].

[0032] [Formula 1]

[0033] The guidance device 1 with this structure guides the operator to perform the optimal coke ratio operation by executing the model learning process and guidance prompting process shown below. Hereinafter, the operation of the guidance device 1 during the execution of the model learning process and guidance prompting process will be described with reference to the flowcharts shown in Figures 2 and 6.

[0034] [Model Learning Processing] First, referring to Figures 2 to 5, the operation of the guidance device 1 during model learning processing will be explained.

[0035] FIG2 is a flowchart showing the process of model learning processing as an embodiment of the present invention. The flowchart shown in FIG2 begins when the execution command for model learning processing is input to the guidance device 1, and the model learning processing enters the processing of step S1.

[0036] In step S1, the model learning processing unit 11 acquires historical time-series data of observations including ventilation parameters, tapping speed, and molten iron temperature from the blast furnace operation DB 2, as well as historical operation data based on the operator's coke ratio time series. These observations can be calculated based on sensor information (e.g., coke moisture content, burden line, furnace pressure, carbon powder ratio, blast pressure, ventilation parameters, and molten iron temperature) and target values ​​for ironmaking speed. Thus, step S1 is completed, and the model learning process proceeds to step S2.

[0037] In step S2, the model learning processing unit 11 assigns a label representing the operation direction of the coke ratio as a prediction object (output data) based on the time series operation history data based on the operator's coke ratio obtained in step S1, for each unit of time (e.g., every hour) of the operation history data. Hereinafter, the policy for assigning labels will be explained with reference to FIG3. FIG3 is a diagram showing an example of the operation history data based on the time series of the operator's coke ratio (top CR) (FIG3(a)) and the labels assigned to the operation history data (FIG3(b)).

[0038] Generally, increasing the coke ratio leads to an increase in CO2 emissions or a rise in the manufacturing cost of molten iron. Therefore, increasing the coke ratio is infrequent, usually occurring after an operational malfunction (such as a drop in tapping speed or molten iron temperature deviating from the managed range) or after the operator notices signs of an impending malfunction. That is, the operator sometimes increases the coke ratio later than the appropriate time. Therefore, in such cases, the coke ratio can be increased before the actual operator's intended time.

[0039] On the other hand, from the perspective of reducing CO2 emissions and the manufacturing cost of molten iron, the operation of lowering the coke ratio should be performed frequently at the optimal time, while taking into account concerns about the deterioration of furnace ventilation or the decrease in molten iron temperature. Therefore, regarding the operation of lowering the coke ratio, it is sufficient to learn (imitate) the timing when the operator actually lowers the coke ratio. Based on the above, in order to implement the operation of increasing the coke ratio at the appropriate time, it is particularly important to ensure that the timing of implementing the operation of increasing the coke ratio is faster than the timing when the operator actually increases the coke ratio.

[0040] Furthermore, based on the aforementioned reasons, the amount of each operation (change in coke ratio) that increases the coke ratio is greater than the amount of each operation that decreases the coke ratio. Additionally, the frequency of operations that decrease the coke ratio is higher than the frequency of operations that increase the coke ratio. Therefore, when performing machine learning, it is ideal to eliminate the imbalance in the number of data points between operations that increase the coke ratio and operations that decrease the coke ratio.

[0041] Therefore, as shown in Figures 3(a) and 3(b), the model learning processing unit 11 assigns a label indicating a decrease in the coke ratio to each unit of time in the operation history data within a time range of α hours before and after the operator actually lowers the coke ratio. Additionally, within a time range of β hours before the operator actually raises the coke ratio, the model learning processing unit 11 assigns a label indicating an increase in the coke ratio to each unit of time in the operation history data. Furthermore, in other time periods, the model learning processing unit 11 assigns a label indicating maintaining the coke ratio as "wait and see" to each unit of time in the operation history data.

[0042] The model learning processing unit 11 can set the value of the time parameter β to be greater than the value of the time parameter α, and increase the number of operation history data items labeled "coke ratio increase" so that the number of operation history data items labeled "coke ratio decrease" is approximately the same as the number of operation history data items labeled "coke ratio increase". Furthermore, the time parameters α and β can be set between 0 hours and 8 hours, taking into account the elapsed time from when the coke charged from the top of the furnace descends to the front of the tuyeres. Additionally, the time parameter β is preferably adjusted based on the deviation of the control indicators (tap iron speed and molten iron temperature) from the target value after a predetermined time T following the operator's operation of the coke ratio. Furthermore, the predetermined time T, taking into account the elapsed time from when the coke charged from the top of the furnace descends to the height position of the front of the tuyeres, can be set between 8 hours and 12 hours.

[0043] Furthermore, if the blast furnace control parameters (such as tapping speed and molten iron temperature) do not increase and deviate from the target value (such as the central value of the control parameter's management range) when the raw material layer (charge) that increases the coke ratio descends to the tuyeres height position after a specified time T, then the timing of increasing the coke ratio is inappropriate, and it is considered that the timing of increasing the coke ratio should be brought forward. On the other hand, if the control parameters recover to near the target value after a specified time T following the operation that increases the coke ratio, then the timing of the operation that increases the coke ratio is appropriate, and the "coke ratio increase" label should be assigned in accordance with the timing of the operator increasing the coke ratio.

[0044] Therefore, as shown in Figure 4, the model learning processing unit 11 can change the time parameter β in stages by increasing the coke ratio over a predetermined time T after the operation. The larger the deviation of the control index from the target value (such as the central value of each management range), the larger the value of the time parameter β. The value d on the horizontal axis of Figure 4 is the value after standardizing the deviation δ by dividing the deviation δ of the control index from the target value R by the target value R. The control index can be set as the tapping rate and molten iron temperature, which are management indicators for the reduction reaction of iron ore. In this case, the value d can be calculated for the tapping rate and molten iron temperature in each historical data, and the arithmetic mean or weighted average of the values ​​d calculated for each historical data can be used as the value of the horizontal axis. Thus, the processing of step S2 is completed, and the model learning processing proceeds to the processing of step S3.

[0045] In step S3, the model learning processing unit 11 generates learning data. The learning data is generated by associating input data with output data. The input data is set as the observations obtained in step S1, representing the ventilation index, tapping speed, and molten iron temperature in the blast furnace, which are based on the operator's coke ratio operation. Then, the model learning processing unit 11 associates the input data as image data for a predefined continuous time interval (from time t-γ to time t) and the output data as a label representing the operation direction of the coke ratio at time t. In order to unify the value range or dimension, the input data can be standardized before being used as image data. The image data is in the form of a two-dimensional image, which uses one axis as the time axis and the other axis as the sensor information and the target value of the ironmaking speed. For historical data of the input data, more than one input data is arranged in a direction different from the time axis, aligned with the time axis. In addition, if the number of data items labeled "coke ratio decreases", "coke ratio increases", and "wait and see" after step S2 is unbalanced, for the learning data with fewer data items that are output variables, oversampling techniques such as Synthetic Minority Oversampling Technique (SMOTE) can be performed to make the ratio of data items close to 1:1:1.

[0046] Furthermore, data on erroneous coke ratio operations performed by the operator should be excluded from the learning data to indicate the optimal coke ratio operation. As an example, Figures 5(a) and 5(b) show the time-varying furnace pressure when the operator's coke ratio operation is erroneous. In the situation shown in Figures 5(a) and 5(b), the operator excessively lowered the coke ratio at the 16-hour mark, causing a deterioration in the blast furnace's ventilation and an increase in furnace pressure. Therefore, the operator then increased the coke ratio at the 27-hour mark. Thus, if the coke ratio is operated in the opposite direction to the previous operation during the period from when the operator operates the coke ratio until a predetermined time τ has elapsed, it is preferable to consider the previous operation as an erroneous coke ratio operation and exclude it from the learning data. N, W, S, and E in Figure 5 indicate the orientation of the blast furnace.

[0047] Furthermore, the predetermined time τ for determining the operation in the opposite direction is preferably set between 8 and 12 hours, considering the elapsed time from when the coke charged from the top of the furnace descends to the front of the tuyeres. Additionally, if, during the period from when the operator operates the coke ratio until the predetermined time τ elapses, the molten iron temperature or tapping rate deviates from the target range, or the ventilation in the furnace exceeds the management limit, this can be considered an incorrect coke ratio operation and excluded from the learning data. Thus, step S3 is completed, and the model learning process proceeds to step S4.

[0048] In step S4, the model learning processing unit 11 uses the learning data generated in step S3 to generate a machine learning model that predicts the optimal coke ratio operation based on time-series data of sensor information and the target value of ironmaking speed. The model generated by machine learning can be a convolutional neural network (CNN). Furthermore, when the image data is a two-dimensional image with a time axis arranged on one axis and historical data arranged on the other, the convolution operation of the CNN can be performed only relative to the time axis direction. Thus, step S4 is completed, and the model learning process proceeds to step S5.

[0049] In step S5, the model learning processing unit 11 outputs the machine learning model generated in step S4. This machine learning model uses image data obtained by visualizing the temporal changes of observations during the guidance period (time t-γ to time t) as input variables, and the operation action of the coke ratio at time t as output variables. It predicts the optimal operation action for the coke ratio based on sensor information and the temporal changes of the target value of the ironmaking speed. Considering the descent time of the raw materials in the blast furnace and the coke replacement time in the furnace core, time γ can be set between 8 hours and 120 hours. Thus, step S5 is completed, and the series of model learning processes ends.

[0050] When new historical data is added to the blast furnace operation DB 2, the model learning processing unit 11 preferably updates the model by using the newly added sensor information and ironmaking speed as additional learning data to learn the coke ratio action prediction model.

[0051] [Guidance prompt processing] Next, the operation of the guidance device 1 when performing guidance prompt processing will be described with reference to FIG6.

[0052] FIG6 is a flowchart of the guidance prompting process as an embodiment of the present invention. The flowchart shown in FIG6 begins when the execution command for guidance prompting is input to the guidance device 1, and the guidance prompting process enters the processing of step S11.

[0053] In step S11, the action guidance prompting unit 13 acquires historical time series data from the blast furnace operation DB 2, including observations of the blast furnace ventilation index, tapping rate, and molten iron temperature during the guidance prompting period (time t0-γ to time t0, where time t0 is the current time). Thus, step S11 is completed, and the guidance prompting process proceeds to step S12.

[0054] In step S12, the action guidance prompting unit 13 inputs the historical data of the time series of observations obtained in step S11 into the machine learning model as input variables, and calculates the optimal operation amount (action prediction value P) for the current time t0 of the historical data of the time series of observations during the guidance prompting period (time t0-γ to time t0). Specifically, in the machine learning model generated by model learning, the probabilities of the operation actions of "coke ratio decrease", "coke ratio increase", and "wait and see" are output respectively. Therefore, if the probabilities of each are set as P1, P2, and P3 (P1+P2+P3=1), the action prediction value P of coke ratio is represented by P=P2-P1. Thus, the process of step S12 is completed, and the guidance prompting process proceeds to the process of step S13.

[0055] In step S13, the action guidance prompting unit 13 prompts the operator with the recommended coke ratio operation based on the action prediction value P calculated in step S12. Specifically, when the action prediction value P exceeds the upper limit value ε (P > ε), the action guidance prompting unit 13 prompts an operation to increase the coke ratio. Furthermore, when the action prediction value P is lower than the lower limit value (-η) (P < -η), the action guidance prompting unit 13 prompts an operation to decrease the coke ratio. Additionally, when the action prediction value P is below the upper limit value ε and above the lower limit value (-η), the action guidance prompting unit 13 prompts an operation to maintain the coke ratio. To avoid deterioration of the ventilation in the blast furnace or a decrease in the molten iron temperature, it is necessary to urgently perform an operation to increase the coke ratio. On the other hand, an operation to decrease the coke ratio carries the risk of deterioration of the ventilation in the blast furnace or a decrease in the molten iron temperature, therefore, it is necessary to provide careful prompts. Therefore, the upper limit value ε and the lower limit value (-η) can be set within the range of 0 < ε ≦ η < 1. Additionally, the motion guidance prompt unit 13 may also include setting components for the upper limit value ε and the lower limit value (-η). Examples of setting components include tuning screens or operator control panels that can set and change the upper limit value ε and the lower limit value (-η). With this, the processing in step S13 is completed, and the series of guidance prompt processing ends.

[0056] As clearly explained above, in the coke ratio action guidance device 1 for a blast furnace, which is an embodiment of the present invention, the model learning processing unit 11 takes image data obtained by image processing historical data including observations of ventilation, tapping speed and molten iron temperature in the blast furnace process as input data, and takes historical data including labels representing the operation direction of the coke ratio assigned by the operator within a specified time range for the timing of the operation of the coke ratio as output data for machine learning, thereby generating a machine learning model that takes image data during the guidance prompt period as input variables and action prediction values ​​representing the operation direction of the coke ratio during the guidance prompt period as output variables. Then, the coke ratio action prediction unit 12 calculates the action prediction values ​​during the guidance prompt period by inputting image data during the guidance prompt period into the generated machine learning model, and the action guidance prompt unit 13 provides guidance on the operation action of the coke ratio based on the calculated action prediction values.

[0057] Furthermore, according to this structure, since machine learning is performed using historical data representing the operation direction of the coke ratio, assigned within a predetermined time range including the operator's coke ratio operation timing, the optimal coke ratio operation action can be prompted at the optimal time. In addition, by operating the coke ratio based on the prompted coke ratio operation action, the blast furnace can be operated efficiently and stably. Furthermore, by operating the coke ratio based on the prompted coke ratio operation action to produce molten iron, molten iron can be produced efficiently and with a good yield. [Example]

[0058] In this embodiment, the time γ for setting the guidance prompt period is set to 24 hours. Sensor information acquired at 1-hour intervals and the target value of ironmaking speed are used, and features are extracted using CNN to predict the optimal coke ratio operation. When using 22,500 pieces (approximately 600 days' worth) of sensor information and the target value of ironmaking speed to correlate input and output data, the ratio of data points for "coke ratio decrease," "coke ratio increase," and "wait and see" becomes unbalanced at 2:1:20. SMOTE is applied to the remaining 22,000 pieces (excluding the 500 pieces used for model accuracy verification) to generate 57,500 pieces of learning data with a ratio of "coke ratio decrease," "coke ratio increase," and "wait and see" close to 1:1:1. Subsequently, machine learning of the CNN-based model is performed using the initial 57,000 points of learning data, and the accuracy of the model is verified using the remaining 500 points of learning data.

[0059] The structure of the CNN used is shown in Figure 7. The blocks in Figure 7 represent convolutional layers, pooling layers, pooling layers, fully connected layers (affine), dropout layers, fully connected layers (affine), dropout layers, and fully connected layers (affine). Additionally, the numbers in parentheses in Figure 7 indicate the size of the data arrangement. For example, (13, 24, 64) represents a three-dimensional arrangement of 13 vertical (variables) × 24 horizontal (time) × 64 channels. The CNN structure shown in Figure 7 is an example; other structures besides those shown in Figure 7 can also be used.

[0060] The verification results of this embodiment are shown in Figure 8. Figure 8 is as follows: The total number of cases where the model predicts "coke ratio decrease" or "coke ratio increase" using the same verification data and the following three methods of machine learning is set as N, and the total number of cases where the molten iron temperature and tapping speed are within the target range and the ventilation in the blast furnace is less than the management limit is set as N1, is N1 / N as the accuracy rate and compared. The first method (Example 1) is to use the following learning data, that is, the operator's misoperation of coke ratio is also included in the learning data, and α and β are both set to 0 hours and the label is only assigned when the operator actually operates on the coke ratio, to perform machine learning on the model.

[0061] The second method (Example 2) involves excluding learning data on operator errors in coke ratio operation and performing machine learning. The third method (Example 3) involves excluding learning data on operator errors in coke ratio operation and using learning data where α is set to 1 hour, and β varies between 1 hour and 8 hours based on the deviation of the molten iron temperature and tapping speed from the target value after a specified time T=8 hours. This results in machine learning of the model using learning data with expanded labels for operator coke ratio operation timing. As shown in Figure 8, the accuracy increases in the order of Examples 1, 2, and 3. This confirms that, according to the present invention, appropriate coke ratio operation can be predicted.

[0062] Furthermore, when the model predicts an increase in the coke ratio, and the operator's actual action is "wait and see," as shown in Figures 9(a) and 9(b), starting from 2 to 3 hours before the time indicated by the dashed line predicting the increase in the coke ratio, the increase in the ventilation index or air supply pressure included in the input variables can indicate the deterioration of ventilation within the blast furnace. This confirms that, according to the present invention, it is possible to predict and guide operational actions regarding the coke ratio that override the operator's control.

[0063] The embodiments of the invention made by the inventors have been described above, but the present invention is not limited to the descriptions and drawings that form part of the disclosure of the present invention based on this embodiment. For example, the image data can be set as a two-dimensional image with a time axis arranged on one axis and a range of values ​​that the variable can take arranged on the other axis, according to each item of the input variable, and multiple image data can be associated with labels for coke ratio operation to perform CNN convolution operations in the time axis direction and the variable axis direction. In addition, the labels may not be the three categories of "coke ratio decrease", "coke ratio increase" and "wait and see", but may be the operation quantity of coke ratio itself, or a model for predicting the operation quantity of coke ratio.

[0064] Alternatively, a coke ratio action guidance device for a blast furnace, as an embodiment of the present invention, can be used to construct a coke ratio action guidance system for a blast furnace. FIG10 is a block diagram showing a structural example of a coke ratio action guidance system for a blast furnace. The coke ratio action guidance system 10 for a blast furnace shown in FIG10 includes the blast furnace operation DB2, information processing device 20, and information output device 30 shown in FIG1. ​​The blast furnace operation DB2, information processing device 20, and information output device 30 are configured, for example, to communicate via a network N such as the Internet or a network cable. The information output device 30 in the coke ratio action guidance system is, for example, located in an ironmaking plant.

[0065] The information processing device 20 includes a model learning processing unit 11, a coke ratio action prediction unit 12, and an output unit 21, as shown in FIG1. ​​The output unit 21 sends the action prediction value P, which is the output of the coke ratio action prediction unit 12, to the information output device 30 or other information processing device via the network N. Other information processing devices that send the action prediction value P from the output unit 23 may include, for example, a blast furnace control computer. The control computer can also control the coke ratio of the blast furnace based on the sent action prediction value P. In addition, the control computer can also control the coke ratio of the blast furnace based on operator corrections.

[0066] The information output device 30 includes an output information acquisition unit 31 and an action guidance prompting unit 13 shown in FIG1. ​​The output information acquisition unit 31 acquires an action prediction value P from the information processing device 20, and the action guidance prompting unit 13 prompts the operation action of the coke ratio based on the action prediction value P. The action guidance prompting unit 13 may also include a tuning screen, which can display and set the judgment threshold values ​​(upper limit value ε, lower limit value -η) of the action prediction value when prompting the operation action of the coke ratio based on the action prediction value.

[0067] The information processing device 20 and the information output device 30 may not include all the constituent elements shown in FIG10. Furthermore, the information processing device 20 and the information output device 30 may include constituent elements other than those shown in FIG10. Additionally, the constituent elements included in the information processing device 20 and the information output device 30 are not limited to the example shown in FIG10. For example, a portion of the constituent elements included in the information processing device 20 may be disposed on the information output device 30 side. Similarly, a portion of the constituent elements included in the information output device 30 may be disposed on the information processing device 20 side. Therefore, the coke ratio action guidance system can be configured as a whole, including the model learning processing unit 11, the coke ratio action prediction unit 12, and the action guidance prompt unit 13, using the information processing device 20 and the information output device 30.

[0068] Thus, all other embodiments, examples, and application technologies developed by those skilled in the art based on this embodiment are included within the scope of this invention. [Industrial Applicability]

[0069] According to the present invention, a method for guiding the operation of a blast furnace to achieve the optimal coke ratio, a system for guiding the operation of a blast furnace to achieve the optimal coke ratio at the optimal time, a program for guiding the operation of a blast furnace to achieve the optimal coke ratio, and an information output device can be provided. Furthermore, according to the present invention, a blast furnace control device and a method for operating a blast furnace can be provided to stably operate the blast furnace. Additionally, according to the present invention, a method for manufacturing molten iron that can produce molten iron with a good yield can be provided. [Simplified Explanation of the Diagram]

[0027] Figure 1 is a block diagram showing the structure of a coke ratio action guidance device for a blast furnace according to an embodiment of the present invention. Figure 2 is a flowchart showing the process of model learning processing according to an embodiment of the present invention. Figure 3 is a diagram showing an example of operation history data based on the operator's coke ratio time series and the labels assigned to the operation history data. Figure 4 is a diagram showing the relationship between the deviation of the control index relative to the target value and the time parameter β. Figure 5 is a diagram showing the time variation of the furnace pressure when the operator's coke ratio operation is incorrect. Figure 6 is a flowchart showing the process of guidance prompt processing according to an embodiment of the present invention. Figure 7 is a diagram showing the machine learning model of the embodiment. Figure 8 is a diagram showing the accuracy of the embodiment. Figure 9 is a diagram showing the time variation of the ventilation index and the air supply pressure. Figure 10 is a block diagram showing the structure of a coke ratio action guidance system according to an embodiment of the present invention.

Claims

1. A method for guiding the operation of coke ratio in a blast furnace, comprising: a model learning processing step, wherein image data obtained by visualizing historical data including observations of ventilation, tapping speed and molten iron temperature in the blast furnace process is used as input data, and historical data including labels representing the operation direction of coke ratio assigned by the operator within a specified time range for the timing of coke ratio operation is used as output data for machine learning, thereby generating a machine learning model that takes the image data during the guidance prompt period as input variables and the predicted value of the operation direction of coke ratio during the guidance prompt period as output variables; an operation prediction value calculation step, wherein the predicted value of the operation during the guidance prompt period is calculated by inputting the image data during the guidance prompt period into the machine learning model generated in the model learning processing step; and an operation guidance prompt step, wherein the operation of coke ratio is guided based on the operation prediction value calculated in the operation prediction value calculation step.

2. The method for guiding the coke ratio action of a blast furnace as described in claim 1, wherein, The historical data of the tags is generated in the following manner: tags indicating a decrease in the coke ratio are assigned within an α-hour range before and after the time when the operator lowers the coke ratio, and tags indicating an increase in the coke ratio are assigned within a β-hour range before the time when the operator increases the coke ratio, with the time parameter β set to be greater than the time parameter α.

3. The method for guiding the coke ratio action of a blast furnace as described in claim 2, wherein, The value of the time parameter β changes based on the deviation of the molten iron temperature and tapping speed from the target value after a specified time following the operator's operation of the coke ratio.

4. The method for guiding the coke ratio action of a blast furnace as described in any one of claims 1 to 3, wherein, If, during the period from when the operator operates on the coke ratio until a specified time has elapsed, the coke ratio is operated in the opposite direction to the previous operation, the previous operation on the coke ratio is considered an erroneous operation, and the historical data related to the operation is excluded from the learning data used during machine learning.

5. The method for guiding the coke ratio action of a blast furnace as described in any one of claims 1 to 3, wherein, If, after a specified time has elapsed since the operator manipulated the coke ratio, the molten iron temperature or tapping speed deviates from the target range, or the ventilation exceeds the management limit, the operation of the coke ratio is considered an erroneous action, and the historical data related to the operation will be excluded from the learning data used during machine learning.

6. A coke ratio control system for a blast furnace, comprising: The model learning processing unit takes image data obtained by visualizing historical data including observations of ventilation, tapping speed, and molten iron temperature in the blast furnace process as input data, and takes historical data including labels representing the operation direction of the coke ratio assigned by the operator within a specified time range for the timing of the operation of the coke ratio as output data for machine learning, thereby generating a machine learning model that takes the image data during the guidance prompt period as input variables and the predicted value of the operation direction of the coke ratio during the guidance prompt period as output variables; the coke ratio action prediction unit calculates the predicted value of the action during the guidance prompt period by inputting the image data during the guidance prompt period into the machine learning model generated by the model learning processing unit; And an action guidance prompting unit, which provides guidance on the operation action of the coke ratio based on the action prediction value calculated by the coke ratio action prediction unit.

7. The coke ratio action guidance system for a blast furnace as described in claim 6, wherein, The action guidance prompting unit provides prompts based on the action prediction value, suggesting actions to increase the coke ratio, maintain the coke ratio, or decrease the coke ratio as guidance for the operation actions related to the coke ratio.

8. The coke ratio action guidance system for a blast furnace as described in claim 7, wherein, The action guidance prompt unit prompts an action to increase the coke ratio when the action prediction value exceeds the upper limit, prompts an action to decrease the coke ratio when the action prediction value is below the lower limit, and prompts an action to maintain the coke ratio when the action prediction value is below the upper limit and above the lower limit.

9. The coke ratio action guidance system for a blast furnace as described in claim 8, wherein, The action guidance prompt unit includes a setting component for the upper limit value and the lower limit value.

10. The coke ratio action guidance system for a blast furnace as claimed in any one of claims 6 to 9, comprising an output unit that outputs the action prediction value calculated by the coke ratio action prediction unit.

11. A blast furnace control device, comprising a component for controlling the blast furnace based on the predicted action value output by the output unit of the blast furnace coke ratio action guidance system of claim 10.

12. A coke ratio action guidance program for a blast furnace, comprising a computer functioning as follows: a model learning processing unit, which takes as input image data historical data including observations of ventilation, tapping speed, and molten iron temperature in the blast furnace process, and as output historical data containing labels representing the operation direction of the coke ratio assigned by the operator within a predetermined time range for the timing of the coke ratio operation, and performs machine learning to generate a machine learning model that takes the image data during the guidance prompt period as input variables and a predicted value representing the operation direction of the coke ratio during the guidance prompt period as output variables; a coke ratio action prediction unit, which calculates the predicted value of the action during the guidance prompt period by inputting the image data during the guidance prompt period into the machine learning model generated by the model learning processing unit; and an action guidance prompt unit, which provides guidance on the operation action of the coke ratio based on the predicted value calculated by the coke ratio action prediction unit.

13. An information output device comprising a coke ratio action guidance system for a blast furnace as described in claim 6, the information output device including the action guidance prompting unit, which prompts guidance on the operation action of the coke ratio based on the action prediction value output from the coke ratio action guidance system of the blast furnace; prompts an operation action to increase the coke ratio when the action prediction value exceeds an upper limit value; prompts an operation action to decrease the coke ratio when the action prediction value is below a lower limit value; and prompts an operation action to maintain the coke ratio when the action prediction value is below the upper limit value and above the lower limit value.

14. The information output device as claimed in claim 13, wherein, The action guidance prompt unit includes a setting component for the upper limit value and the lower limit value.

15. A method of operating a blast furnace, comprising the step of: operating the coke ratio based on guidance of the operation of the coke ratio prompted by the coke ratio operation guidance method of the blast furnace as described in any one of claims 1 to 5.

16. A method for manufacturing molten iron, comprising the steps of: operating the coke ratio based on the guidance of the operation of the coke ratio indicated by the coke ratio operation guidance method of the blast furnace as described in any one of claims 1 to 5, thereby manufacturing molten iron.

Images