molten iron temperature control method, molten iron temperature control device, operation guidance method, blast furnace operation method and molten iron production method
Patent Information
- Application Number
- KR1020267024250
- Authority / Receiving Office
- KR · KR
- Patent Type
- Applications
- Current Assignee / Owner
- Priority Date
- 2024-03-14
- Filing Date
- 2025-01-21
- Publication Date
- 2026-08-14
Smart Images
Figure P1020267024250_ABST
Abstract
Description
Technology Field
[0001] The present disclosure relates to a method for controlling molten iron temperature, a molten iron temperature control device, an operation guidance method, a method for operating a blast furnace, and a method for producing molten iron. Background Technology
[0002] In a blast furnace, iron ore and coke are fed from the top of the furnace. As the fed iron ore and coke descend toward the bottom of the furnace, they are heated by hot air blown in from the bottom of the furnace and pulverized coal. During this process, the iron oxide in the iron ore is reduced by the carbon contained in the coke and pulverized coal, and is discharged as molten iron at approximately 1500°C.
[0003] Blast furnaces operate continuously without interruption, and once shut down due to operational troubles, it takes a long time to restart. Therefore, stable operation of the blast furnace is required. Meanwhile, in recent blast furnace operations, a reduction in the amount of reducing agents (coke and pulverized coal) is required to lower CO2 emissions and manufacturing costs. However, due to the demand for reduced reducing agent usage, stable operation of the blast furnace can sometimes be difficult. Furthermore, with the progression of low birth rates and an aging population, it is expected that securing operators will become difficult in the future. Consequently, high-efficiency and stable blast furnace operation through process automation is desired.
[0004] Controlling the molten iron temperature is particularly important for realizing high-efficiency and stable blast furnace operations. If the molten iron temperature drops drastically, the molten iron or byproduct slag may solidify inside the furnace, potentially leading to prolonged operational shutdowns. Conversely, setting a high target temperature to avoid a drop in temperature results in the excessive consumption of reducing agents. By suppressing fluctuations in the molten iron temperature, it becomes possible to lower the target temperature only after maintaining it above a lower limit. Consequently, suppressing temperature variations reduces the amount of reducing agent used, thereby achieving the effects of reducing CO2 emissions and cutting costs.
[0005] Against this background, a method for controlling molten iron temperature using a physical model (see Patent Document 1) and a method for controlling molten iron temperature using a statistical model (see Patent Document 2) have been proposed. For example, Patent Document 1 discloses a method for controlling molten iron temperature based on a predicted value to compensate for the error between the calculated value and the actual value of at least one of the reducing agent ratio, the amount of carbon lost from the solution, the shipbuilding speed, and the gas utilization rate. The technology of Patent Document 1 calculates a predicted value of molten iron temperature using a physical model, assuming that the current amount of operation is maintained while adjusting the gas reduction equilibrium parameter or the coke ratio parameter in the furnace top.
[0006] Patent Document 2 discloses a method for predicting a furnace heat evaluation index by using information regarding operating conditions within a performance data set to create a prediction formula for a furnace heat evaluation index that represents the relationship between the operating conditions of a blast furnace and the furnace heat evaluation index, and by inputting the operating conditions of the target for prediction. The technology of Patent Document 2 determines the parameters of the prediction formula by solving an optimization problem using a function that has similarity to the operating conditions of the target for prediction as a weight, as an evaluation function that evaluates the prediction error of the prediction formula.
[0007] In addition, Patent Document 3 discloses a molten iron temperature control method in which an operator determines an operation action using a one-dimensional convolutional neural network, and uses the constructed operation model to determine an operation action to set the molten iron temperature of a blast furnace to a target temperature. The operation model has at least one of molten iron temperature, water hole filling temperature, and coke ratio as an observed quantity (input data), and at least one of pulverized coal ratio, blowing moisture, and coke ratio as an operated quantity (output data). Prior art literature
[0008] Japanese Published Patent Application No. 2018-024935, Japanese Published Patent Application No. 2015-117431, Japanese Published Patent Application No. 2021-018569 The problem to be solved
[0009] The method described in Patent Document 1 can predict changes in molten iron temperature after a long period of time (e.g., 4 to 12 hours). However, it is difficult to predict sudden changes in molten iron temperature accompanied by hanging, slip, and furnace abnormalities caused by sudden deterioration of permeability using the method described in Patent Document 1. The method described in Patent Document 2 can input various observational information, such as the results of reactions within the furnace. Therefore, the method described in Patent Document 2 can predict sudden changes in molten iron temperature that are difficult to predict with physical models, but the prediction range is limited to a short period of time (e.g., 1 to 4 hours). The method described in Patent Document 3 makes it possible to learn and mimic past exemplary operational actions performed by an operator based on experience, even under conditions where the prediction accuracy of physical or statistical models is reduced. However, relying on training data used for machine learning, the prediction accuracy of the machine learning model generated decreases when there is little data similar to the situation to be addressed. As such, since each model has its advantages and disadvantages, using a single model may result in a decrease in prediction accuracy and may lead to a failure to suggest appropriate operational actions.
[0010] In addition, to control the molten iron temperature, the pulverized coal ratio, blower moisture content, and blower temperature are manipulated as operational actions. The pulverized coal ratio is the amount of pulverized coal per ton of molten iron. Blower moisture content is the amount of moisture added to the hot air. Blower temperature is the temperature of the hot air. Recently, to reduce the reduction agent ratio, the blower temperature is manipulated to remain constant near the upper limit. Furthermore, it is known that manipulating the blower moisture content is highly effective as a control of the molten iron temperature. Here, the decomposition of water vapor is an endothermic reaction. Therefore, by making the blower moisture content greater than the atmospheric moisture content, the amount of reducing agent required to maintain a constant molten iron temperature increases. Conversely, by lowering the blower moisture content to be equal to the atmospheric moisture content and manipulating the pulverized coal ratio, the amount of reducing agent required to maintain a constant molten iron temperature can be reduced. However, when the high-temperature coke burning at the tip of the nozzle is replaced with pulverized coal at room temperature, the gas temperature at the tip of the nozzle decreases, so the molten iron temperature rises after about 5 hours. In other words, the immediate effect of adjusting the pulverized coal ratio is lower compared to adjusting the air-drying moisture ratio. Therefore, in order to achieve both a reduction in the reducing agent ratio and high-precision control of the molten iron temperature, it is necessary to appropriately distinguish and use the operating variables of the operation action.
[0011] The objective of the present disclosure, made in consideration of these circumstances, is to provide a molten iron temperature control method, a molten iron temperature control device, an operation guidance method, a method for operating a blast furnace, and a method for producing molten iron that can achieve both a reduction in the reducing agent ratio and high-precision molten iron temperature control. means of solving the problem
[0012] (1) A molten iron temperature control method related to one embodiment of the present disclosure is,
[0013] A model calculation step for calculating an action determination value of an operation action for molten iron temperature or molten iron temperature control using multiple models that take blast furnace operation data as input, and
[0014] It includes an operation action determination step that uses the calculated molten iron temperature or the action determination value, selects one of the plurality of models based on a determination criterion based on the trend of the prediction accuracy of the plurality of models, and determines the operation variable and the amount of operation as the operation action.
[0015] (2) As an embodiment of the present disclosure, in (1),
[0016] The above plurality of models are,
[0017] A physical model capable of representing transient states through calculations of heat transfer, chemical reactions, and fluid flow inside the blast furnace, and
[0018] It includes a statistical model that predicts the molten iron temperature by extracting similar operating conditions based on past operating data of the above-mentioned blast furnace.
[0019] (3) As an embodiment of the present disclosure, in (1) or (2),
[0020] The above plurality of models are,
[0021] It further includes a machine learning model generated by machine learning using training data that correlates past operation data of the blast furnace with exemplary operation actions performed by an operator, which takes the operation data of the blast furnace as input and outputs an operation action determined to be appropriate.
[0022] (4) As an embodiment of the present disclosure, in any one of (1) to (3),
[0023] The above-mentioned operation action determination step determines the operation variables and the amount of operation as the operation action based on the predicted molten iron temperature after a long time calculated using the physical model, the predicted molten iron temperature after a short time calculated using the statistical model, the action judgment value calculated using the machine learning model, the control range of the blowing moisture determined by considering the reduction of the reducing agent ratio, and the restriction on the use of pulverized coal based on operation constraints.
[0024] (5) As an embodiment of the present disclosure, in any one of (1) to (4),
[0025] The above operation action determination step determines the operation variables and the amount of operation as the operation action based on the action judgment value when the predicted molten iron temperature after the above long time and the predicted molten iron temperature after the above short time are both within the target range.
[0026] (6) As an embodiment of the present disclosure, in any one of (1) to (5),
[0027] The above operation action determination step determines the operation variables and amounts as the operation action based on the predicted molten iron temperature after the short time, when both the predicted molten iron temperature after the long time and the predicted molten iron temperature after the short time deviate from the target range and the corresponding operation actions match, or when the predicted molten iron temperature after the long time is within the target range and the predicted molten iron temperature after the short time deviates from the target range.
[0028] (7) As an embodiment of the present disclosure, in any one of (1) to (6),
[0029] The above operation action determination step is such that when the predicted molten iron temperature after a long time deviates from the target range and the respective operation actions corresponding to the predicted molten iron temperature after a long time and the predicted molten iron temperature after a short time do not match, and when the operation action corresponding to the predicted molten iron temperature after a long time and the operation action corresponding to the action judgment value are at least not contradictory, the operation amount and operation amount are determined as operation actions based on the predicted molten iron temperature after a long time, and when the operation action corresponding to the predicted molten iron temperature after a long time and the operation action corresponding to the action judgment value do not match, an active operation action is not executed.
[0030] (8) As an embodiment of the present disclosure, in any one of (1) to (7),
[0031] The above operating constraints include at least one of the upper limit of the gas temperature at the tip of the ox, the lower limit of the gas temperature at the tip of the ox, the upper limit of the reducing agent ratio, the lower limit of the reducing agent ratio, and the lower limit of the gas temperature at the top of the furnace.
[0032] (9) As an embodiment of the present disclosure, in any one of (1) to (8),
[0033] The above-mentioned range of airflow moisture control is determined by adding a value that allows for at least one reduction of the amount of airflow moisture, which is determined as a single operation amount, with respect to atmospheric moisture.
[0034] (10) A molten metal temperature control device related to one embodiment of the present disclosure is,
[0035] A model calculation unit that calculates action determination values for molten iron temperature or operation actions for molten iron temperature control using multiple models that take blast furnace operation data as input, and
[0036] The operating action determining unit is provided, which uses the calculated molten iron temperature or the action determination value, selects one of the plurality of models based on a determination criterion based on the trend of the prediction accuracy of the plurality of models, and determines the operating variable and the amount of operation as the operating action.
[0037] (11) As one embodiment of the present disclosure, in (10),
[0038] The above plurality of models are,
[0039] A physical model capable of representing transient states through calculations of heat transfer, chemical reactions, and fluid flow inside the blast furnace, and
[0040] It includes a statistical model that predicts the molten iron temperature by extracting similar operating conditions based on past operating data of the above-mentioned blast furnace.
[0041] (12) As an embodiment of the present disclosure, in (10) or (11),
[0042] The above plurality of models are,
[0043] It further includes a machine learning model generated by machine learning using training data that correlates past operation data of the blast furnace with exemplary operation actions performed by an operator, which takes the operation data of the blast furnace as input and outputs an operation action determined to be appropriate.
[0044] (13) As an embodiment of the present disclosure, in any one of (10) to (12),
[0045] The above-mentioned operation action determination unit determines the operation variables and the amount of operation as the operation action based on the predicted molten iron temperature after a long time calculated using the physical model, the predicted molten iron temperature after a short time calculated using the statistical model, the action judgment value calculated using the machine learning model, the management range of the blowing moisture determined by considering the reduction of the reducing agent ratio, and the restriction on the use of pulverized coal based on operation constraints.
[0046] (14) An operation guidance method related to one embodiment of the present disclosure is,
[0047] The above-mentioned operation action determined by any one of the molten iron temperature control methods (1) to (9) is presented so that the operator of the blast furnace can verify it.
[0048] (15) A method of operating a blast furnace related to one embodiment of the present disclosure is,
[0049] The method includes a step of controlling the blast furnace according to the operation action determined by the molten iron temperature control method described in any one of (1) to (9).
[0050] (16) A method for manufacturing molten iron related to one embodiment of the present disclosure is,
[0051] (15) A step of controlling the blast furnace according to the method of operating the blast furnace to produce molten iron is included. Effects of the invention
[0052] According to the present disclosure, a molten iron temperature control method, a molten iron temperature control device, an operation guidance method, a method for operating a blast furnace, and a method for producing molten iron can be provided, which are capable of achieving both a reduction in the reducing agent ratio and high-precision molten iron temperature control. Brief explanation of the drawing
[0053] FIG. 1 is a block diagram showing the configuration of a molten iron temperature control device related to one embodiment of the present disclosure. Figure 2A is a flowchart showing the flow of processing of the operation action decision unit. Figure 2B is a flowchart showing the flow of processing of the operation action decision unit. FIG. 3 is a diagram showing changes in molten iron temperature, coke ratio, pulverized coal ratio, blowing moisture, and atmospheric moisture during a blast furnace operation method using a molten iron temperature control device related to one embodiment of the present disclosure for 4 days. FIG. 4 is a drawing showing a comparison of the molten iron temperature deviation and the reducing agent ratio in the case where a method of operating a blast furnace using a molten iron temperature control device related to one embodiment of the present disclosure is carried out and a conventional method of operating by an operator. FIG. 5 is a drawing showing an example of a system configuration having a molten iron temperature control device related to one embodiment of the present disclosure. Specific details for implementing the invention
[0054] Hereinafter, with reference to the drawings, a molten iron temperature control method, a molten iron temperature control device (10) (see FIG. 1), an operation guidance method, a method for operating a blast furnace, and a method for manufacturing molten iron related to one embodiment of the present disclosure are described.
[0055] (Device configuration)
[0056] FIG. 1 is a block diagram showing the configuration of a molten metal temperature control device (10) according to one embodiment of the present disclosure. The molten metal temperature control device (10) according to one embodiment of the present disclosure is configured by an information processing device such as a computer. When the molten metal temperature control device (10) is a computer, a computational processing device such as a CPU (Central Processing Unit) functions as a model calculation unit (11) and an operation action determination unit (12) by executing a program. The processing executed by the model calculation unit (11) may be referred to as a model calculation step. Also, the processing executed by the operation action determination unit (12) may be referred to as an operation action determination step. The functions of each unit will be described later.
[0057] An operation database (20) is connected to the molten iron temperature control device (10) in a data-readable form. In this embodiment, the operation database (20) stores operation factors such as furnace top coke ratio, blower flow rate, enriched oxygen amount, blower temperature, blower moisture, and pulverized coal flow rate. Additionally, the operation database (20) stores actual values of process variables calculated based on the volume fraction of CO and CO2 of the discharged furnace top gas. Furthermore, the operation database (20) stores historical data of temperature and pressure measured at least one of the furnace body, furnace top, and furnace bottom. Here, process variables may include molten iron temperature, shipbuilding speed, solution loss carbon amount, gas utilization rate, etc.
[0058] The molten iron temperature control device (10) executes each process of the molten iron temperature control method by the model calculation unit (11) and the operation action determination unit (12), specifies the operation variables for molten iron temperature control, and outputs the optimal action as guidance.
[0059] (Operation of the model calculation unit (11))
[0060] The model calculation unit (11) calculates an action determination value for the molten iron temperature or an operation action for controlling the molten iron temperature by using a plurality of models that take the blast furnace operation data as input. The action determination value will be described later. In this embodiment, the model calculation unit (11) executes a process using a physical model, a statistical model, and a machine learning model at the timing when an execution command for model calculation is input to the molten iron temperature control device (10). In this embodiment, three models are used, but the plurality of models are not limited to three models and may be, for example, two or four or more. The plurality of models used by the model calculation unit (11) are created in advance and stored in a memory device accessible by the molten iron temperature control device (10). Before executing the calculation process, the model calculation unit (11) reads the models from the memory device.
[0061] The physical model is a model capable of expressing transient states through the calculation of heat transfer, chemical reactions, and fluid flow inside the blast furnace. The fluid includes not only gases but also solids. The physical model in this embodiment is the same model as the method described in Reference 1 (Michiharu Hatano et al., “Review of ignition operation using a blast furnace unsteady model,” Iron and Steel, vol. 68, p. 2369). That is, the physical model is composed of a group of partial differential equations that consider multiple physical phenomena such as the reduction of iron ore, heat exchange between iron ore and coke, and the melting of iron ore, and it is possible to calculate variables (output variables) representing the state inside the blast furnace in an unsteady state. Since the physical model can predict the reaction itself occurring inside the furnace, long-term predictions are possible for a long period (4 to 12 hours) in advance, including cases where operating variables are changed immediately. Here, the long period (4 to 12 hours) in the prediction of the physical model takes into account the time for the raw materials charged from the top of the blast furnace to descend.
[0062] The statistical model is a model that predicts molten iron temperature by extracting similar operating conditions based on past operating data of the blast furnace. The statistical model in this embodiment is a model that uses the Just-In-Time method described in Reference 2 (Shigeru Yamamoto, “Just-In-Time Predictive Control: Predictive Control Based on Accumulated Data,” Measurement and Control, Vol. 52, No. 10, p. 878). The statistical model is a model that predicts molten iron temperature after a short period (1 to 4 hours) by using measured values of temperature and pressure at the furnace body, furnace top, and furnace bottom of the blast furnace, as well as actual values of process variables, as explanatory variables. The statistical model sequentially constructs a local linear function using data near the explanatory variables at the current time. Correlating the set of explanatory variables at the current time with the molten iron temperature after 1 to 4 hours is intended to take into account the residence time of the molten iron at the bottom of the furnace.
[0063] The machine learning model is generated by machine learning using training data that correlates past operation data of the blast furnace with exemplary operation actions performed by an operator. The generated machine learning model takes the blast furnace operation data as input and outputs an operation action determined to be appropriate. The machine learning model may use a one-dimensional convolutional neural network method, for example, similar to the method described in Patent Document 3. The machine learning model in this embodiment outputs a value obtained by subtracting the determination probability of a molten iron temperature decrease action from the determination probability of a molten iron temperature increase action (hereinafter referred to as the action determination value). Here, the molten iron temperature increase action is an operation action intended to raise the molten iron temperature. Also, the molten iron temperature decrease action is an operation action intended to lower the molten iron temperature. In short, the machine learning model in this embodiment includes the action determination value as the operation action determined to be appropriate to output. In addition, in machine learning using a one-dimensional convolutional neural network of a machine learning model, learning may be performed that includes not only the direction of action (increase or decrease of pulverized coal ratio, blowing moisture, etc.) but also the manipulated quantity (amount to increase or decrease).
[0064] Here, in order to control the molten iron temperature, the following three factors must be considered when determining the operational variables of the operation action. The first is the relationship between the model selection and the time delay in which changes to the operational variables are reflected in the molten iron temperature. As mentioned above, when comparing the pulverized coal ratio and the blowing moisture content, the blowing moisture content has a higher immediate effect (in other words, good responsiveness). Therefore, if it is predicted based on a statistical model that the molten iron temperature will rise or fall rapidly in the short term, the operation of the blowing moisture content is prioritized. If it is predicted based on a physical model that the molten iron temperature will deviate from the target in the long term, the operation of either the pulverized coal ratio or the blowing moisture content is selected. A machine learning model is used to execute an appropriate operation action in advance when it is determined that no action is necessary regardless of whether the physical or statistical model is used, or when the prediction accuracy of the physical or statistical model deteriorates due to the influence of disturbances.
[0065] Physical and statistical models may experience a significant decrease in prediction accuracy when there are large, difficult-to-measure disturbances. Consequently, operational actions corresponding to the molten iron temperature after a long period based on the physical model and the molten iron temperature after a short period based on the statistical model may conflict. In such cases, it becomes difficult to determine which operational action to execute or to judge the timing for switching between actions. Furthermore, if the molten iron temperature after a long period based on the physical model is predicted to deviate from the target range, but the molten iron temperature after a short period based on the statistical model is expected to remain within the target range, it becomes difficult to determine the timing for executing the operational action corresponding to the long-term molten iron temperature. However, even in such cases, skilled operators observe the trends in the operational data and execute the appropriate operational action at the appropriate timing.
[0066] Furthermore, even when it is determined that operational action is unnecessary because both the molten iron temperature after a long period based on a physical model and the molten iron temperature after a short period based on a statistical model are within the target range, a skilled operator may decide that it is better to perform operational action at an appropriate timing by observing the trends in the operational data.
[0067] Machine learning models are used to execute appropriate operational actions in advance at the appropriate timing, even in such cases, by mimicking the operational actions of exemplary operators. In addition, machine learning models can be used to determine the validity of operational actions based on physical and statistical models.
[0068] The second is a guideline for determining the amount of air-blowing moisture based on a desirable range, considering the realization of low-reducing agent ratio operation and the response to sudden changes in molten iron temperature. When the air-blowing moisture is high, the air-blowing moisture is lowered as an action to raise the molten iron temperature. As the air-blowing moisture is lowered, the reducing agent ratio is reduced. However, if the air-blowing moisture is lowered while it is close to atmospheric moisture, the air-blowing moisture becomes almost equal to atmospheric moisture; consequently, the operational variable for controlling the molten iron temperature to raise the molten iron temperature is limited to the pulverized coal ratio, and the action is limited to raising the pulverized coal ratio. Consequently, when the molten iron temperature fluctuates rapidly, the responsiveness of the operation action deteriorates. It is considered desirable to continue operation while maintaining the air-blowing moisture at a predetermined value so that the responsive air-blowing moisture can be lowered when the molten iron temperature drops sharply, in order to achieve both molten iron temperature control and reduction of the reducing agent ratio. Here, the predetermined value may be determined by adding a value (see β in FIGS. 2A and 2B) that allows for at least one reduction of the blower moisture, which is determined as a single operation amount with respect to atmospheric moisture. Additionally, the predetermined value may be given a range so that the blower moisture is maintained within a desirable range. That is, in cases where it is predicted based on a physical model that the molten iron temperature will decrease over the long term, if the blower moisture exceeds the desirable range, the action to reduce the blower moisture is prioritized. In cases where it is predicted based on a physical model that the molten iron temperature will increase over the long term, if the blower moisture falls below the desirable range, the action to increase the blower moisture is prioritized. In the following, the desirable range of blower moisture may be referred to as the "control range."
[0069] The third point concerns the restriction on the use of pulverized coal based on operational constraints. For example, if the pulverized coal ratio is increased while the reducing agent ratio is high, the amount of powder at the bottom of the furnace may increase, potentially worsening air permeability. Additionally, when the gas temperature at the tip of the trough is low, the combustibility of pulverized coal is poor; therefore, increasing the pulverized coal ratio may result in unburned pulverized coal, potentially worsening air permeability. Consequently, when the reducing agent ratio is at a predetermined upper limit or the gas temperature at the tip of the trough is at a predetermined lower limit, it is desirable to take action to reduce the moisture content of the blowing air in order to raise the molten iron temperature. Furthermore, if the pulverized coal ratio is lowered while the reducing agent ratio is low, the risk of troubles such as worsening air permeability or a sudden drop in molten iron temperature due to sudden disturbances increases. If the pulverized coal ratio is lowered while the gas temperature at the tip of the trough is high, the gas temperature at the tip of the trough may rise further, potentially damaging the equipment surrounding the trough. In addition, if the pulverized coal ratio is lowered while the furnace top gas temperature is low, the furnace top gas temperature may drop further, causing zinc in the raw material to adhere to the furnace wall, thereby worsening air permeability, or soot may be generated at the top of the furnace, potentially deteriorating the furnace top equipment. Therefore, when the reducing agent ratio is at a predetermined lower limit, when the gas temperature at the tip of the mouth is at a predetermined upper limit, or when the furnace top gas temperature is at a predetermined lower limit, it is desirable to increase the moisture content of the blowing air to lower the molten iron temperature.
[0070] In addition, the manipulation amount based on the physical model and the statistical model may be calculated by, for example, a known method of model predictive control or a known method of defining influence coefficients, or it may be an integer. In addition, the manipulation amount based on the machine learning model may be calculated by defining influence coefficients for the action judgment value, or it may be an integer. Based on the concept of determining the manipulation variables and the manipulation amount, an operation action determination unit (12) is constructed.
[0071] (Operation of the operation action decision unit (12))
[0072] FIGS. 2A and FIGS. 2B are flowcharts showing the flow of processing of the operation action determination unit (12). A single flowchart is configured in FIGS. 2A and FIGS. 2B. By processing the flowchart, one of the multiple models is selected based on a judgment criterion based on the trend of the prediction accuracy of the multiple models. The operation of the operation action determination unit (12) is described with reference to FIGS. 2A and FIGS. 2B. The flowcharts of FIGS. 2A and FIGS. 2B are initiated at the timing when an execution command for operation action determination is input to the molten metal temperature control device (10), and proceed to the processing of step S1.
[0073] In the processing of step S1, the operation action determination unit (12) determines the difference between the predicted value of the molten iron temperature based on the physical model calculated by the model calculation unit (physical model predicted molten iron temperature, predicted molten iron temperature after a long time) and the target value of the molten iron temperature (target molten iron temperature). The target range of the difference between the physical model predicted molten iron temperature and the target molten iron temperature (target range of the predicted molten iron temperature after a long time) is greater than or equal to -α1 and less than or equal to α1. If the physical model predicted molten iron temperature after τ1 hours is less than -α1 compared to the target molten iron temperature, the operation action determination processing proceeds to step S11 so that an operation action to raise the molten iron temperature is taken. If the physical model predicted molten iron temperature after τ1 hours is greater than α1 compared to the target molten iron temperature, the operation action determination processing proceeds to step S21 so that an operation action to lower the molten iron temperature is taken. If the molten iron temperature predicted by the physical model after τ1 hours is greater than or equal to -α1 and less than or equal to α1 compared to the target molten iron temperature, it is determined that no action is required in the prediction based on the physical model, and the operation action determination process proceeds to Step S31. Here, τ1 is a future time predicted using the physical model. It is preferable that τ1 be set to 4 to 12 hours, taking into account the time delay until the raw materials charged from the top of the blast furnace descend to the bottom of the furnace (top hole) and the time delay until pulverized coal burns at the top hole and is reflected in the molten iron temperature. Additionally, it is preferable that α1 be set to 1 to 30°C according to the molten iron temperature control target.
[0074] In the processing of Step S11, the operation action determining unit (12) determines the relationship with the control range of the blower moisture determined by considering the reduction of the reducing agent ratio, with respect to the difference between the blower moisture and the ambient moisture at the current time. If the blower moisture at the current time is greater than the ambient moisture + β (g / N㎥), an action to lower the blower moisture is suggested to raise the molten iron temperature, and the series of operation action determination processes is completed. If the blower moisture at the current time is less than or equal to the ambient moisture + β (g / N㎥), the operation action determination process proceeds to Step S12. Here, β is a value that allows for at least one reduction of the blower moisture, which is determined as a single operation amount.
[0075] In the processing of Step S12, the operation action determining unit (12) performs a determination regarding the reduction agent ratio and the gas temperature at the tip of the mouth, which is a limitation on the use of pulverized coal based on operation constraints. If the reduction agent ratio is at the upper limit or the gas temperature at the tip of the mouth is at the lower limit, the operation action determination process proceeds to Step S15 because, in order to raise the molten iron temperature, it is more desirable to lower the moisture content of the blowing air rather than raise the pulverized coal ratio. If the reduction agent ratio is below the upper limit and the gas temperature at the tip of the mouth is above the lower limit, the operation action determination process proceeds to Step S13. Here, in FIG. 2A, the case where the reduction agent ratio is at the upper limit or the gas temperature at the tip of the mouth is at the lower limit corresponds to Yes in Step S12. Also, the case where the reduction agent ratio is below the upper limit and the gas temperature at the tip of the mouth is above the lower limit corresponds to No in Step S12. In addition, the upper limit of the reducing agent ratio and the lower limit of the gas temperature at the tip of the nozzle may be predetermined based, for example, on past performance data or experimental data.
[0076] In the processing of step S13, the operation action determination unit (12) determines the difference between the predicted value of the molten iron temperature based on the statistical model calculated by the model calculation unit (11) (statistical model predicted molten iron temperature, predicted molten iron temperature after a short time) and the target value of the molten iron temperature (target molten iron temperature). The target range of the difference between the statistical model predicted molten iron temperature and the target molten iron temperature (target range of the predicted molten iron temperature after a short time) is -α2 or greater and α2 or less. If the statistical model predicted molten iron temperature after τ2 hours is less than -α2 compared to the target molten iron temperature, it is determined that an action to reduce the responsiveness of the blowing moisture is desirable because there is a high possibility that the molten iron temperature will drop rapidly, and the operation action determination processing proceeds to step S15. That is, if the predicted molten iron temperature after a long time (Step S1) and the predicted molten iron temperature after a short time (this Step S13) both deviate from the target range and the corresponding respective operation actions match the operation actions that raise the molten iron temperature, the process proceeds to Step S15, which determines the operation variables and amounts of the action to lower the blast moisture content, which has good responsiveness, as an operation action based on the predicted molten iron temperature after a short time. If the molten iron temperature predicted by the statistical model after τ2 hours is -α2 or higher compared to the target molten iron temperature, that is, if the respective operation actions corresponding to the predicted molten iron temperature after a long time and the predicted molten iron temperature after a short time do not match, the operation action determination process proceeds to Step S14. Here, τ2 is a future time predicted using the statistical model. Considering the prediction accuracy of the statistical model, it is preferable to set τ2 to 1 to 4 hours. In addition, it is preferable to set α2 to 1 to 30°C depending on the molten iron temperature control target. Here, it is desirable to make α2 a larger value than α1 so that the action is executed in response to a sudden change in molten iron temperature.
[0077] In the processing of step S14, the operation action determination unit (12) makes a determination on the action determination value (machine learning model action determination value) based on the machine learning model calculated by the model calculation unit (11). If the machine learning model action determination value is less than -α3, the operation action determination unit (12) does not execute an active action because the molten iron temperature rise action based on the prediction of molten iron temperature drop based on the physical model and the molten iron temperature drop action based on the machine learning model are in opposition. That is, if the operation action of molten iron temperature drop corresponding to the machine learning model action determination value and the operation action of molten iron temperature rise corresponding to the predicted molten iron temperature after a long time based on the physical model are in opposition, the active action is not executed. Therefore, the series of operation action determination processes is terminated without action. Also, when the machine learning model action judgment value is -α3 or higher, the operation action based on the machine learning model action judgment value is a molten iron temperature increase action or observing the situation (not taking active action), so it is not at least contrary to the operation action of increasing the molten iron temperature corresponding to the prediction of a molten iron temperature decrease after a long time based on the physical model. That is, when the operation action for the predicted molten iron temperature after a long time based on the physical model and the operation action corresponding to the machine learning model action judgment value are not at least contrary to each other, the operation action determination unit (12) suggests an action to increase the pulverized coal ratio based on the prediction of a molten iron temperature decrease after a long time based on the physical model, and the series of operation action judgment processing is completed. Here, the machine learning model action judgment value is the value obtained by subtracting the judgment probability of the molten iron temperature decrease action from the judgment probability of the molten iron temperature increase action calculated as above. Here, the machine learning model action decision value only needs to represent numerically which of the molten iron temperature raising action and molten iron temperature lowering action is appropriate, and is not limited to specific types of values such as probability.For example, machine learning may be performed using the amount of manipulation of an action performed by an exemplary operator, which is obtained as past performance data, and the amount of manipulation itself may be output as the machine learning model action decision value. Also, α3 may be determined based on the machine learning model action decision value, for example, past performance data or experimental data.
[0078] In the processing of Step S15, the operation action determination unit (12) determines the relationship with the control range of the blower moisture determined by considering the reduction of the reducing agent ratio, with respect to the difference between the blower moisture and the ambient moisture at the current time. If the blower moisture at the current time is less than the ambient moisture + γ (g / N㎥), it is determined that there is no room to lower the blower moisture, so no action is performed, and the series of operation action determination processes is terminated. Here, γ is set to a value smaller than the above-mentioned β. The blower moisture at the current time being less than the ambient moisture + γ (g / N㎥) indicates that the blower moisture is almost equal in amount to the ambient moisture. If the blower moisture at the current time is greater than or equal to the ambient moisture + γ (g / N㎥), an action to lower the blower moisture is suggested to raise the molten iron temperature, and the series of operation action determination processes is terminated.
[0079] In the processing of Step S21, the operation action determination unit (12) determines the relationship with the control range of the blower moisture determined by considering the reduction of the reducing agent ratio, with respect to the difference between the blower moisture and the ambient moisture at the current time. If the blower moisture at the current time is less than the ambient moisture + β (g / N㎥), an action to increase the blower moisture is suggested to lower the molten iron temperature, and the series of operation action determination processes is completed. If the blower moisture at the current time is greater than or equal to the ambient moisture + β (g / N㎥), the operation action determination process proceeds to Step S22 to avoid excessive blower moisture.
[0080] In the processing of Step S22, the operation action determining unit (12) performs a determination regarding the reduction agent ratio, the gas temperature at the tip of the mouth, and the gas temperature at the top of the furnace, which is accompanied by a limitation on the use of pulverized coal based on operation constraints. If the reduction agent ratio is at the lower limit, if the gas temperature at the tip of the mouth is at the upper limit, or if the gas temperature at the top of the furnace is at the lower limit, the operation action determination process proceeds to Step S25 because, in order to lower the molten iron temperature, it is more desirable to increase the moisture content of the blowing air rather than lowering the pulverized coal ratio. If the reduction agent ratio is greater than the lower limit, and the gas temperature at the tip of the mouth is below the upper limit, and the gas temperature at the top of the furnace is above the lower limit, the operation action determination process proceeds to Step S23. Here, in FIG. 2B, if the reduction agent ratio is at the lower limit, if the gas temperature at the tip of the mouth is at the upper limit, or if the gas temperature at the top of the furnace is at the lower limit, this corresponds to Yes in Step S22. In addition, the case where the reducing agent ratio is greater than the lower limit, the gas temperature at the tip of the right end is lower than the upper limit, and the gas temperature at the top of the furnace is higher than the lower limit corresponds to No. of Step S22. In addition, the lower limit of the reducing agent ratio, the upper limit of the gas temperature at the tip of the right end, and the lower limit of the gas temperature at the top of the furnace may be predetermined based, for example, past performance data or experimental data.
[0081] In the processing of step S23, the operation action determination unit (12) determines the difference between the statistical model predicted molten iron temperature calculated by the model calculation unit (11) and the target molten iron temperature. If the statistical model predicted molten iron temperature after τ2 hours is greater than α2 compared to the target molten iron temperature, it is determined that an action to increase the responsive blowing moisture is desirable because there is a high possibility that the molten iron temperature will rise rapidly, and the operation action determination processing proceeds to step S25. That is, if the predicted molten iron temperature after a long time (step S1) and the predicted molten iron temperature after a short time (step S23) both deviate from the target range and the corresponding operation action matches an operation action that lowers the molten iron temperature, the process proceeds to step S25, which determines the operation variables and the amount of operation for an action to increase the responsive blowing moisture as an operation action based on the predicted molten iron temperature after a short time. If the molten iron temperature predicted by the statistical model after τ2 hours is α2 or less compared to the target molten iron temperature, that is, if the respective operation actions corresponding to the predicted molten iron temperature after a long time and the predicted molten iron temperature after a short time do not match, the operation action determination process proceeds to step S24.
[0082] In the processing of step S24, the operation action determination unit (12) makes a determination on the machine learning model action determination value calculated by the model calculation unit (11). If the machine learning model action determination value is greater than α3, the operation action determination unit (12) does not execute an active action because the molten iron temperature lowering action based on the prediction of molten iron temperature rise based on the physical model and the molten iron temperature rising action based on the machine learning model are in opposition. That is, if the operation action corresponding to the machine learning model action determination value and the operation action corresponding to the predicted molten iron temperature after a long time based on the physical model are in opposition, an active action is not executed. Therefore, the series of operation action determination processes is terminated without action. Also, when the machine learning model action judgment value is α3 or less, the operation action based on the machine learning model action judgment value is a molten iron temperature lowering action or observing the situation (not taking active action), so it is not at least contrary to the operation action of lowering the molten iron temperature corresponding to the prediction of molten iron temperature rise after a long time based on the physical model. That is, when the operation action for the predicted molten iron temperature after a long time based on the physical model and the operation action corresponding to the machine learning model action judgment value are not at least contrary to each other, the operation action determination unit (12) suggests an action to lower the pulverized coal ratio based on the prediction of molten iron temperature rise after a long time based on the physical model, and the series of operation action judgment processes are completed.
[0083] In the processing of Step S25, the operation action determination unit (12) determines the relationship with the control range of the blower moisture determined by considering the reduction of the reducing agent ratio, with respect to the difference between the blower moisture and the ambient moisture at the current time. If the blower moisture at the current time is greater than the ambient moisture +δ (g / N㎥), it indicates that the blower moisture is added in a larger amount than the ambient moisture. Therefore, if the blower moisture is increased, the moisture introduced into the furnace becomes excessive, and the risk of the molten iron temperature dropping rapidly increases. Therefore, if the blower moisture at the current time is greater than the ambient moisture +δ (g / N㎥), it is determined that there is no action, and the series of operation action determination processes is terminated. Here, δ is set to a value smaller than the above β. For example, δ may be the same value as γ. If the blowing humidity at the current time is less than or equal to atmospheric humidity +δ (g / N㎥), an action to increase the blowing humidity is suggested to lower the molten iron temperature, and a series of operation action determination processes are completed.
[0084] In the processing of step S31, the operation action determination unit (12) determines the difference between the statistical model predicted molten iron temperature calculated by the model calculation unit (11) and the target molten iron temperature. In the processing of step S31, if it is determined based on the physical model that the molten iron temperature after a long time in the future will not change significantly (the predicted molten iron temperature after a long time is within the target range) and if it is predicted based on the statistical model that the molten iron temperature after a short time will change rapidly (the predicted molten iron temperature after a short time deviates from the target range), the prediction of the molten iron temperature after a short time in the statistical model is prioritized. If the statistical model predicted molten iron temperature after τ2 hours is less than -α2 compared to the target molten iron temperature, it is determined that an action to reduce the responsive blowing moisture is desirable because there is a high possibility that the molten iron temperature will drop rapidly, and the operation action determination processing proceeds to step S15. If the molten iron temperature predicted by the statistical model after τ2 hours is greater than α2 compared to the target molten iron temperature, it is determined that an action to increase the responsiveness of the blowing moisture is desirable because there is a high probability that the molten iron temperature will rise rapidly, and the operation action determination process proceeds to step S25. If the molten iron temperature predicted by the statistical model after τ2 hours is greater than or equal to -α2 and less than or equal to α2 compared to the target molten iron temperature, the operation action determination process proceeds to step S32.
[0085] In the processing of step S32, the operation action determination unit (12) makes a determination on the machine learning model action determination value calculated by the model calculation unit (11). In step S32, if it is determined that the future molten iron temperature will not change significantly whether by a physical model or a statistical model, that is, if the predicted molten iron temperature after a long time based on the physical model and the predicted molten iron temperature after a short time based on the statistical model are both within the target range, the machine learning model action determination value is prioritized. If the machine learning model action determination value is greater than α3, the operation action determination unit (12) suggests an action to increase the pulverized coal ratio, and the series of operation action determination processes are terminated. If the machine learning model action determination value is less than -α3, the operation action determination unit (12) suggests an action to lower the pulverized coal ratio, and the series of operation action determination processes are terminated. When the machine learning model action judgment value is -α3 or higher and α3 or lower, the operation action determination unit (12) determines that no action is needed and the series of operation action judgment processes are terminated.
[0086] In this way, the operation action determination unit (12) determines the operation variables and the amount of operation as an operation action based on the value calculated by the model calculation unit (11), the management range of the blowing moisture determined by considering the reduction of the reducing agent ratio, and the limit on the use of pulverized coal based on the operation constraint. The value calculated by the model calculation unit (11) includes the predicted molten iron temperature after a long time calculated using a physical model, the predicted molten iron temperature after a short time calculated using a statistical model, and the action determination value calculated using a machine learning model.
[0087] FIG. 5 is a diagram showing an example of a system configuration equipped with a molten metal temperature control device (10). The system may be configured to include a molten metal temperature control device (10), an operation data server (94), a control calculator (74), and a display unit (95). The operation data server (94) accumulates data regarding the molten metal manufacturing process. The molten metal temperature control device (10) obtains actual values, etc., in the manufacturing process executed in the molten metal manufacturing facility from the operation data server (94). A memory device (e.g., memory) in which the operation data server (94) accumulates data regarding the manufacturing process corresponds to the operation database (20). The control calculator (74) controls the molten metal manufacturing facility. The display unit (95) displays the guidance operation amount (operation action determined according to the above flowchart) output from the molten metal temperature control device (10) which functions as an operation guidance device. Here, the display unit (95) may be a display device such as a Liquid Crystal Display or an Organic Electro-Luminescence Panel. Additionally, the display unit (95) may be realized by a display of a terminal device such as a smartphone or tablet. The operation data server (94), the control calculator (74), and the display unit (95) can communicate with the molten metal temperature control device (10) via a network. The network is, for example, the Internet.
[0088] The molten iron temperature control device (10) may execute an operation guidance method that presents the operation action determined by the above processing so that the blast furnace operator can verify it. The presentation may include display on the above display unit (95), but voice may be used, and is not limited to a specific presentation method. The operator may change the operation conditions of the molten iron manufacturing facility based on the guidance operation amount (operation variable and operation amount) shown on the display unit (95). Such operation guidance for the molten iron manufacturing facility may be executed as part of a manufacturing method for producing molten iron. In addition, the molten iron temperature control device (10) may output the determined operation variable and operation amount to a control calculator (74), and the control calculator (74) may control the molten iron manufacturing facility according to these. The contents of the operation action determined by the molten iron temperature control device (10) may be automatically reflected in the manufacturing or operation by using, for example, a communication function between a control calculator (74) and a molten iron manufacturing facility. That is, as part of the molten iron manufacturing method and operation method, the output and automatic change of such operation variables and operation quantities may be carried out. Here, each of the operation data server (94), the control calculator (74), and the display unit (95) may be located in the same place as the molten iron temperature control device (10) (for example, within the same factory) or may be physically separated.
[0089] As described above, the molten iron temperature control method executed by the molten iron temperature control device (10) may be part of the blast furnace operation method. For example, the blast furnace operation method may include a step of controlling the blast furnace according to an operation action determined by the molten iron temperature control method. Additionally, the molten iron temperature control method or the blast furnace operation method may be part of the molten iron manufacturing method. For example, the molten iron manufacturing method may include a step of manufacturing molten iron by controlling the blast furnace according to the blast furnace operation method.
[0090] (Example)
[0091] Operation was carried out by controlling the pulverized coal ratio and blowing moisture using a molten iron temperature control device (10). Here, α1 and α2 were set to 30 ℃. α3 was set to 0.3. β was set to 7 g / N㎥. γ was set to 0 g / N㎥. Also, δ was set to 12 g / N㎥. Here, in this embodiment, α3 is a probability and is a dimensionless number.
[0092] FIG. 3 is a diagram showing the changes in molten iron temperature, coke ratio, pulverized coal ratio, blowing moisture, and atmospheric moisture after operating a blast furnace for 4 days using the molten iron temperature control device (10) related to the present embodiment. The numbers on the horizontal axis represent the number of days. The graph of molten iron temperature in FIG. 3 shows the difference from the target molten iron temperature at zero time. In addition, other graphs set the period average value (average value over 4 days) to zero. In the graph of molten iron temperature in FIG. 3, the dotted line represents the target value and the solid line represents the actual value. In addition, in the graph of blowing moisture and atmospheric moisture in FIG. 3, the dotted line represents the blowing moisture and the solid line represents the atmospheric moisture. In the graph of coke ratio in FIG. 3, a large fluctuation occurred due to manual operation by the operator, but afterwards, the molten iron temperature control device (10) manipulated the blowing moisture and pulverized coal ratio as an operation action. Due to the operation action of the molten iron temperature control device (10), the influence of fluctuations in the coke ratio on the molten iron temperature decreased over time, and the molten iron temperature was controlled to be near the target value. As a result of automatically controlling the blowing moisture ratio and pulverized coal ratio by the molten iron temperature control device (10) throughout the entire period, the molten iron temperature could be maintained within the range of -30 to 30 ℃.
[0093] FIG. 4 is a diagram showing a comparison of the molten iron temperature deviation and reducing agent ratio in the case where a blast furnace operation method using the molten iron temperature control device (10) related to the present embodiment is performed and a conventional operation method by an operator is performed. In the comparison of FIG. 4, standardization is implemented such that the molten iron temperature deviation and reducing agent ratio in the result of the conventional operation method are set to 1. As shown in FIG. 4, by using the molten iron temperature control device (10), it was possible to achieve both improved precision of molten iron temperature control and a reduction in the reducing agent ratio compared to the conventional method.
[0094] As described above, the molten iron temperature control method, molten iron temperature control device (10), operation guidance method, blast furnace operation method, and molten iron manufacturing method related to the present embodiment make it possible to achieve both a reduction in the reducing agent ratio and high-precision molten iron temperature control. That is, by using the execution results of multiple models regarding blast furnace operation separately, the disadvantages of each model can be compensated for, and an optimal operation action including the determination of operation variables for molten iron temperature control can be performed.
[0095] Although embodiments have been described, the present disclosure is not limited by the technology and drawings that constitute part of the present disclosure according to the present embodiments. That is, other embodiments, examples, and operational technologies, etc., made by those skilled in the art based on the present embodiments are all included within the technical scope of the present disclosure.
[0096] For example, in the above embodiment, a process is performed to supplement the molten iron temperature prediction by a statistical model and the molten iron temperature control action prediction by a machine learning model, based on the molten iron temperature prediction by a physical model. For example, the priority of the model serving as the basis for action presentation may be changed by rearranging the order of the flowchart. An embodiment in which the priority of the model is changed is included within the scope of the present disclosure. Furthermore, while the above embodiment uses a physical model, a statistical model, and a machine learning model, the machine learning model may be omitted, for example. For example, in cases where it is difficult to anticipate that the prediction accuracy of the physical model and the statistical model will deteriorate due to the influence of disturbances, it is possible to achieve both a reduction in the reducing agent ratio and high-precision molten iron temperature control simultaneously even if the machine learning model is omitted. Explanation of the symbols
[0097] 10: Molten iron temperature control device 11 : Model calculation section 12: Operation Action Decision Unit 20: Operation Database 74 : Control calculator 94 : Operation Data Server 95 : Display section
Claims
Claim 1 A method for controlling molten iron temperature, comprising: a model calculation step for calculating an action judgment value of an operation action for controlling molten iron temperature using a plurality of models that take operation data of a blast furnace as input; and an operation action determination step for selecting one of the plurality of models based on a judgment criterion based on the trend of the prediction accuracy of the plurality of models using the calculated molten iron temperature or the action judgment value, and determining an operation variable and an operation amount as the operation action. Claim 2 A method for controlling molten iron temperature according to claim 1, wherein the plurality of models include a physical model capable of expressing a transient state by calculating heat transfer, chemical reaction, and fluid flow inside the blast furnace, and a statistical model that predicts the molten iron temperature by extracting similar operating conditions based on past operating data of the blast furnace. Claim 3 A method for controlling molten iron temperature according to claim 2, wherein the plurality of models are generated by machine learning using learning data that associates past operation data of the blast furnace with exemplary operation actions performed by an operator, and further comprises a machine learning model that takes the operation data of the blast furnace as input and outputs an operation action determined to be appropriate. Claim 4 A molten iron temperature control method according to claim 3, wherein the operation action determination step determines the operation variables and the amount of operation as the operation action based on the predicted molten iron temperature after a long time calculated using the physical model, the predicted molten iron temperature after a short time calculated using the statistical model, the action determination value calculated using the machine learning model, the management range of the blowing moisture determined considering the reduction of the reducing agent ratio, and the restriction on the use of pulverized coal based on operation constraints. Claim 5 A molten iron temperature control method according to claim 4, wherein the operation action determination step determines an operation variable and an operation amount as the operation action based on the action determination value when the predicted molten iron temperature after a long time and the predicted molten iron temperature after a short time are both within the target range. Claim 6 A molten iron temperature control method according to claim 4 or 5, wherein the operation action determination step determines an operation variable and an operation amount as the operation action based on the predicted molten iron temperature after a short time, when both the predicted molten iron temperature after a long time and the predicted molten iron temperature after a short time deviate from the target range and the corresponding operation actions match, or when the predicted molten iron temperature after a long time is within the target range and the predicted molten iron temperature after a short time deviates from the target range. Claim 7 A molten iron temperature control method according to any one of claims 4 to 6, wherein the operation action determination step determines an operation amount and an operation amount as an operation action based on the predicted molten iron temperature after a long time, in the case where the predicted molten iron temperature after a long time deviates from a target range and the respective operation actions corresponding to the predicted molten iron temperature after a short time do not match, and the operation action corresponding to the predicted molten iron temperature after a long time and the operation action corresponding to the action judgment value are at least not opposite, and if the operation action corresponding to the predicted molten iron temperature after a long time and the operation action corresponding to the action judgment value do not match. Claim 8 A method for controlling molten iron temperature according to any one of claims 4 to 7, wherein the operating constraint comprises at least one of an upper limit of the gas temperature at the tip of the mouth, a lower limit of the gas temperature at the tip of the mouth, an upper limit of the reducing agent ratio, a lower limit of the reducing agent ratio, and a lower limit of the gas temperature at the top of the furnace. Claim 9 A molten iron temperature control method according to any one of claims 4 to 8, wherein the management range of the blowing moisture is determined by adding a value that allows for at least one reduction of the amount of blowing moisture, determined as a single operation amount, with respect to atmospheric moisture. Claim 10 A molten iron temperature control device comprising: a model calculation unit that calculates an action determination value of a molten iron temperature or an operation action for controlling the molten iron temperature using a plurality of models that take operation data of a blast furnace as input; and an operation action determination unit that selects one of the plurality of models based on a determination criterion based on the trend of the prediction accuracy of the plurality of models using the calculated molten iron temperature or the action determination value, and determines an operation variable and an operation amount as the operation action. Claim 11 A molten iron temperature control device according to claim 10, wherein the plurality of models include a physical model capable of expressing a transient state by calculating heat transfer, chemical reactions, and fluid flow inside the blast furnace, and a statistical model that predicts the molten iron temperature by extracting similar operating conditions based on past operating data of the blast furnace. Claim 12 A molten iron temperature control device according to claim 11, wherein the plurality of models are generated by machine learning using learning data that associates past operation data of the blast furnace with exemplary operation actions performed by an operator, and further comprises a machine learning model that takes the operation data of the blast furnace as input and outputs an operation action determined to be appropriate. Claim 13 In claim 12, the above-mentioned operation action determining unit determines an operation variable and an operation amount as the operation action based on the predicted molten iron temperature after a long time calculated using the physical model, the predicted molten iron temperature after a short time calculated using the statistical model, the action determination value calculated using the machine learning model, a management range of blowing moisture determined considering the reduction of the reducing agent ratio, and a restriction on the use of pulverized coal based on operation constraints. Claim 14 An operation guidance method that presents the operation action determined by the molten iron temperature control method described in any one of claims 1 to 9 so that the operator of the blast furnace can verify it. Claim 15 A method for operating a blast furnace, comprising a step of controlling the blast furnace according to the operating action determined by the molten iron temperature control method described in any one of claims 1 to 9. Claim 16 A method for producing molten iron, comprising the step of controlling the blast furnace according to the method of operating the blast furnace described in claim 15 to produce molten iron.