Temperature prediction system, control device, temperature prediction method and control method

Abstract

The temperature prediction system is a temperature prediction system that predicts the heated material temperature, which is the temperature of the heated material in a continuous heating furnace having multiple continuous zones, and includes a collection unit that collects operational data and stores the overall heat absorptivity in association with the operational data for each predetermined range, a learning unit that generates a machine learning model for each zone, a prediction unit that predicts the heated material temperature at the outlet side of the continuous heating furnace using the predicted overall heat absorptivity value of each zone, and an output unit that outputs calculation results including the predicted heated material temperature.

Description

[Technical Field]

[0001] The present disclosure relates to a temperature prediction system, a control device, a temperature prediction method, and a control method.The present disclosure particularly relates to a temperature control, operator guidance, and steel strip temperature prediction technology used in furnace operation in a continuous heating furnace that heats materials to be heated, including continuous steel strips, and can be widely used in industrial fields such as the steel manufacturing industry or the steel plate processing industry. [Background technology]

[0002] Heat treatment of steel sheets is an important process that affects the quality of the steel sheets, and various studies have been conducted on methods for estimating the steel sheet temperature after heat treatment, which is used for temperature control and operator guidance.

[0003] A widely used method for estimating the steel plate temperature is to calculate the temperature transition of the steel plate using a known heat transfer mathematical model. The heat transfer mathematical model treats the thermal radiation heat transfer phenomenon (radiative heat transfer and convective heat transfer) in a heating furnace as radiation heat transfer from the furnace gas temperature (furnace temperature) to the steel plate, and has a parameter called the overall heat absorptivity (or overall heat transfer coefficient).

[0004] Here, the overall heat absorptance varies depending on the characteristics of the equipment, such as the furnace shape, etc. The overall heat absorptance is determined based on, for example, numerical calculations such as heat transfer analysis or heat transfer experiments.

[0005] However, the overall heat absorptivity varies depending on the position of the steel sheet in the furnace, the operating conditions (heating conditions), etc. Therefore, techniques have been proposed to correct a preset (standard) overall heat absorptivity for various operating conditions.

[0006] Patent Document 1 discloses a method for determining the overall heat transfer coefficient of a heat transfer mathematical model, which is applied to combustion gas flow rate control in a continuous annealing furnace, as a function of the combustion gas flow rate and the width of the strip. Patent Document 1 states that by controlling the combustion gas flow rate using this overall heat transfer coefficient, it is possible to compensate for the effect on the temperature control of the subsequent strip caused by the change in strip width from the preceding strip to the subsequent strip.

[0007] Patent Document 2 discloses a technique for correcting the overall heat absorption rate of a heating furnace based on the flame length of an axial flow burner in the heating furnace.

[0008] Patent Document 3 discloses a technique for correcting the overall heat absorption rate of a heating furnace based on the number of combustions of burners in the heating furnace (total combustion amount).

[0009] Patent Document 4 discloses a method for predicting the overall heat absorptance of a continuous heating furnace, in which a plurality of overall heat absorptances are calculated in advance by heat transfer analysis (numerical calculation) within the range of change of variable items in actual operation, and the overall heat absorptance is predicted during actual operation using a regression equation with the variable items as variables. The variable items include the billet charging temperature, billet withdrawal temperature, production volume, time the billet has been in the furnace, furnace temperature of each zone, and the dimensionless position in the furnace length direction of the zone for which the overall heat absorptance is to be calculated, and a regression equation for the overall heat absorptance is calculated for each zone.

[0010] Furthermore, Patent Document 5 discloses a technology for sequentially estimating the overall heat absorptivity of multiple zones in a continuous heating furnace online by performing a recursive least squares method based on the estimation error of the slab temperature at the outlet of the furnace, and updating the overall heat absorptivity of each zone. [Prior art documents] [Patent documents]

[0011] [Patent Document 1] Japanese Patent Application Publication No. 4-246130 [Patent Document 2] Japanese Patent Application Publication No. 2019-014953 [Patent Document 3] Japanese Patent Application Publication No. 2018-003084 [Patent Document 4] Japanese Patent Application Publication No. 10-324926 [Patent Document 5] Japanese Patent Application Publication No. 6-264153 Summary of the Invention [Problem to be solved by the invention]

[0012] Here, the heat transfer mathematical model having the overall heat absorptivity as a parameter is a simplified model of various thermal radiation heat transfer phenomena in a furnace, and does not consider, for example, thermal radiation heat transfer in the longitudinal direction of a heating furnace. Actual thermal radiation heat transfer phenomena are complex. Therefore, it is difficult to estimate the steel plate temperature with high accuracy by calculations in which the overall heat absorptivity is set to a single value.

[0013] As described above, in the methods of Patent Documents 1 to 4, the reference overall heat absorption rate can be corrected for operational factors such as the number of combustions, the combustion gas flow rate, the time in the furnace, the furnace temperature of each zone, and the flame length using a correction means (for example, a correction coefficient, a regression equation, etc.).

[0014] However, actual heating furnaces are configured as continuous heating furnaces with multiple consecutive zones (heating zones), and temperature distribution exists within the heating furnace. It is also assumed that there are many other factors affecting the overall heat absorptivity in addition to the factors disclosed in Patent Documents 1 to 4. The methods in Patent Documents 1 to 4 are based on the use of a heat transfer mathematical model that does not take into account thermal radiation heat transfer in the longitudinal direction of the heating furnace, and it is difficult to select all of the influencing factors or estimate the effects of all of the influencing factors. Furthermore, the methods in Patent Documents 1 to 4 have the problem of being difficult to immediately reflect changes in the state of the heating furnace (e.g., deterioration of refractories inside the furnace over time) in the overall heat absorptivity, resulting in a lack of real-time performance.

[0015] Here, the method of Patent Document 5 performs a recursive least squares method to sequentially estimate the overall heat absorptivity of multiple zones in the furnace online, so that changes in the state of the heating furnace can be immediately reflected in the overall heat absorptivity. However, the method of Patent Document 5 cannot distinguish between the influence of operating conditions (short-term factors) and the influence of changes in the state of the heating furnace (long-term factors), so sufficient improvement in accuracy cannot be expected.

[0016] In view of the above circumstances, the present disclosure aims to provide a temperature prediction system, a control device, a temperature prediction method, and a control method that are capable of predicting the temperature of a material to be heated in a heating furnace while taking into account changes in operating conditions and the equipment state of the heating furnace. The content of the present disclosure can also be applied to predicting the temperature in a heating furnace of a material to be heated, such as a steel strip, a steel plate obtained by cutting the steel strip into plates, or a steel ingot including a slab. In the following embodiments, the material to be heated is described as a steel strip, and the temperature of the material to be heated is described as the steel strip temperature. When the material to be heated is a steel plate or a steel ingot, the steel strip temperature in the following description can be read as the steel plate temperature and the steel ingot temperature, respectively. [Means for solving the problem]

[0017] (1) A temperature prediction system according to an embodiment of the present disclosure includes: A temperature prediction system for predicting a temperature of a material to be heated, which is a temperature of a material to be heated in a continuous heating furnace having a plurality of continuous zones, comprising: a collection unit that collects operational data including the temperature of the heated material for each predetermined range of the heated material in each zone, calculates an overall heat absorptivity of each zone for each predetermined range based on a change in the temperature of the heated material from the inlet side to the outlet side of each zone, and stores the overall heat absorptivity in association with the operational data for each predetermined range; a learning unit that sets any one zone among the plurality of zones as a target zone, sequentially learns a machine learning model of the target zone for each of the predetermined ranges, the machine learning model having the overall heat absorptivity of the target zone as an output variable and the operational data of the target zone and a zone upstream of the target zone as input variables, and generates the machine learning model for each zone; a prediction unit that inputs the operational data of the target zone and a zone upstream of the target zone into the machine learning model of the target zone, calculates a predicted value of the overall heat absorptance of the target zone, and predicts the temperature of the heated material at the outlet side of the continuous heating furnace using the predicted value of the overall heat absorptance of each zone; and an output unit that outputs the calculation result including the predicted temperature of the material to be heated.

[0018] (2) As one embodiment of the present disclosure, in (1), The collected operational data includes at least the furnace temperature, exhaust gas temperature, fuel flow rate, air flow rate, inlet temperature of the heated material, and outlet temperature of the heated material in each zone, The input variables of the machine learning model include at least the furnace temperature, exhaust gas temperature, fuel flow rate, and air flow rate in each zone.

[0019] (3) As an embodiment of the present disclosure, in (2), The furnace temperature is measured at a plurality of positions spaced apart in the conveying direction of the material to be heated.

[0020] (4) As an embodiment of the present disclosure, in any one of (1) to (3), The machine learning model is generated using machine learning techniques including Lasso regression, ridge regression, linear regression, local regression, principal component regression, PLS regression, and neural network.

[0021] (5) A control device according to an embodiment of the present disclosure includes: The continuous heating furnace is controlled based on the predicted temperature of the material to be heated output by any one of the temperature prediction systems (1) to (4).

[0022] (6) A temperature prediction method according to an embodiment of the present disclosure includes: A temperature prediction method executed by a temperature prediction system for predicting a temperature of a material to be heated, which is a temperature of a material to be heated in a continuous heating furnace having a plurality of continuous zones, comprising: a collection step of collecting operational data including the temperature of the heated material for each predetermined range of the heated material in each zone, calculating an overall heat absorptivity of each zone for each predetermined range based on a change in the temperature of the heated material from the inlet side to the outlet side of each zone, and storing the overall heat absorptivity in association with the operational data for each predetermined range; a learning step of sequentially learning a machine learning model of the target zone for each of the predetermined ranges, with any one of the plurality of zones as a target zone, the overall heat absorptivity of the target zone as an output variable, and the operational data of the target zone and a zone upstream of the target zone as input variables, to generate the machine learning model for each zone; a prediction step of inputting the operational data of the target zone and a zone upstream of the target zone into the machine learning model of the target zone, calculating a predicted value of the overall heat absorptance of the target zone, and predicting the temperature of the heated material at the outlet side of the continuous heating furnace using the predicted value of the overall heat absorptance of each zone; and an output step of outputting the calculation result including the predicted temperature of the material to be heated.

[0023] (7) As an embodiment of the present disclosure, in (6), The collected operational data includes at least the furnace temperature, exhaust gas temperature, fuel flow rate, air flow rate, inlet temperature of the heated material, and outlet temperature of the heated material in each zone, The input variables of the machine learning model include at least the furnace temperature, exhaust gas temperature, fuel flow rate, and air flow rate in each zone.

[0024] (8) A control method according to an embodiment of the present disclosure includes: The continuous heating furnace is controlled based on the predicted temperature of the material to be heated output by the temperature prediction method (6) or (7). [Effects of the Invention]

[0025] According to the present disclosure, it is possible to provide a temperature prediction system, a control device, a temperature prediction method, and a control method that are capable of predicting the temperature of a material to be heated in a heating furnace while taking into account changes in operating conditions and the equipment state of the heating furnace. [Brief explanation of the drawings]

[0026] [Figure 1] FIG. 1 is a schematic diagram illustrating an example configuration of a temperature prediction system according to an embodiment of the present disclosure. [Figure 2] FIG. 2 is a flowchart showing the process of a temperature prediction method according to an embodiment of the present disclosure. [Figure 3] FIG. 3 is a diagram for explaining a machine learning model that outputs an overall heat absorption rate. [Figure 4] FIG. 4 is a diagram showing the top feature importance of the machine learning model. [Figure 5] FIG. 5 is a diagram illustrating an example of changes in exhaust gas temperature and overall heat absorption rate. [Figure 6] FIG. 6 is a diagram showing the prediction accuracy of the overall heat absorptivity. [Figure 7A] FIG. 7A is a diagram showing the predicted results of the steel strip temperature in the comparative example. [Figure 7B] FIG. 7B is a diagram showing the predicted results of the steel strip temperature in the example. [Figure 8] FIG. 8 is a diagram for explaining the heat balance in each zone of the continuous heating furnace. DETAILED DESCRIPTION OF THE INVENTION

[0027] A temperature prediction system, a control device, a temperature prediction method, and a control method according to an embodiment of the present disclosure are described below. The temperature prediction system and the temperature prediction method executed by the temperature prediction system according to this embodiment predict the temperature of a material to be heated, which is the temperature of a material to be heated in a continuous heating furnace having multiple continuous zones (i.e., multiple continuous heating zones, see zones 1 to 3 in FIG. 1 ). As described above, this embodiment will be described assuming that the material to be heated is a steel strip (continuous steel strip) and that the temperature of the material to be heated is the steel strip temperature. However, as described above, the material to be heated is not limited to a steel strip and may be, for example, a steel plate or a steel ingot. When the material to be heated is a steel plate or a steel ingot, the steel strip temperature in the following description will be interpreted as the steel plate temperature or the steel ingot temperature.

[0028] The inventors have investigated the heating conditions (i.e., the operating conditions for the continuous heating furnace) that affect the overall heat absorption rate for each zone of the continuous heating furnace. Then, the inventors have investigated in detail the heat balance (relationship between heat input and heat output) of each zone of the continuous heating furnace shown in Fig. 8 and the heating conditions that are thought to be influencing factors.

[0029] The heat input includes the heat content of the steel strip, the heat of combustion or sensible heat of the fuel, the sensible heat of the combustion air, and the sensible heat of the furnace gas entering from the adjacent heating zone. The heat output includes the heat content of the steel strip, the sensible heat of the furnace gas, the sensible heat of the exhaust gas, and other heat outputs. Other heat outputs include, for example, the heat dissipation of the furnace body, the sensible heat of the gas flowing into the adjacent heating zone, and other heat losses.

[0030] The above heat input or heat output can be calculated based on influencing factors. The heat content of the steel strip, which is heat input, can be calculated from the steel strip temperature at the entrance to the zone. The heat of combustion or sensible heat of the fuel can be calculated from the fuel flow rate (gas flow rate) of the zone. The sensible heat of the combustion air can be calculated from the air flow rate of the zone. The sensible heat of the furnace gas entering from an adjacent zone can be calculated from the heating conditions of the adjacent zone (furnace temperature, fuel flow rate, air flow rate, etc.). In addition, the heat content of the steel strip, which is heat output, can be calculated from the steel strip temperature at the exit of the zone. The sensible heat of the furnace gas can be calculated from the furnace temperature (ambient temperature) of the zone. The sensible heat of the exhaust gas can be calculated from the exhaust gas temperature of the zone. Other heat outputs can be calculated from the furnace equipment status (furnace body radiation, furnace opening radiation heat loss, other heat losses).

[0031] The overall heat absorptivity is calculated based on the difference in heat content between the steel strip at the inlet and outlet of a zone. In other words, the overall heat absorptivity is calculated from the difference in steel strip temperature. From the perspective of heat balance, the overall heat absorptivity is thought to be affected by fluctuations in heating conditions. The inventors have conceived a new method for accurately predicting the overall heat absorptivity in a continuous heating furnace consisting of multiple consecutive zones. In this method, the overall heat absorptivity for each zone is calculated based on changes in the steel strip temperature within each zone. For each zone, a machine learning model is sequentially generated using training, with the overall heat absorptivity as the output variable and the heating conditions of the zone and adjacent zones as input variables. One of the features of this method is that it incorporates factors not considered in conventional methods (heating conditions of adjacent upstream zones, exhaust gas temperature) as input variables for the overall heat absorptivity prediction model. In addition, this method simultaneously allows for the selection of factors with high influence importance (variable selection) and the estimation of the overall heat absorptivity. In addition, the machine learning model can be generated in real time, and changes in the state of the heating furnace can be immediately reflected in the overall heat absorption rate. Details will be described below with reference to the drawings.

[0032] FIG. 1 is a schematic diagram showing an example of the configuration of a temperature prediction system according to this embodiment. In this embodiment, the temperature prediction system predicts the temperature of a steel strip in a continuous heating furnace of a hot-dip galvanizing production line. The hot-dip galvanizing production line is an example of a production line for steel products and includes a continuous annealing process in which thin plates are continuously annealed in a continuous heating furnace. The temperature prediction system predicts the temperature of the steel strip passing through the furnace in the continuous annealing process. As shown in FIG. 1, the hot-dip galvanizing production line includes a pot (galvanizing tank), cooling equipment, and a continuous heating furnace located upstream of the cooling equipment. The continuous heating furnace is composed of a first zone, a second zone, and a third zone. In this embodiment, the continuous heating furnace has three zones (heating zones), but the number is not limited to three and may be any number. Furthermore, the zones may be divided, for example, by equipment unit when the continuous heating furnace is subdivided, or by control unit.

[0033] As shown in FIG. 1, the temperature prediction system according to this embodiment includes a data collection device and a continuous steel strip temperature prediction device.

[0034] The data collection device includes a first collection unit that collects operational data such as steel strip temperature, furnace temperature, fuel flow rate, air flow rate, line speed (i.e., steel strip conveying speed), and exhaust gas temperature from the hot-dip galvanizing production line in real time. The data collection device also calculates φ cg The data collection device includes a second collection unit that calculates (actual value). The collection unit is composed of the first collection unit and the second collection unit. In other words, the data collection device includes a collection unit, and the collection unit includes the first collection unit and the second collection unit. The data collection device also includes a performance database that stores the collected information and calculated values. Here, φ cg is the overall heat transfer coefficient, and the parentheses distinguish between the measured value and the predicted value. cg (Actual value) is the overall heat transfer coefficient of the actual value obtained from the steel strip temperature. cg (Predicted value) is the predicted overall heat transfer coefficient calculated using the prediction model described below.

[0035] The continuous steel strip temperature prediction device is a data acquisition device. cg The continuous steel strip temperature prediction device includes a learning unit that acquires operational data such as the furnace temperature, fuel flow rate, air flow rate, line speed, and exhaust gas temperature, and generates a prediction model of the overall heat absorption rate used to calculate the steel strip temperature. cg (predicted value) is calculated in real time, and φ cgThe continuous steel strip temperature prediction device includes a prediction unit that predicts the temperature of the steel strip before entering the continuous heating furnace from the (predicted value). The continuous steel strip temperature prediction device also includes an output unit that outputs the calculation results by the prediction unit, such as the predicted temperature of the steel strip before entering. The output unit may output the calculation results to, for example, a calculation result display monitor that is viewed by an operator operating the factory, allowing the operator to adjust operating conditions such as the furnace temperature based on the calculation results. The output unit may also directly output the calculation results to a control device that controls, for example, the fuel flow rate and line speed of a hot-dip galvanizing production line, enabling automatic control.

[0036] In the example of Figure 1, the temperature prediction system is composed of two devices: a data collection device and a continuous steel strip temperature prediction device, which can communicate with each other. However, the device configuration is not limited to this. The temperature prediction system may be composed of one or more devices. If composed of multiple devices, each device may be located in multiple locations and configured to be able to send and receive data between them via a network. For example, the temperature prediction system may be composed of a single computer, and the computer's processor may function as the data collection device and the continuous steel strip temperature prediction device via a program. In this case, the performance database may be stored in the computer's memory. For example, the temperature prediction system may be configured such that one computer functions as the data collection device and another computer functions as the continuous steel strip temperature prediction device. Furthermore, for example, the temperature prediction system may be configured such that at least one of the data collection device and the continuous steel strip temperature prediction device is connected to multiple computers via a network.

[0037] 2 is a flowchart showing the processing of the temperature prediction method executed by the temperature prediction system according to this embodiment. The processing of the temperature prediction method is executed while steel products are being manufactured by the hot-dip galvanizing production line (i.e., during operation), and the generation of a prediction model by the learning unit is also carried out in real time.

[0038] The first collection unit collects operational data in real time from an operating hot-dip galvanizing production line (step S1). The data collection device may also transmit the collected operational data to a continuous steel strip temperature prediction device (step S2). The operational data may be collected from sensors within the line or a process computer (the control device in FIG. 1). Steps S1 and S2 correspond to the first collection step. The collected operational data includes the steel strip temperature for each predetermined range of the continuous steel strip in each zone. Here, the "predetermined range" refers to a certain range in the longitudinal direction (i.e., the conveying direction) of the continuous steel strip. The predetermined range may be determined based on the moving distance of the continuous steel strip in the conveying direction per unit time (i.e., the threading speed) and the sampling period of the operational data. The predetermined range may also be expressed in units of the length of the continuous steel strip (e.g., 1 m).

[0039] The operational data for each zone collected by the first collection unit preferably includes heating conditions that are considered to be influential factors in the heat balance of each zone of the continuous heating furnace (see FIG. 8). Specifically, the collected operational data includes at least the furnace temperature, exhaust gas temperature, fuel flow rate, air flow rate, inlet steel strip temperature, and outlet steel strip temperature for each zone. Here, the inlet steel strip temperature and the outlet steel strip temperature may be given as the steel strip temperature change, which is the difference between these temperatures.

[0040] Furthermore, it is preferable that the furnace temperature be measured at multiple positions spaced apart in the conveying direction of the continuous steel strip. Measurement at multiple positions can incorporate the influence of the temperature distribution in the conveying direction inside the heating furnace, improving the prediction accuracy of the machine learning model for the overall heat absorption rate. For example, since the third zone in Figure 1 is long in the conveying direction, multiple furnace thermometers are arranged along the conveying direction. Then, the furnace temperature may be measured when each predetermined area of ​​the continuous steel strip passes the position where each furnace thermometer is arranged.

[0041] The second collection section calculates the temperature of the steel strip from the measured value of φ cg The second collection unit calculates φ (actual value) based on, for example, the amount of change in the steel strip temperature and the standard (specification) of the continuous steel strip (step S3). cgStep S3 corresponds to the second collection step. Steps S1 to S3 correspond to a collection step that combines the first collection step and the second collection step.

[0042] In this embodiment, a device (temperature measuring device) for measuring temperature in the longitudinal direction within the continuous heating furnace is provided, and the temperature of the continuous steel strip is continuously measured by the temperature measuring device and stored as time-series data by the control device. From the time-series data, the temperature change amount of the continuous steel strip before and after passing through the continuous heating furnace can be calculated. First, from the distance from the entrance of the continuous heating furnace to the temperature measuring device and the conveying speed, the time it takes for a predetermined position on the continuous steel strip to reach the temperature measuring device from the entrance of the continuous heating furnace can be calculated. Even if there are multiple temperature measuring devices within the furnace, the time it takes to reach each temperature measurement point can be calculated similarly from the distance between the temperature measuring devices and the conveying speed. Using the time-series data of the steel strip temperature at each temperature measurement point and the time it takes to reach each temperature measurement point, the temperature change amount of the predetermined position on the continuous steel strip before and after passing through the continuous heating furnace can be calculated, and φ can be calculated from the change in steel strip temperature. cg Furthermore, for a continuous heating furnace having multiple continuous zones, the overall heat absorption rate of each zone can be calculated for each predetermined range of the continuous steel strip based on the change in steel strip temperature from the inlet to the outlet of each zone.

[0043] The second collection section measures the overall heat absorption rate (φ cg The overall heat absorption rate (actual value) is calculated for each predetermined range of the continuous steel strip, and the calculated overall heat absorption rate is associated with the operation data for each predetermined range and stored in the performance database. The operation data collected by the first collection unit is also stored in the performance database.

[0044] Here, the second collecting unit calculates φ based on, for example, the temperature change of the steel strip and the specifications of the continuous steel strip. cg The actual value is calculated, and the following formula (1) can be used as a specific calculation formula.

[0045] In the case of a continuous steel strip with a relatively thin thickness, where the temperature gradient in the thickness direction can be ignored, the temperature change of the continuous steel strip can be calculated by the overall heat absorption rate (φ cg ) can be calculated using the heat transfer model of Equation (1), where ρ is the specific gravity of the steel strip [kg / m 3 ]. C is the specific heat of the steel strip [kcal / (kg·K)]. h is the plate thickness [m]. T s is the average temperature within a given range of the continuous steel strip in the target zone [°C]. T w is the furnace temperature [℃]. cg is the overall heat absorption coefficient. σ is the Stefan-Boltzmann coefficient. Here, the target zone refers to the zone that is the target of calculation among multiple zones. φ cg The calculation of the (actual value) is performed for all of the multiple zones. Therefore, the target zone can be said to be any one zone, not a specific zone among the multiple zones.

[0046]

number

[0047] Also, the heat flux (q[kcal / (m 2 ·s)] can be calculated from the temperature gradient in the longitudinal direction of the continuous steel strip as shown in equation (2). s / dt can be calculated by dividing the temperature change of the continuous steel strip within a specified range in each zone by the transport time of the continuous steel strip in each zone. The temperature change of the continuous steel strip within a specified range in each zone is calculated as "the steel strip temperature at the exit of the zone - the steel strip temperature at the entry of the zone."

[0048]

number

[0049] From equations (1) and (2), the overall heat absorption rate (φ cg [Dimensionless quantity]) is calculated using equation (3).

[0050]

number

[0051] As described above, the learning unit generates a prediction model of the overall heat absorption rate used to calculate the steel strip temperature. cg A prediction model for calculating the (predicted value) is generated (or updated) at a predetermined update period or at a timing when the operating conditions are changed. If the update period or the timing when the operating conditions are changed is met (Yes in step S4), the learning unit generates a prediction model for the overall heat absorption rate (step S5). Steps S4 and S5 correspond to the learning step. As an example, the update period in step S4 may be the period from immediately after the actual operation data for each predetermined range along the entire length of the coil of continuous steel strip is collected until the actual operation data for each coil is stored in the actual database. As another example, the update period may be the period until the actual operation data for each specific portion of the coil (e.g., the tip, middle, or tail end) is stored. As yet another example, the update period may be a fixed period determined based on the processing capacity of the device that generates the prediction model, or may be determined based on the timing when the operating conditions are changed by the operator.

[0052] In this embodiment, a machine learning model (trained model) generated by a machine learning technique is used as a prediction model for the overall heat absorptance. The learning unit sets any one of the multiple zones as a target zone, and sequentially learns a machine learning model for the target zone for each predetermined range, with the overall heat absorptance of the target zone as an output variable and operational data for the target zone and a zone upstream of the target zone as input variables. The learning unit then generates a machine learning model for each zone. As described above, operational data is collected and accumulated in real time in the performance database. The learning unit acquires at least the furnace temperature, exhaust gas temperature, fuel flow rate, and air flow rate for each zone from the performance database and uses these as input variables for the machine learning model. That is, the input variables for the machine learning model include at least the furnace temperature, exhaust gas temperature, fuel flow rate, and air flow rate for each zone. The learning unit then uses the performance values ​​(performance data) of this operational data as input variables for the training data. Furthermore, the actual values ​​of the overall heat absorption rate corresponding to the input variables of the learning data (actual values ​​of at least the furnace temperature, exhaust gas temperature, fuel flow rate, and air flow rate) are used as output variables of the learning data. The learning unit can perform machine learning (sequential learning) using newly accumulated actual data at timings such as update cycles to update the machine learning model.

[0053] The machine learning method is not limited, and the machine learning model is generated using a machine learning method including, for example, Lasso regression, ridge regression, linear regression, local regression, principal component regression, PLS regression, and neural network.

[0054] Figure 3 shows the overall heat absorption rate (φ cg The machine learning model for the first zone includes the furnace temperature, exhaust gas temperature, fuel flow rate, and air flow rate (furnace temperature 1, exhaust gas temperature 1, fuel flow rate 1, and air flow rate 1), which are the heating conditions of the first zone, as input variables, and outputs the overall heat absorption rate (φ cg1 ) is the output variable.

[0055] The machine learning model for the second zone includes the heating conditions of the second zone (furnace temperature 2, exhaust gas temperature 2, fuel flow rate 2, and air flow rate 2) and the heating conditions of the upstream first zone as input variables, and estimates the overall heat absorption rate (φ cg2 ) is the output variable.

[0056] The machine learning model for the third zone includes the heating conditions of the third zone (furnace temperatures 3-1 to 3-5, exhaust gas temperature 3, fuel flow rate 3, air flow rate 3) and the heating conditions of the first and second upstream zones as input variables, and estimates the overall heat absorption rate (φ cg3 ) is the output variable. Furnace temperatures 3-1 to 3-5 correspond to the furnace temperatures measured by five furnace thermometers spaced apart in the conveying direction of the steel strip in the third zone.

[0057] Furthermore, as shown in FIG. 3, the input variables of the machine learning model for each zone may further include factors influencing the heat content of the steel strip entering each zone (the steel strip temperature and temperature change in the upstream zone). Furthermore, the input variables of the machine learning model for each zone may further include steel strip size information (thickness and width) and operating conditions other than heating conditions (the steel strip conveying speed). Here, the combustion gases from zones 1 to 3 may be combined into a single flue. In this case, the exhaust gas temperatures 1 to 3 may be the temperatures after being combined into a single flue (i.e., the same temperature).

[0058] The prediction unit predicts the steel strip temperature using the machine learning model for each zone (step S6). The prediction unit predicts the steel strip temperature at the outlet side of the continuous heating furnace for the continuous steel strip before it enters the continuous heating furnace. The prediction unit may further predict the temperature of the continuous steel strip while it is passing through the continuous heating furnace. Step S6 corresponds to the prediction step.

[0059] In detail, the prediction unit inputs the operation data of the target zone and the zone upstream of the target zone into the machine learning model of the target zone, and calculates the predicted value of the overall heat absorption rate of the target zone (φ cg (predicted value) is calculated. Here, the target zone refers to the zone that is the target of the calculation among multiple zones. φ cgThe calculation of the (predicted value) is performed for all of the multiple zones. Therefore, the target zone can be said to be any one zone, rather than a specific zone among the multiple zones. The prediction unit calculates the predicted value of the overall heat absorptance for each zone, and predicts the steel strip temperature at the outlet side of the continuous heating furnace using the predicted value of the overall heat absorptance for each zone.

[0060] Actual continuous heating furnaces are often divided into multiple zones, each using a different heating method (e.g., direct flame or radiant tube). The method disclosed herein matches the configuration of an actual continuous heating furnace, and a machine learning model for each zone is generated as described above, taking into account the heating conditions and exhaust gas temperature of adjacent upstream zones. This enables accurate calculations that match the actual situation. Furthermore, the machine learning model is updated through sequential learning, making it possible to determine an accurate value for the overall heat transfer coefficient of the continuous heating furnace during operation (a value that reflects changes in the state of the continuous heating furnace), thereby enabling sustained accurate prediction of the steel strip temperature.

[0061] The output unit outputs the calculation results including the steel strip temperature predicted by the prediction unit (Step S7). Step S7 corresponds to the output step.

[0062] The output unit may output, as the calculation results, the steel strip temperature at the outlet of the continuous heating furnace and the steel strip temperature at the outlet of each zone, predicted for the continuous steel strip before entering the continuous heating furnace. The calculation results may be displayed on a calculation result display monitor (an example of a display device) connected to the continuous steel strip temperature prediction device. The display device may also be configured to be connected to the prediction unit via a network. Here, if the predicted steel strip temperature falls outside a predetermined allowable range (the difference is large) with respect to a predetermined target temperature, not only may the calculation results be displayed but an alarm may also be output.

[0063] The control device may also control the continuous heating furnace based on the predicted steel strip temperature output from the temperature prediction system. The control device sets the fuel flow rate so that the predicted steel strip temperature becomes a target temperature.

[0064] The effects of the present disclosure will be specifically described below based on examples, but the present disclosure is not limited to these examples.

[0065] By the method of the above embodiment, the overall heat absorption rate (φ cg A machine learning model was created to predict φ (predicted value). cg A linear model was generated using Lasso regression, with the actual values ​​as the objective variable (output variable) and the operational data, including the heating conditions, as the explanatory variables (input variables). Scikit-learn, a Python machine learning library, was used to generate this machine learning model.

[0066] Figure 4 shows the top feature importance of the machine learning model for the third zone. Among the heating conditions, the feature importance of the furnace temperature in the third zone, the furnace temperature in the second zone, and the exhaust gas temperature is high. Here, the exhaust gas temperature is common to the first to third zones. Other than the heating conditions, the feature importance of the steel strip temperature change amount in the second zone is high. In contrast, among the heating conditions, the feature importance of the fuel flow rate and air flow rate is relatively low. Furthermore, the feature importance of the steel strip width and conveying speed is relatively low.

[0067] In machine learning, it is preferable to select appropriate input variables based on feature importance. For example, a threshold close to zero may be set, and input variables with feature importance below the threshold may not be incorporated into the model. Alternatively, Lasso regression may be applied to machine learning to bring the coefficients of input variables with low feature importance closer to zero, thereby selecting appropriate input variables based on feature importance. In this example, it is shown that the feature importance of the steel strip temperature change amount in the second zone is high. Since the method disclosed herein performs machine learning in each zone, it is reasonable that the feature importance of the steel strip temperature change amount is high.

[0068] Furthermore, this example shows that the feature importance of the exhaust gas temperature, which was not taken into account in the conventional method, is high. It is reasonable that the feature importance of the exhaust gas temperature is high because it is thought that the portion of the thermal energy generated in the continuous heating furnace that is not absorbed by the steel strip is represented by the exhaust gas temperature.

[0069] Figure 5 is a diagram illustrating the changes in exhaust gas temperature and overall heat absorptivity. The horizontal axis represents time. Approximately 2,400 seconds after the timing when the exhaust gas temperature suddenly drops (around 11,000 seconds), the overall heat transfer coefficient also drops, indicating a correlation between the exhaust gas temperature and the overall heat absorptivity. This corresponds to the high feature importance of the exhaust gas temperature.

[0070] Regarding fuel flow rate (gas flow rate), various parameters such as furnace temperature and steel strip temperature are fed back and controlled on the factory line, so it is thought that changes in fuel flow rate in other zones affect the heat transfer in the target zone. Therefore, taking into account fuel flow rate, which was not taken into account in conventional methods, is thought to contribute to improving prediction accuracy.

[0071] Figure 6 shows the prediction accuracy of the overall heat absorption rate of the third zone. The horizontal axis is φ cg (Actual value) and is shown as a calculated value. The vertical axis is φ cg (Predicted value) and is indicated as predicted value. cg (Actual value) is calculated using equation (3). The predicted value is output by the machine learning model. The plot shown by squares is the prediction accuracy using data other than the training data. Data other than the training data is verification data used to evaluate the general performance of the machine learning model. The plot shown by circles is the prediction accuracy using the training data. The training data is part of the data used to train the machine learning model. The overall heat absorption rate (φ cg3 It was confirmed that this example exhibited good prediction accuracy in predicting the above.

[0072] Also, φ cgA numerical simulation was performed to predict the steel strip temperature at the outlet of a continuous heating furnace using the (predicted value). In this simulation, operational data was collected at 2-second intervals, and the overall heat transfer coefficient was changed at that interval, with the simulation performed as a first-order lag. The overall heat absorption rate was set for each of zones 1 to 3, and the steel strip temperature at the outlet of the continuous heating furnace (steel strip temperature at the outlet of zone 3) was calculated based on the differential equation in equation (1).

[0073] FIG. 7A is a comparative example, and φ cg 7B shows the results of predicting the steel strip temperature when the conventional method is used, in which φ is constant. cg (predicted value) and φ cg The predicted results of the steel strip temperature based on (predicted value) are shown.

[0074] "Data" in Figures 7A and 7B indicates the steel strip temperature (actual steel strip temperature) on the actual line. "Sim" in Figures 7A and 7B indicates the predicted steel strip temperature by simulation. In the comparative example, the prediction error for the actual steel strip temperature was a maximum of approximately 13°C, resulting in a large error. In the example, the prediction error for the actual steel strip temperature was within the range of -5°C to +5°C, demonstrating good prediction accuracy.

[0075] As is clear from the comparison with the comparative example, the disclosed method is capable of predicting the steel strip temperature with high accuracy in accordance with fluctuations in operating conditions. Here, a short calculation cycle is preferable, but is not limited to the short cycle (2 seconds) used in the simulation of this embodiment. For example, it is preferable to recalculate the overall heat transfer coefficient when fluctuations in the line's operating conditions occur, such as when the steel strip passing through the continuous heating furnace is switched or when an operator manually changes the settings.

[0076] Although the embodiments of the present disclosure have been described based on the drawings and examples, it should be noted that those skilled in the art would easily be able to make various modifications or alterations based on the present disclosure. Therefore, it should be noted that these modifications and alterations are included within the scope of the present disclosure. For example, the functions included in each component or step can be rearranged so as not to cause logical inconsistencies, and multiple components or steps can be combined or divided into one. The embodiments of the present disclosure can also be realized as a program executed by a processor included in an apparatus or a storage medium on which a program is recorded. It should be understood that these are also included within the scope of the present disclosure.

Claims

1. A temperature prediction system for predicting a temperature of a material to be heated, which is a temperature of a material to be heated in a continuous heating furnace having a plurality of continuous zones, comprising: a collection unit that collects operational data including the temperature of the heated material for each predetermined range of the heated material in each zone, calculates an overall heat absorptivity of each zone for each predetermined range based on a change in the temperature of the heated material from the inlet side to the outlet side of each zone, and stores the overall heat absorptivity in association with the operational data for each predetermined range; a learning unit that sets any one zone among the plurality of zones as a target zone, sequentially learns a machine learning model of the target zone for each of the predetermined ranges, the machine learning model having the overall heat absorptivity of the target zone as an output variable and the operational data of the target zone and a zone upstream of the target zone as input variables, and generates the machine learning model for each of the zones; a prediction unit that inputs the operational data of the target zone and a zone upstream of the target zone into the machine learning model of the target zone, calculates a predicted value of the overall heat absorptance of the target zone, and predicts the temperature of the heated material at the outlet side of the continuous heating furnace using the predicted value of the overall heat absorptance of each zone; an output unit that outputs a calculation result including the predicted temperature of the material to be heated, The collected operational data includes at least the furnace temperature, exhaust gas temperature, fuel flow rate, air flow rate, inlet temperature of the heated material, and outlet temperature of the heated material in each zone, A temperature prediction system, wherein input variables of the machine learning model include at least a furnace temperature, an exhaust gas temperature, a fuel flow rate, and an air flow rate in each zone.

2. The temperature prediction system according to claim 1 , wherein the furnace temperature is measured at a plurality of positions spaced apart in a conveying direction of the material to be heated.

3. The temperature prediction system according to claim 1 or 2, wherein the machine learning model is generated using a machine learning technique including Lasso regression, ridge regression, linear regression, local regression, principal component regression, PLS regression, and neural network.

4. A control device that controls the continuous heating furnace based on the predicted temperature of the material to be heated output by the temperature prediction system according to claim 1 or 2.

5. A temperature prediction method executed by a temperature prediction system for predicting a temperature of a material to be heated, which is a temperature of a material to be heated in a continuous heating furnace having a plurality of continuous zones, comprising: a collection step of collecting operational data including the temperature of the heated material for each predetermined range of the heated material in each zone, calculating an overall heat absorptivity of each zone for each predetermined range based on a change in the temperature of the heated material from the inlet side to the outlet side of each zone, and storing the overall heat absorptivity in association with the operational data for each predetermined range; a learning step of sequentially learning a machine learning model of the target zone for each of the predetermined ranges, with any one of the plurality of zones as a target zone, the overall heat absorptivity of the target zone as an output variable, and the operational data of the target zone and a zone upstream of the target zone as input variables, to generate the machine learning model for each zone; a prediction step of inputting the operational data of the target zone and a zone upstream of the target zone into the machine learning model of the target zone, calculating a predicted value of the overall heat absorptance of the target zone, and predicting the temperature of the heated material at the outlet side of the continuous heating furnace using the predicted value of the overall heat absorptance of each zone; an output step of outputting a calculation result including the predicted temperature of the material to be heated, The collected operational data includes at least the furnace temperature, exhaust gas temperature, fuel flow rate, air flow rate, inlet temperature of the heated material, and outlet temperature of the heated material in each zone, A temperature prediction method, wherein input variables of the machine learning model include at least a furnace temperature, an exhaust gas temperature, a fuel flow rate, and an air flow rate in each zone.

6. A control method for controlling the continuous heating furnace based on the predicted temperature of the material to be heated output by the temperature prediction method according to claim 5.

Images