Processing device, parameter creation device, processing method, parameter creation method, and program

Abstract

Adaptively change the pre-configured parameters set in an input / output model that takes input variables as input and outputs output variables. [Solution] The processing unit 120 acquires parameter selection conditions, selects a pre-configured parameter that satisfies the acquired parameter selection conditions from among the candidates of pre-configured parameters stored in the storage medium, and applies the selected pre-configured parameter to the input / output model to derive the output variable.

Description

[Technical Field]

[0001] This disclosure relates to an processing apparatus, a parameter creation apparatus, a processing method, a parameter creation method, and a program. [Background technology]

[0002] Some input-output models, such as evaluation functions for solving optimization problems and pre-trained models using machine learning (e.g., neural networks), require pre-configured parameters to be set. These models take input variables as input variables and output variables as output variables. Pre-configured parameters are parameters that must be set as fixed values ​​within the input-output model in advance, separate from the input and output variables. An optimization problem is a problem in which the evaluation metrics included in the evaluation function are used as decision variables (solutions) to find a solution that minimizes or maximizes the value of the evaluation function. In an evaluation function for solving optimization problems, weight coefficients for multiple evaluation metrics included in the evaluation function are an example of pre-configured parameters. The weight coefficients are parameters used to balance the evaluation of multiple evaluation metrics and are determined in advance according to the relative importance and units of each evaluation metric. For example, if the optimization problem is to minimize the value of the evaluation function, the weight coefficients of evaluation metrics with relatively high importance are increased. In pre-trained models, hyperparameters are an example of pre-configured parameters.

[0003] One technique for deriving such pre-set parameters is described in Patent Document 1. Patent Document 1 discloses a method for searching for the optimal solution of weight coefficients for multiple evaluation indicators included in an evaluation function by performing inverse reinforcement learning, using the evaluation indicator when a desired result (e.g., operational result) is obtained as a given optimal value. [Prior art documents] [Patent Documents]

[0004] [Patent Document 1] Japanese Patent Publication No. 2023-039710 [Overview of the project] [Problems that the invention aims to solve]

[0005] However, the technology described in Patent Document 1 is limited to obtaining weight coefficients suitable for the desired result (e.g., operational results) at a specific point in time. Therefore, with the technology described in Patent Document 1, the weight coefficients found as the optimal solution at a certain point in time may not be appropriate at a later point in time. In this case, with the technology described in Patent Document 1, it is necessary to search again for the optimal weight coefficients that are suitable for that later point in time. Therefore, with the technology described in Patent Document 1, it is not easy to quickly derive weight coefficients that are suitable for that later point in time. Thus, for example, when using the technology described in Patent Document 1 to search for weight coefficients of an evaluation function used for control in manufacturing a product, it is not easy to adaptively change the weight coefficients in response to changes in manufacturing conditions. In this way, the technology described in Patent Document 1 has the problem that it is not possible to adaptively change the pre-set parameters set in an input / output model that takes input variables as input and outputs output variables as output.

[0006] This disclosure was made in view of the above-mentioned problems, and aims to adaptively change the pre-configured parameters set in an input / output model that takes input variables as input and outputs output variables. [Means for solving the problem]

[0007] The apparatus of the present disclosure is an apparatus that derives an output variable using an input / output model that takes an input variable as input and outputs an output variable, and comprises: a condition acquisition unit that acquires selection conditions for pre-set parameters that are set in advance for the input / output model in order to perform the derivation; a parameter selection unit that selects a pre-set parameter that satisfies the selection conditions from a plurality of candidate pre-set parameters stored in advance on a storage medium; and an output variable derivation unit that applies the pre-set parameter selected by the parameter selection unit to the input / output model and performs the derivation. The parameter creation device of this disclosure is a parameter creation device for deriving candidate pre-set parameters, comprising: a performance acquisition unit that acquires time series data of actual values ​​of at least one variable among the input variables and the output variables that can be measured; a division unit that divides the time series data into data for each of the multiple time intervals; a time interval selection unit that selects a time interval from the multiple time intervals that has data satisfying the selection conditions based on the data for each of the multiple time intervals; and a parameter candidate derivation unit that derives candidate pre-set parameters using the time series data included in the time interval selected by the time interval selection unit.

[0008] The processing method of the present disclosure is a processing method for deriving an output variable using an input / output model that takes an input variable as input and outputs an output variable, comprising: a condition acquisition step of acquiring selection conditions for pre-configured parameters that are set in advance for the input / output model in order to perform the derivation; a parameter selection step of selecting pre-configured parameters that satisfy the selection conditions from a plurality of candidates of pre-configured parameters that are stored in advance on a storage medium; and an output variable derivation step of applying the pre-configured parameters selected in the parameter selection step to the input / output model and performing the derivation. The parameter creation method of this disclosure is a parameter creation method for deriving candidate pre-set parameters, comprising: a performance acquisition step of acquiring time series data of actual values ​​of at least one variable among the input variables and the output variables that can be measured; a division step of dividing the time series data into data for each of a plurality of time intervals; a time interval selection step of selecting a time interval from the plurality of time intervals that has data that satisfies the selection conditions based on the data for each of the plurality of time intervals; and a parameter candidate derivation step of deriving candidate pre-set parameters using the time series data included in the time interval selected in the time interval selection step.

[0009] A first example of the program of this disclosure causes a computer to function as a part of the processing unit. A second example of the program of this disclosure involves a computer functioning as a component of the parameter creation device. [Effects of the Invention]

[0010] According to the present disclosure, as a preset parameter set in the input / output model, a preset parameter that satisfies the selection condition of the preset parameter is selected from candidates of the preset parameters stored in a storage medium in advance, so that the preset parameter can be adaptively changed.

Brief Description of the Drawings

[0011] [Figure 1] It is a diagram showing an example of the functional configuration of a parameter creation device and a processing device. [Figure 2] It is a diagram showing an example of input variable actual time series data. [Figure 3A] It is a diagram showing an example of the in-section average value of input variable actual time interval data that does not distinguish operators. [Figure 3B] It is a diagram showing an example of the in-section average value of input variable actual time interval data for each operator. [Figure 4] It is a diagram showing an example of the in-section average value after scaling of input variable actual time interval data that does not distinguish operators. [Figure 5] It is a diagram showing an example of the in-section average value after scaling of input variable actual time interval data when each operation is in operation. [Figure 6] It is a diagram showing an example of the interval data evaluation value when the parameter selection condition is a parameter selection condition independent of the operator. [Figure 7] It is a diagram for explaining an example of a time interval that satisfies a parameter selection condition independent of the operator. [Figure 8] It is a diagram showing an example of the overall interval average value of input variable actual time interval data when each operator is in operation. [Figure 9] It is a diagram showing an example of a time interval that satisfies a parameter selection condition determined for each operator. [Figure 10] It is a diagram showing an example of a weight coefficient. [Figure 11]This flowchart illustrates one example of a parameter creation method. [Figure 12] This is a flowchart illustrating one example of a processing method. [Figure 13] This figure shows an example of a system that shares pre-configured parameters across multiple locations. [Modes for carrying out the invention]

[0012] Hereinafter, an embodiment of this disclosure will be described with reference to the drawings. Note that the values ​​of the physical quantities to be compared, such as length, position, size, and spacing, are the same, and this includes not only cases where they are strictly the same, but also cases where they differ to the extent that they do not depart from the spirit of the invention. For example, even if multiple values ​​of a certain physical quantity differ within the tolerance range determined at the time of design, these multiple values ​​may be considered the same. Also, for example, if each of the multiple values ​​of a certain physical quantity is within the tolerance range, these multiple values ​​may be considered the same. In this case, these multiple values ​​may be measured values. If these multiple values ​​are measured values, the tolerance may be, for example, the maximum tolerance specified in the measuring instrument used to measure the physical quantity. Also, these multiple values ​​may be estimated values ​​(calculated values). If these multiple values ​​are estimated values ​​(calculated values), the tolerance may be the error that arises from the calculation of the physical quantity being performed by a computer. Also, the tolerance may be the error that arises from the modeling of the actual phenomenon using mathematical formulas.

[0013] Figure 1 shows an example of the functional configuration of the parameter creation device 110 and the processing device 120. The parameter creation device 110 and the processing unit 120 have, for example, one or more hardware processors such as a CPU (Central Processing Unit) and one or more memory such as RAM (Random Access Memory) and ROM (Read Only Memory) as hardware, and perform various calculations by executing one or more programs stored in memory using one or more hardware processors. Furthermore, the parameter creation device 110 and the processing unit 120 have input devices and output devices as hardware. At least one of the input devices and output devices may be located outside the parameter creation device 110 and the processing unit 120.

[0014] Furthermore, this embodiment illustrates a case where the parameter creation device 110 and the processing device 120 are implemented using separate hardware (devices). The parameter creation device 110 and the processing device 120 may be configured to communicate with each other. In this case, the parameter creation device 110 and the processing device 120 may communicate via wired communication or wireless communication. Alternatively, the parameter creation device 110 and the processing device 120 may communicate via a communication network. The communication network may include, for example, at least one of a LAN (Local Area Network), a WAN (Wide Area Network), and the Internet.

[0015] Furthermore, the functions of the parameter creation device 110 and the processing device 120 may be implemented, for example, by a single piece of hardware (the same hardware). Also, the hardware of the parameter creation device 110 and the processing device 120 may be implemented by dedicated hardware such as an ASIC (Application Specific Integrated Circuit).

[0016] The processing unit 120 derives output variables using an input / output model that takes input variables as input and outputs output variables. Pre-configured parameters are set for the input / output model. The processing unit 120 obtains the selection criteria for the pre-configured parameters. The processing unit 120 then selects pre-configured parameters that satisfy the selection criteria from among several candidate pre-configured parameters stored in the storage medium, applies the selected pre-configured parameters to the input / output model, and then derives output variables corresponding to those input variables by inputting the input variables into the input / output model. The selection criteria for the pre-configured parameters are defined using information about the application target of the input / output model, specifying the conditions that pre-configured parameters suitable for the input / output model should satisfy. The input / output model may be, for example, an evaluation function for solving an optimization problem, a trained model created by machine learning, or other input / output models (e.g., mathematical models). As mentioned above, an optimization problem is a problem in which the evaluation metric included in the evaluation function is used as the decision variable (solution), and the solution is sought that maximizes or minimizes the value of the evaluation function. A trained model created by machine learning is a neural network, etc.

[0017] Input variables are the variables that are input to an input / output model. Output variables are the variables that are output from the input / output model, and are derived by inputting the values ​​of the input variables into the input / output model. For example, in an evaluation function for solving an optimization problem, the design variables included in the evaluation metric are an example of input variables, and the variables derived by providing the values ​​of the design variables to the evaluation function are output variables (i.e., the value of the evaluation function is the value of the output variable). Also, in a trained model, the explanatory variables are an example of input variables, and the target variable derived by providing the explanatory variables to the trained model is an example of an output variable.

[0018] As mentioned above, pre-configured parameters are parameters that need to be set as fixed values ​​in advance for the input / output model, separate from the input and output variables. In an evaluation function for solving an optimization problem, pre-configured parameters are weight coefficients for multiple evaluation metrics. As previously mentioned, weight coefficients are parameters used to balance the evaluation of multiple evaluation metrics and are predetermined according to the relative importance and units of each evaluation metric. For example, if the optimization problem is to minimize the value of the evaluation function, the weight coefficients of evaluation metrics with relatively high importance are increased. In a trained model, pre-configured parameters are hyperparameters. Specifically, in a neural network, which is an example of a trained model, pre-configured parameters include, for example, the number of neurons, the number of layers, the drop rate, and the learning rate.

[0019] The applications of an input / output model are not limited. An input / output model may be used for processing (e.g., controlling) a product. Alternatively, an input / output model may be used for various estimations and predictions, such as product quality or state.

[0020] Furthermore, if the input / output model is an evaluation function for solving an optimization problem, the evaluation metrics included in the evaluation function should also be determined according to the purpose of the evaluation function. For example, the evaluation metrics may include evaluation metrics related to the items being processed. In this case, the evaluation metrics may include variables that represent physical quantities whose values ​​change as a result of the processing of the items. Examples of such physical quantities include at least one of the following: a variable that represents a physical quantity representing the state of the equipment used to process the items, and a variable that represents a physical quantity representing the state of the items. Examples of item processing include the transportation (logistics) of items and the manufacturing of products. Examples of physical quantities representing the state of the equipment used to process the items include physical quantities representing the state of the equipment that transports the items, and physical quantities representing the state of product manufacturing equipment or inspection equipment. Examples of physical quantities representing the state of the items include the temperature of the items. However, the evaluation metrics that constitute the evaluation function are not limited to evaluation metrics related to the items being processed, but can be any evaluation metrics (multiple input variables) that can construct an evaluation function.

[0021] As described above, the input / output model is not limited, but in this embodiment, similar to the specific example in Patent Document 1, an example of the parameter creation device 110 and processing device 120 will be described by illustrating the case where the input / output model is an evaluation function used to control the amount of heat input during non-steady-state operation of a coke oven. In the following description, a detailed explanation of what is described in Patent Document 1 will be omitted.

[0022] In this embodiment, we illustrate the case in which the value of the evaluation function J (goodness-of-fit function) in equation (1) is derived when the input / output model is equation (1) below and the processing unit 120 satisfies the constraint equation (2) below. Equations (1) and (2) are described in Patent Document 1.

[0023]

number

[0024] In equation (1), Tr(t) is the predicted furnace bed temperature at time t. The furnace bed temperature is a representative value of the temperature of all combustion chambers in the coke oven. Tr_ref is the target furnace bed temperature (target value of furnace bed temperature). Tc(t e ) is the end time of non-regular operations t e This is the coke temperature. The coke temperature is the temperature (predicted value) of the coke discharged from the carbonization chamber of the coke oven. Tc_sv is the target coke temperature. The target coke temperature is the target value of the coke temperature at the end of non-steady-state operation and is set according to the required quality of the coke, etc. Q(t) is the amount of heat input (predicted value) at time t. The amount of heat input is the amount of heat input (supplied) to the combustion chamber. The amount of heat input may be controlled by operating a single control valve installed for all combustion chambers, or by operating control valves installed for each combustion chamber. The control valves are operated via actuators (not shown) under the control of the processing device 120.

[0025] In equation (2), time1 is the time to reach the minimum furnace temperature. The time to reach the minimum furnace temperature is the time (hr) required for the furnace group temperature to change from the steady state furnace group temperature immediately before the start of transient operation to the minimum furnace group temperature during transient operation. time2 is the time to maintain the minimum furnace temperature. The time to maintain the minimum furnace temperature is the time (hr) required for the furnace group temperature to change from the minimum furnace group temperature during transient operation to the steady state furnace group temperature immediately after the end of transient operation. time0 is the duration of transient operation (hr).

[0026] The range for the summation of t in the first and third terms on the right-hand side of equation (1) is the start time t of the non-steady operation. s From end time t e This is within the range up to (t s ≦t≦t e ). The first term on the right-hand side of equation (1) is an evaluation index that evaluates the difference between the candidate target furnace temperature orbital Tr_ref and the predicted value of the furnace group temperature Tr(t). The second term on the right-hand side of equation (1) is the difference between the target coke temperature Tc_sv and the coke temperature Tc(t eThis is an evaluation index that evaluates the difference between the predicted value of ) and . The third term on the right side of equation (15) is an evaluation index that evaluates the input heat Q(t). In the following explanation, |Tr(t)-Tr_ref(t)| will be referred to as the furnace temperature error as needed, and |Tc(t e )-Tc_sv| is referred to as the coke temperature error as needed. Also, in equation (1), w1, w2, and w3 are weighting coefficients w1, w2, and w3, and are an example of a set of pre-set parameters.

[0027] In the example shown in equation (1), the smaller the value of each term on the right-hand side of equation (1), the higher the evaluation according to each evaluation index. Therefore, it is preferable that the value of the evaluation function J is as close to 0 as possible. That is, as will be described later, the processing unit 120 searches for the optimal solution for the design variables that minimize the value of the evaluation function J (in the range of 0 or greater) within the range that satisfies the constraint equation in equation (2). The design variables are the change in furnace temperature at the start of unsteady-state operation ΔTr1, the change in furnace temperature at the end of unsteady-state operation ΔTr2, the time to reach the minimum furnace temperature time1, and the time to maintain the minimum furnace temperature time2. The change in furnace temperature at the start of unsteady-state operation ΔTr1 is the change in furnace temperature (°C) from the furnace group temperature in the steady state immediately before the start of unsteady-state operation (= furnace group temperature at the start of unsteady-state operation) to the minimum furnace group temperature during unsteady-state operation. The change in furnace temperature ΔTr2 at the end of the unsteady operation is the change in furnace temperature (°C) from the lowest furnace bed temperature in the steady state immediately before the start of the unsteady operation to the furnace bed temperature in the steady state immediately after the end of the unsteady operation (= furnace bed temperature at the end of the unsteady operation).

[0028] Alternatively, the optimal solution may be found by searching for the design variables that maximize the value of the evaluation function. In this case, for example, the evaluation function is obtained by multiplying each term on the right-hand side of equation (1) by (-1). In this case, the processing unit 120 searches for the optimal solution for the design variables that maximize the value of the evaluation function within the range that satisfies the constraint equation of equation (2).

[0029] In addition, instead of or in addition to at least one of the evaluation indices shown in each term on the right side of the formula (1), other evaluation indices (for example, an evaluation index for evaluating the productivity of coke, etc.) may be included as evaluation indices in the evaluation function of the formula (1). The evaluation index for evaluating the productivity of coke is, for example, ton / hr (production amount (ton) per hour).

[0030] As described above, in the present embodiment, the input variables are the furnace temperature error |Tr(t) - Tr_ref(t)|, the coke temperature error |Tc(t e ) - Tc_sv|, and the input heat amount |Q(t)|, and the case where the evaluation function J (J on the left side of the formula (1)) is the output variable is exemplified.

[0031] [Parameter creation device 110] First, an example of the parameter creation device 110 will be described. In the present embodiment in FIG. 1, the case where the parameter creation device 110 includes a performance acquisition unit 111, a division unit 112, a time interval selection unit 113, a parameter candidate derivation unit 114, a storage unit 115, and a parameter output unit 116 is exemplified.

[0032] <Performance acquisition unit 111> The performance acquisition unit 111 acquires time-series data of the performance values of at least one variable that can be actually measured among the input variables and output variables in the input-output model. As described above, in the present embodiment, the case where the input-output model is an evaluation function (more specifically, the formula (1)) used for solving an optimization problem is exemplified. In this case, the evaluation function J (J on the left side of the formula (1)), which is the output variable, is not a variable that can be actually measured. On the other hand, the furnace mass temperature Tr(t), the coke temperature Tc(t e ) and the input heat amount Q(t) can be actually measured, so the furnace temperature error |Tr(t) - Tr_ref(t)|, the coke temperature error |Tc(t e)-Tc_sv| and input heat |Q(t)| are variables that can be measured. Therefore, in this embodiment, we will illustrate the case in which the performance acquisition unit 111 acquires time-series data of the actual values ​​of the furnace temperature error, the actual values ​​of the coke temperature error, and the actual values ​​of the input heat. In the following description, the time-series data of each actual value will be referred to as performance time-series data as needed. Also, in the following description, if we do not distinguish between the time-series data of the actual values ​​of the furnace temperature error, the actual values ​​of the coke temperature error, and the actual values ​​of the input heat, we will refer to them collectively as input variable performance time-series data as needed.

[0033] Figure 2 is a graph illustrating an example of actual time-series data for input variables. A wider time range ΔT for the actual time-series data of input variables is preferable because it includes data for coke production under various manufacturing conditions. However, if the time range ΔT of the actual time-series data of input variables is too wide, the computational load will increase. From this perspective, the time range ΔT of the actual time-series data of input variables is set in advance.

[0034] The performance acquisition unit 111 may acquire time-series data of input variables by, for example, reading time-series data of input variables from an external storage medium that stores operational performance data of a coke oven. Alternatively, the performance acquisition unit 111 may acquire time-series data of input variables by, for example, receiving time-series data of input variables transmitted from an external device. It is preferable for the performance acquisition unit 111 to acquire time-series data of all measurable variables among the input and output variables in the input / output model, but it may acquire time-series data of only some of the measurable variables. For example, it is not necessary to acquire time-series data of variables with a low measurement frequency. Furthermore, if the output variables can be measured, such as in a trained model that estimates physical quantities, the performance acquisition unit 111 may acquire time-series data of the output variables.

[0035] <Divided part 112> The splitting unit 112 divides the input variable actual time series data into data for multiple time intervals. Figure 2 illustrates the case where each of the input variable actual time series data (actual time series data of input heat 201, actual time series data of coke temperature error 202, and actual time series data of furnace temperature error 203) is divided into n time intervals 1 to n, which are integers of 2 or more. Figure 2 illustrates the case where all time intervals 1 to n have the same time width TL. However, the time width of at least one of the multiple time intervals 1 to n may differ from the time widths of the other time intervals. Also, Figure 2 illustrates the case where the multiple time intervals 1 to n are consecutive (i.e., the end time of the earlier time interval and the start time of the later time interval are the same for two time intervals that are adjacent in time). However, at least one of the multiple time intervals 1 to n may not be consecutive with at least one of the time intervals that are adjacent in time. Multiple time intervals 1 to n (time width TL) are pre-set so that the number of actual values ​​necessary to derive candidate pre-set parameters is included in one time interval. As will be described later, in this embodiment, we will illustrate a case in which pre-set parameters are derived by performing inverse reinforcement learning using the data in one time interval of the input variable actual time series data as the optimal training action sequence in order to derive candidate pre-set parameters. Therefore, in this embodiment, we will illustrate a case in which multiple time intervals 1 to n (time width TL) are pre-set so that the number of actual values ​​necessary to perform one inverse reinforcement learning is included in one time interval. In the following description, the data included in one time interval of the input variable actual time series data will be referred to as input variable actual time interval data as needed. Also, the data included in one time interval of the input heat quantity actual time series data, furnace temperature error actual time series data, and coke temperature error actual time series data will be referred to as input heat quantity actual time interval data, furnace temperature error actual time interval data, and coke temperature error actual time interval data as needed.

[0036] Inverse reinforcement learning assumes that an optimal training sequence of actions is given, and is a method that minimizes the objective function defined by the sequence of actions under the reward function and the optimal training sequence of actions. In other words, inverse reinforcement learning assumes that the optimal actions are given, and is a method that determines a reward function that minimizes the difference between the actions determined under the reward function and the pre-set optimal actions.

[0037] <Time interval selection unit 113> The time interval selection unit 113 selects a time interval from a plurality of time intervals 1 to n that satisfies the selection conditions for a pre-set parameter, based on the input variable actual time interval data. In the following description, the selection conditions for a pre-set parameter will be referred to as the parameter selection conditions as necessary. As will be described later, in this embodiment, we will illustrate a case in which the processing unit 120 selects a pre-set parameter that satisfies the parameter selection conditions from a list of candidate pre-set parameters as a pre-set parameter to be applied to the input / output model. The parameter selection conditions are set in advance according to the application of the input / output model. For example, if the input / output model includes an evaluation function for solving an optimization problem for formulating a transport plan, the parameter selection conditions may be determined based on the planning policy. The planning policy may include conditions relating to at least one of the following: the delivery date of the transported goods, the transport equipment that transports the transported goods, and the transport cost of the transported goods. For example, avoiding delays in the delivery date of the transported goods may be included in the planning policy. Also, for example, reducing the number of transports by the transport equipment, shortening the travel distance of the transport equipment, or increasing the amount of transported goods that the transport equipment transports at one time may be included in the planning policy. Furthermore, reducing the transportation costs of the transported items may be included in the planning guidelines.

[0038] In this embodiment, when the derived results of an input / output model are used for control during the production of a product (coke), the parameter selection conditions may be determined based on the product manufacturing conditions. The product manufacturing conditions are conditions imposed during the production of the product, and may be conditions that are artificially determined or conditions that cannot be artificially determined (e.g., weather conditions). The product manufacturing conditions may include, for example, the product manufacturing policy and the manufacturing environment. Specifically, the product manufacturing conditions may include conditions relating to at least one of the following: product quality, product manufacturing cost, product productivity, manufacturing equipment used for manufacturing, raw materials used for manufacturing, and the manufacturing environment. For example, improving product quality may be included in the product manufacturing conditions. Also, lowering the product manufacturing cost may be included in the product manufacturing conditions. Also, improving the control performance of the control for manufacturing the product may be included in the product production conditions. Also, increasing product productivity may be included in the product manufacturing conditions. Also, lowering the equipment load of the manufacturing equipment used for manufacturing may be included in the product manufacturing conditions. Also, using specific raw materials (e.g., coal with a high moisture content) as raw materials used for manufacturing may be included in the product manufacturing conditions. Furthermore, the manufacturing conditions may include manufacturing in a specific environment (for example, a high-temperature environment). The manufacturing conditions may also include operating in the same manner as when a specific operator (for example, a skilled operator) is operating. It should be noted that even skilled operators may have strengths and weaknesses, so it is preferable that specific operators are distinguished based on these strengths and weaknesses. Additionally, multiple manufacturing policies may be used to define the manufacturing conditions. For example, the manufacturing conditions may include balancing product quality with product manufacturing costs.

[0039] In this embodiment, the parameter selection conditions are determined based on the coke production conditions, and an example is given where the coke production conditions include improving the quality of the coke, lowering the energy cost during coke production, improving the control performance of the control system for coke production, and operating in the same manner as when an operator is operating the system.

[0040] Specifically, an example is given where the parameter selection criteria include selecting a pre-set parameter that maximizes coke quality, selecting a pre-set parameter that minimizes energy costs during coke production, selecting a pre-set parameter that maximizes the control performance of the control system for coke production, selecting a pre-set parameter that can balance two or three of the following: coke quality, energy costs during coke production, and control performance of the control system for coke production, and selecting a pre-set parameter that allows the system to operate in the same way as an operator would.

[0041] If the parameter selection conditions include selecting pre-configured parameters that allow the operator to perform operations in the same manner as when an operator is performing operations, the performance acquisition unit 111 may acquire information that identifies which operator performed the operations at each time point in the input variable performance time series data. Furthermore, the division unit 112 may divide the time intervals 1 to n at the timing when operators change.

[0042] In this embodiment, the time interval selection unit 113 derives characteristic values ​​(characteristic values ​​for each time interval 1 to n of the actual time series data of the input variable) in multiple time intervals 1 to n, and based on the derived characteristic values, selects a time interval from the multiple time intervals 1 to n that contains data satisfying the parameter selection conditions, as an example of this process. The characteristic values ​​of the actual time interval data of the input variable can be any value that indicates the characteristics of the actual time interval data of the input variable, for example, a representative value (mean (derived mean), median, mode, etc.), or the absolute value of the difference between the maximum and minimum values. In this embodiment, the case where the characteristic value of the actual time interval data of the input variable is the mean is given as an example. In this way, the characteristics of the actual time interval data of the input variable in multiple time intervals 1 to n can be represented without performing complex calculations. In the following description, the mean value when the characteristic value of the actual time interval data of the input variable is the mean will be referred to as the mean value within the interval, as needed.

[0043] In this embodiment, an example is given in which the time interval selection unit 113 derives the average value within the interval of the actual time interval data for the input heat, the average value within the interval of the actual time interval data for the furnace temperature error, and the average value within the interval of the actual time interval data for the coke temperature error for each of the multiple time intervals 1 to n.

[0044] For example, due to disturbances or abnormal operation of equipment, the characteristic values ​​(in this embodiment, the average value within the interval) of the input variable actual time series data may not be appropriate. Also, from the perspective of product specifications, manufacturing costs, and optimizing, simplifying, and stabilizing operations, it may be preferable to set the values ​​of the input variables (heat input, coke temperature error, furnace temperature error) within an acceptable range. In this embodiment, an example is given in which the time interval selection unit 113 excludes characteristic values ​​from the input variable actual time series data that are outside a preset acceptable range from the above perspectives.

[0045] Figure 3A shows an example of the average value within an interval of actual time interval data for input variables without distinguishing between operators. Figure 3B shows an example of the average value within an interval of actual time interval data for input variables for each operator. In Figures 3A and 3B, for the sake of simplicity, the average value within an interval of actual time interval data for input heat is used as the vertical axis and the average value within an interval of actual time interval data for furnace temperature error is used as the horizontal axis, representing the average value within an interval of input variables as points on a two-dimensional Cartesian coordinate system. In the examples shown in Figures 3A and 3B, the average value within an interval of actual time interval data for input variables that does not fall within the tolerance range ΔH of the actual time interval data for input heat and the tolerance range ΔC of the actual time interval data for furnace temperature error (white circles) are excluded.

[0046] In Figure 3B, the average value of the input variable's actual interval data within the dashed line 321a represents the average value of the input variable's actual interval data when operator A performs the operation. Similarly, the average value of the input variable's actual interval data within the dashed line 321b represents the average value of the input variable's actual interval data when operator B performs the operation. Furthermore, the average value of the input variable's actual interval data within the dashed line 321c represents the average value of the input variable's actual interval data when operator C performs the operation. By setting an acceptable range for the average value of the input variable's actual interval data, it is possible to exclude, for example, the average value of the input variable's actual interval data when an operator with relatively lower skill (operator C) performs the operation, based on objective indicators.

[0047] In this embodiment, we illustrate a case in which the time interval selection unit 113 selects a time interval that satisfies the parameter selection condition from a plurality of time intervals 1 to n, based on the average values ​​within the interval of the input variable actual time interval data, excluding the average values ​​within the interval that have been excluded as described above, for each of the parameter selection conditions.

[0048] As described above, this embodiment exemplifies a case where the parameter selection conditions include selecting a pre-set parameter that maximizes coke quality, selecting a pre-set parameter that minimizes energy costs during coke production, selecting a pre-set parameter that maximizes the control performance of the control system for coke production, selecting a pre-set parameter that can balance two or three of the following: coke quality, energy costs during coke production, and control performance of the control system for coke production, and selecting a pre-set parameter that allows the system to operate in the same manner as when an operator is operating it.

[0049] When balancing two or three of the following: coke quality, energy cost during coke production, and control performance of the control system for coke production, it is preferable to treat coke quality, energy cost during coke production, and control performance of the control system for coke production equally. That is, it is preferable to standardize the units of coke quality, energy cost during coke production, and control performance of the control system for coke production (for example, by making them dimensionless). For this reason, in this embodiment, we illustrate a case in which the time interval selection unit 113 scales the in-interval average values ​​of the actual time interval data of the input variables in multiple time intervals 1 to n for each input variable (heat input, furnace temperature error, coke temperature error), and then selects a time interval that satisfies the parameter selection conditions from the multiple time intervals 1 to n based on the scaled in-interval average values ​​of the actual time interval data of the input variables in multiple time intervals 1 to n, for each parameter selection condition. Note that scaling refers to converting the units of multiple values ​​that have different units to standardize those units. Scaling can be achieved, for example, by standardization or normalization. Standardization is the process of transforming data values ​​so that the mean becomes 0 and the variance becomes 1. Normalization, on the other hand, is the process of changing data values ​​so that the maximum value becomes 1 and the minimum value becomes 0.

[0050] Figure 4 shows an example of scaled in-interval mean values ​​for input variable actual time interval data without distinguishing between operators. Figures 5(a) and 5(b) show examples of scaled in-interval mean values ​​for input variable actual time interval data when operators A and B performed operations, respectively. In the following explanation, unless otherwise specified, the in-interval mean values ​​for input variable actual time interval data are assumed to be scaled.

[0051] In this embodiment, we illustrate a case in which the time interval selection unit 113 selects a time interval containing input variable time interval data that satisfies the parameter selection conditions, using the average value within the interval of the actual input variable time interval data derived as described above, for each parameter selection condition. Specifically, in this embodiment, we illustrate a case in which the time interval selection unit 113 derives an evaluation value for each time interval 1 to n to evaluate whether the average value within the interval of the actual input variable time interval data satisfies the parameter selection conditions, and selects a time interval containing input variable time interval data that satisfies the parameter selection conditions based on the derived evaluation value. In the following description, this evaluation value will be referred to as the interval data evaluation value as necessary.

[0052] First, we will explain an example of a method for selecting time intervals containing input variable time interval data that satisfy the parameter selection conditions, using the mean values ​​within the time interval data of the actual input variable, when the parameter selection conditions are other than selecting pre-set parameters that allow operation to be performed in the same way as when an operator is performing the operation. In the following explanation, parameter selection conditions other than selecting pre-set parameters that allow operation to be performed in the same way as when an operator is performing the operation will be referred to as operator-independent parameter selection conditions, as necessary. Also, selecting pre-set parameters that allow operation to be performed in the same way as when an operator is performing the operation will be referred to as operator-specific parameter selection conditions, as necessary.

[0053] As described above, this embodiment exemplifies a case where operator-independent parameter selection conditions include selecting a preset parameter that maximizes coke quality, selecting a preset parameter that minimizes energy costs during coke production, selecting a preset parameter that maximizes the control performance of the control system for coke production, and selecting a preset parameter that can balance two or three of the following: coke quality, energy costs during coke production, and the control performance of the control system for coke production.

[0054] In this embodiment, we illustrate a case in which the time interval selection unit 113 derives interval data evaluation values ​​for each of the multiple parameter selection conditions and for each of the multiple intervals 1 to n using the following equation (3). Interval data evaluation value = (Coke temperature error selection coefficient × Coke temperature error average value within the interval + Input heat quantity selection coefficient × Input heat quantity average value within the interval + Furnace temperature error selection coefficient × Furnace temperature error average value within the interval) ÷ ​​(Coke temperature error selection coefficient + Input heat quantity selection coefficient + Furnace temperature error average value within the interval) ... (3)

[0055] The average value of coke temperature error within a given period is the average value of the actual data for the coke temperature error within that period (see Figure 4). The average value of heat input within a given period is the average value of the actual data for the heat input within that period (see Figure 4). The average value of furnace temperature error within a given period is the average value of the actual data for the furnace temperature error within that period (see Figure 4).

[0056] The coke temperature error selection coefficient is a 0-1 variable, which is 1 if the parameter selection conditions include a condition regarding coke quality, and 0 if the parameter selection conditions do not include a condition regarding coke quality. The input heat selection coefficient is a 0-1 variable, which is 1 if the parameter selection conditions include a condition regarding energy cost, and 0 if the parameter selection conditions do not include a condition regarding energy cost. The coke temperature error selection coefficient is a 0-1 variable, which is 1 if the parameter selection conditions include a condition regarding the control performance of the control system for producing coke, and 0 if the parameter selection conditions do not include a condition regarding the control performance of the control system for producing coke.

[0057] For example, if the parameter selection condition is to select pre-set parameters that maximize coke quality, the interval data evaluation value is the average value of the coke temperature error within the interval (= (1 × average value of coke temperature error within the interval + 0 × average value of input heat within the interval + 0 × average value of furnace temperature error within the interval) ÷ ​​(1 + 0 + 0)). Furthermore, if the parameter selection condition is to select pre-set parameters that can balance coke quality and the energy cost during coke production, the interval data evaluation value is (average value of coke temperature error within the interval + average value of input heat within the interval) ÷ ​​2 (= 1 × average value of coke temperature error within the interval + 1 × average value of input heat within the interval + 0 × average value of furnace temperature error within the interval) ÷ ​​(1 + 1 + 0)).

[0058] Figure 6 shows an example of interval data evaluation values ​​when the parameter selection conditions are operator-independent. In Figure 6, the "Best Quality" evaluation value represents the interval data evaluation value when the parameter selection condition is to select pre-set parameters that maximize coke quality. The "Best Energy Saving" evaluation value represents the interval data evaluation value when the parameter selection condition is to select pre-set parameters that minimize the energy cost during coke production. The "Balanced Quality & Energy Saving" evaluation value represents the interval data evaluation value when the parameter selection condition is to select pre-set parameters that balance coke quality and the energy cost during coke production. The "Balanced Quality, Energy Saving & Control Performance" evaluation value represents the interval data evaluation value when the parameter selection condition is to select pre-set parameters that balance coke quality, the energy cost during coke production, and the control performance of the controls for coke production.

[0059] The time interval selection unit 113 sorts the interval data evaluation values ​​of multiple time intervals 1 to n and selects the time interval having the interval data evaluation value that shows the highest evaluation value for each of the multiple parameter selection conditions. In this embodiment, an example is given where a larger interval data evaluation value indicates a higher evaluation. In this case, the time interval selection unit 113 selects the time interval with the maximum interval data evaluation value for each of the multiple parameter selection conditions. Alternatively, the time interval selection unit 113 may select multiple time intervals in descending order of interval data evaluation value for each of the multiple parameter selection conditions.

[0060] Figure 7 illustrates an example of a time interval that satisfies an operator-independent parameter selection condition. Figure 7 shows the result of sorting the interval data evaluation values ​​in descending order when the parameter selection condition is an operator-independent parameter selection condition. Figure 7 illustrates the case where the interval data evaluation value is maximized in time interval 13 when the parameter selection condition is to select a pre-set parameter that maximizes coke quality. In this case, the time interval with input variable time interval data that satisfies the selection condition for that parameter is time interval 13. Note that in Figure 3A, the interval data evaluation value derived using the in-interval mean value 311a of the input variable actual time interval data corresponds to the interval data evaluation value in time interval 13.

[0061] Furthermore, Figure 7 illustrates a case where the interval data evaluation value is maximized in time interval 103, when the parameter selection condition is to select a pre-set parameter that minimizes the energy cost during coke production. In this case, the time interval that has input variable time interval data satisfying the parameter selection condition is time interval 103. Note that in Figure 3A, the interval data evaluation value derived using the in-interval mean value 311b of the input variable actual time interval data corresponds to the interval data evaluation value in time interval 103.

[0062] The time interval selection unit 113 similarly selects time intervals that have input variable time interval data that satisfy other operator-independent parameter selection conditions. In Figure 3A, the interval data evaluation value derived using the in-interval average value 311c of the actual input variable time interval data corresponds to the interval data evaluation value in time intervals that have input variable time interval data that satisfy the parameter selection condition of selecting pre-set parameters that can balance coke quality and energy costs during coke production.

[0063] Next, we will explain an example of a method for selecting time intervals that have input variable time interval data that satisfy the parameter selection conditions, using the mean value within the time interval data of the input variable actual time interval data, when the parameter selection conditions are determined for each operator.

[0064] In this embodiment, we illustrate a case where the time interval selection unit 113 derives representative values ​​for each time interval 1 to n of the average value within the interval of the input variable's actual time interval data, for each operator and for each input variable. In the examples shown in Figures 5(a) and 5(b), the time interval selection unit 113 derives representative values ​​for each column. The representative value may be the mean (arithmetic mean), the median, the mode, or any other representative value. In this embodiment, we illustrate a case where this representative value is the mean. In the following description, this mean will be referred to as the average value of the entire interval of the input variable's actual time interval data, as needed.

[0065] Figures 8(a) and 8(b) show examples of the average values ​​over the entire interval of the input variable performance interval data when operators A and B perform operations, respectively. In this embodiment, we illustrate a case where the time interval selection unit 113 derives the distance between the in-interval mean of the actual time interval data of the input variable in time interval m (1 ≤ m ≤ n) and the overall interval mean of the actual time interval data of the input variable as an interval data evaluation value. In the following description, this distance will be referred to as the distance from the overall interval mean of time interval m, as needed. In Figure 3B, the distance between the overall interval mean values ​​322a and 322b (black circles) of the actual time interval data of the input variable when operators A and B perform operations and the in-interval mean values ​​(white circles) of the actual time interval data of the input variable within the dashed lines 321a and 321b corresponds to the distance from the overall interval mean of time interval m.

[0066] In this embodiment, we illustrate a case where the time interval selection unit 113 selects the time interval closest in distance from the overall time interval m average from the average value within the input variable's actual time interval data, for each operator and each input variable.

[0067] Specifically, in this embodiment, we will illustrate a case in which the time interval selection unit 113 derives the distance from the average of the entire time interval m for each operator A and B and for each input variable (coke temperature error, heat input amount, furnace temperature error) using the following equation (4). Distance from the average over the entire time interval m = SQRT{(Square distance of coke temperature error over time interval m + Square distance of heat input over time interval m + Square distance of furnace temperature error over time interval m)} 1 / 2 ...(4)

[0068] Here, the squared distance of the coke temperature error in time interval m, the squared distance of the heat input in time interval m, and the squared distance of the furnace temperature error in time interval m are derived by equations (5a), (5b), and (5c), respectively. The squared distance of the coke temperature error over time interval m = (the average value of the coke temperature error within the actual time interval data for time interval m - the average value of the coke temperature error over the entire time interval data) 2 ...(5a) The squared distance of the heat input over time interval m = (the average value of the actual heat input data within time interval m - the average value of the actual heat input data over the entire time interval) 2 ...(5b) The squared distance of the furnace temperature error over time interval m = (the average value of the actual time interval data for the furnace temperature error in time interval m - the average value of the actual time interval data for the furnace temperature error over the entire time interval) 2 ...(5c)

[0069] The time interval selection unit 113 derives the distance of time interval m from the overall mean for each of m=1 to n, as described above, for each parameter selection condition (for each operator A and B). Then, the time interval selection unit 113 derives the time interval that minimizes the distance of time interval m from the overall mean for each parameter selection condition (for each operator A and B). The time interval selection unit 113 then uses the time interval derived in this way as the time interval having input variable time interval data that satisfies the parameter selection conditions determined for each operator.

[0070] Figure 9 shows an example of a time interval that satisfies the parameter selection conditions determined for each operator. In Figure 9(a), the time interval with input variable time interval data that satisfies the parameter selection condition of selecting pre-set parameters that allow operation to be performed in the same manner as when operator A is performing the operation is time interval 26. In Figure 9(b), the time interval with input variable time interval data that satisfies the parameter selection condition of selecting pre-set parameters that allow operation to be performed in the same manner as when operator B is performing the operation is time interval 115.

[0071] Furthermore, the time interval selection unit 113 does not have to select a time interval that has input variable time interval data satisfying the parameter selection conditions in the manner described above. For example, the time interval selection unit 113 may select a time interval that has input variable time interval data satisfying the parameter selection conditions by using a trained model in which all of the input variable actual time interval data for multiple time intervals 1 to n and the parameter selection conditions are used as explanatory variables, and one of the multiple time intervals 1 to n is used as the target variable.

[0072] <Parameter candidate derivation unit 114> The parameter candidate derivation unit 114 derives candidate pre-set parameters using the actual time interval data of the input variables included in the time interval selected by the time interval selection unit 113. In this embodiment, we illustrate a case in which the parameter candidate derivation unit 114 derives candidate pre-set parameters by performing inverse reinforcement learning using the actual time interval data of the input variables included in the time interval selected by the time interval selection unit 113 as training data. In this embodiment, we illustrate the method described in Patent Document 1 as the method for inverse reinforcement learning. As mentioned above, in this embodiment, we illustrate a case in which the input / output model has the same evaluation function as the evaluation function described in Patent Document 1 (see equation (1)). Therefore, when performing inverse reinforcement learning as described in Patent Document 1, the method differs from that described in Patent Document 1 in that, as the optimal value of the evaluation index, the actual time interval data of the input variables included in the time interval selected for each parameter selection condition by the time interval selection unit 113 is used to derive candidate weight coefficients w1 to w3, which are examples of pre-set parameters, for each parameter selection condition. Therefore, this document will only provide an overview of the method for deriving candidate weight coefficients w1 to w3, which are an example of pre-configured parameters, and will omit a detailed explanation.

[0073] To determine candidate weight coefficients w1~w3 using inverse reinforcement learning, first, initial values ​​for weight coefficients w1~w3 are given to the right-hand side of the evaluation function J in equation (1), and the calculated furnace group temperature Tr_sim(t) and coke temperature Tc_sim(t) are obtained. e Then, derive the calculated value Q(t) of the heat input. Tr_sim(t), Tc_sim(t e Q_sim(t) is the derived values ​​of the furnace group temperature, coke temperature, and input heat when the value of the evaluation function J in equation (1) is minimized within the range that satisfies the constraint equation in equation (2). Then, we derive the difference evaluation function J_dif from equation (6) below.

[0074]

number

[0075] Here, Tr_opt(t) is the actual time interval data of the furnace group temperature included in the time interval selected by the time interval selection unit 113 as a time interval that satisfies the parameter selection conditions. Tc_opt(t e Q_opt(t) is the actual time interval data of coke temperature included in the time interval selected by the time interval selection unit 113. Q_opt(t) is the actual time interval data of heat input included in the time interval selected by the time interval selection unit 113.

[0076] Then, if the value of the difference evaluation function J_dif does not satisfy the predetermined conditions, the weight coefficients w1~w3 are updated, and the derivation of the difference evaluation function J_dif is performed again using the updated weight coefficients w1~w3. That is, the derivation of the difference evaluation function J_dif is repeated by changing the weight coefficients w1~w3 until the value of the difference evaluation function J_dif satisfies the predetermined conditions. Then, Tr_sim(t) and Tc_sim(t) are obtained when the value of the difference evaluation function J_dif satisfies the predetermined conditions. e The weight coefficients w1 to w3 set for Q_sim(t) are considered the optimal solution for weight coefficients w1 to w3.

[0077] The parameter candidate derivation unit 114 derives the optimal solution for the weight coefficients w1 to w3 as described above for each parameter selection condition. As a result, for each parameter selection condition, the weight coefficients w1 to w3 used in the input / output model (equation (1)) are obtained as candidates for the weight coefficients w1 to w3.

[0078] The parameter candidate derivation unit 114 does not necessarily have to derive candidate pre-set parameters (weight coefficients w1 to w3 in this embodiment) in the manner described above. For example, the parameter candidate derivation unit 114 may use the actual time interval data of the input variable included in the time interval selected by the time interval selection unit 113 as training data, and derive pre-set parameters that can reproduce the actual time interval data of the input variable through trial and error.

[0079] <Storage section 115> The storage unit 115 stores the candidate pre-set parameters derived by the parameter candidate derivation unit 114 in a storage medium. In this embodiment, the storage unit 115 is shown as storing candidate weight coefficients w1 to w3 (weight coefficients w1 to w3 for each of the multiple parameter conditions) in a storage medium inside the parameter creation device 110. However, the storage medium that stores the candidate weight coefficients w1 to w3 (weight coefficients w1 to w3 for each of the multiple parameter conditions) may be an external storage medium to the parameter creation device 110.

[0080] Figure 10 shows an example of weight coefficients w1 to w3 for each parameter condition. Figure 10(a) shows an example of weight coefficients w1 to w3 when the parameter selection condition is operator-independent. Figure 10(b) shows an example of weight coefficients w1 to w3 when the parameter selection condition is determined for each operator.

[0081] <Parameter output section 116> The parameter output unit 116 outputs the pre-set parameters stored by the storage unit 115 to the outside of the parameter creation device 110. In this embodiment, an example is given in which the parameter output unit 116 transmits the weight coefficients w1 to w3 for each parameter condition to the processing unit 120.

[0082] [Processing device 120] Next, an example of the processing unit 120 will be described. The processing unit 120 derives output variables using an input / output model that takes input variables as input and outputs output variables. In doing so, the processing unit 120 selects candidate pre-configured parameters that satisfy the parameter selection conditions from among the candidate pre-configured parameters created by the parameter creation device 110 and applies them to the input / output model.

[0083] In Figure 1, this embodiment illustrates a case where the processing device 120 comprises a condition acquisition unit 121, a parameter selection unit 122, a data acquisition unit 123, a predicted value derivation unit 124, a target furnace temperature determination unit 125, and a control unit 126.

[0084] <Condition acquisition unit 121> The condition acquisition unit 121 acquires the selection conditions (parameter selection conditions) for pre-configured parameters to be applied to the input / output model.

[0085] As described above, in this embodiment, the parameter selection conditions include selecting a preset parameter that maximizes coke quality, selecting a preset parameter that minimizes energy costs during coke production, selecting a preset parameter that maximizes the control performance of the control system for coke production, selecting a preset parameter that can balance two or three of the following: coke quality, energy costs during coke production, and control performance of the control system for coke production, and selecting a preset parameter that allows the system to operate in the same manner as when an operator is operating it.

[0086] In this embodiment, an example is given in which the condition acquisition unit 121 acquires one of these parameter selection conditions. For example, an operator of the processing unit 120 may input information specifying one of the candidates for a plurality of parameter selection conditions to the processing unit 120 by operating an input device. Alternatively, information specifying the parameter selection conditions may be transmitted to the processing unit 120 from an external device. In these cases, the condition acquisition unit 121 may acquire the selection conditions for pre-set parameters (weight coefficients w1 to w3) to be applied to the input / output model based on the information.

[0087] Furthermore, the condition acquisition unit 121 may acquire parameter selection conditions by deriving them using information that allows for the determination of selection conditions for pre-set parameters (weight coefficients w1 to w3). For example, parameter selection conditions may be classified into parameter selection conditions determined based on manufacturing conditions that depend on the manufacturing policy, parameter selection conditions determined based on manufacturing conditions that depend on disturbance factors (external environment), and parameter selection conditions determined based on manufacturing conditions that depend on the operator. In the example described above, for example, parameter selection conditions determined based on manufacturing conditions relating to product quality, product manufacturing cost, control performance of the control for manufacturing the product, product productivity, manufacturing equipment used for manufacturing, and raw materials used for manufacturing may be considered parameter selection conditions determined based on manufacturing conditions that depend on the manufacturing policy. Also, parameter selection conditions determined based on manufacturing conditions relating to raw materials stored exposed to the external environment and the manufacturing environment may be considered parameter selection conditions determined based on manufacturing conditions that depend on disturbance factors (external environment). Furthermore, a manufacturing condition that the operation is carried out in the same manner as when a specific operator (e.g., a skilled operator) is operating may be considered a parameter selection condition determined based on manufacturing conditions that depend on the operator.

[0088] In this case, information regarding the current manufacturing policy may be transmitted from an external device (a device that manages the operation of the coke oven) to the processing unit 120. The condition acquisition unit 121 may then select one parameter selection condition from among the parameter selection conditions determined based on the manufacturing conditions that depend on the manufacturing policy. For example, if the information regarding the current manufacturing policy includes prioritizing quality, the condition acquisition unit 121 may select a parameter selection condition from among the parameter selection conditions determined based on the manufacturing conditions that depend on the manufacturing policy, which is to select a pre-set parameter that maximizes the quality of the coke.

[0089] Furthermore, information regarding the operator's operational performance may be transmitted to the processing unit 120 from an external device (a device that manages the operation of the coke oven). In this case, the condition acquisition unit 121 may select one parameter selection condition from among the parameter selection conditions determined based on the manufacturing conditions that depend on the operator. For example, if the information regarding the operator's operational performance includes that operator A has a high performance in coke quality, the condition acquisition unit 121 may select a parameter selection condition from among the parameter selection conditions determined based on the manufacturing conditions that depend on the operator, which is a pre-set parameter that allows the operation to be performed in the same manner as when operator A is performing the operation.

[0090] Furthermore, weather information may be transmitted to the processing unit 120 from an external device (a device that manages the operation of the coke oven). In this case, the condition acquisition unit 121 may select one parameter selection condition from among the parameter selection conditions determined based on manufacturing conditions related to the manufacturing environment. For example, if the weather information includes information indicating that it is raining, the condition acquisition unit 121 may select a parameter selection condition that selects pre-set parameters suitable for operation during rainfall from among the parameter selection conditions determined based on manufacturing conditions that depend on the external environment.

[0091] In the above case, the condition acquisition unit 121 may select parameter selection conditions based on two or more manufacturing conditions, including manufacturing conditions dependent on the manufacturing policy, manufacturing conditions dependent on the operator, and manufacturing conditions dependent on the external environment. In this case, the condition acquisition unit 121 may acquire the parameter selection conditions specified by the operator of the processing unit 120 from among these two or more parameter selection conditions as selection conditions for pre-set parameters to be applied to the input / output model.

[0092] <Parameter selection section 122> The parameter selection unit 122 selects a pre-set parameter that satisfies the parameter selection conditions acquired by the condition acquisition unit 121 from among a plurality of pre-set parameter candidates stored in the storage medium. In this embodiment, an example is given in which the parameter selection unit 122 selects the weight coefficients w1 to w3 that satisfy the parameter selection conditions acquired by the condition acquisition unit 121 from the candidate weight coefficients w1 to w3 (weight coefficients w1 to w3 for each parameter condition) created by the parameter creation device 110, as the weight coefficients w1 to w3 in equation (1).

[0093] The parameter selection unit 122 may obtain weight coefficients w1 to w3 for all parameter conditions from the parameter creation device 110 (parameter output unit 116) and select from them weight coefficients w1 to w3 that satisfy the parameter selection conditions obtained by the condition acquisition unit 121. Alternatively, the parameter selection unit 122 may request only the weight coefficients w1 to w3 that satisfy the parameter selection conditions obtained by the condition acquisition unit 121 from the parameter creation device 110, out of the weight coefficients w1 to w3 for each of the multiple parameter conditions.

[0094] <Data acquisition unit 123> The data acquisition unit 123 acquires various data necessary for controlling the coke manufacturing process. The data acquired by the data acquisition unit 123 includes, for example, actual operating values ​​from the present to the past, scheduled values ​​for future operations, target values ​​for operations, and various setting values. Examples of data acquisition methods include at least one of the following: operation of an input device by an operator of the processing unit 120, reception from an external device, and reading from a portable storage medium. Individual data can be input to the processing unit 120 and acquired by the data acquisition unit 123 at any time. Therefore, individual data do not necessarily need to be input to the processing unit 120 and acquired by the data acquisition unit 123 at the same time.

[0095] <Predicted value derivation unit 124> The predicted value derivation unit 124 calculates the predicted value of the furnace group temperature Tr(t), the predicted value of the input heat Q(t), and the coke temperature Tc(t) so that the prediction error with the candidate target furnace temperature orbit Tr_ref(t) output from the target furnace temperature determination unit 125 is less than or equal to a predetermined value. e The predicted value of ) is derived. The predicted value derivation unit 124 is the same as the predicted value derivation unit described in Patent Document 1, so a detailed explanation of the predicted value derivation unit 124 is omitted here.

[0096] <Target furnace temperature determination unit 125> The target furnace temperature determination unit 125 uses the candidate target furnace temperature trajectory Tr_ref(t) output to the prediction value derivation unit 124, and the predicted values ​​of the input heat Q(t), the predicted furnace group temperature Tr(t), and the coke temperature Tc(t) derived (determined) by the prediction value derivation unit 124 for the candidate target furnace temperature trajectory Tr_ref(t). e Using the predicted values ​​of ) and , we derive the value of the evaluation function J (goodness-of-fit function) in equation (1) below when the constraint equation in equation (2) above is satisfied.

[0097] The target furnace temperature determination unit 125 sets the weight coefficients w1 to w3 selected by the parameter selection unit 122 as the weight coefficients w1 to w3 of the evaluation function J in equation (1). Thus, in this embodiment, an example is shown in which the target furnace temperature determination unit 125 includes an output variable derivation unit that derives the value of the evaluation function in equation (1), which is an example of an input / output model, as an example of an output variable.

[0098] The target furnace temperature determination unit 125 determines whether the convergence condition is satisfied based on the derivation of the value of the evaluation function J for each candidate included in the candidate group of target furnace temperature orbital Tr_ref(t). The convergence condition can be any condition used in metaheuristic methods such as genetic algorithms.

[0099] If the convergence conditions are not met, the target furnace temperature determination unit 125 regenerates and updates the candidate group of target furnace temperature orbital Tr_ref(t). The target furnace temperature determination unit 125 then repeatedly performs the derivation of the value of the evaluation function J in equation (1), the determination of whether or not the convergence conditions are met, and the regeneration of the candidate group of target furnace temperature orbital Tr_ref, as described above, until the convergence conditions are met.

[0100] The target furnace temperature determination unit 125 determines the optimal target furnace temperature trajectory Tr_ref(t) from among the multiple candidate target furnace temperature trajectories Tr_ref(t) derived when the convergence conditions are satisfied. This optimal trajectory is determined by the design variables (transient start furnace temperature change ΔTr1, transient end furnace temperature change ΔTr2, minimum furnace temperature arrival time time1, and minimum furnace temperature maintenance time2) at which the value of the evaluation function J is minimized.

[0101] Except for the method of applying the weighting coefficients w1 to w3, the target furnace temperature determination unit 125 is the same as the target furnace temperature determination unit described in Patent Document 1, so a detailed explanation of the target furnace temperature determination unit 125 is omitted here.

[0102] <Control Unit 126> The control unit 126 generates and outputs a control signal to set the amount of heat input to the combustion chamber to a value corresponding to the difference between the target furnace temperature trajectory determined by the target furnace temperature determination unit 125 and the actual value of the furnace bed temperature. Since the control unit 126 is the same as the control unit described in Patent Document 1, a detailed explanation of the control unit 126 will be omitted here.

[0103] [How to create parameters] Next, an example of a parameter creation method using the parameter creation device 110 will be explained with reference to the flowchart in Figure 11. In step S1101, the performance acquisition unit 111 acquires the actual time series data of the input variables. In this embodiment, an example is shown where the actual time interval data of the input variables includes the actual time series data 203 of the furnace temperature error, the actual time series data 202 of the coke temperature error, and the actual time series data 201 of the heat input (see Figure 2).

[0104] Next, in step S1102, the splitting unit 112 splits the input variable actual time series data acquired in step S1101 into data for each time interval 1 to n (input variable actual time interval data) (see Figure 2). Next, in step S1103, the time interval selection unit 113 derives the characteristic value (interval mean) of the input variable actual time interval data obtained in step S1102. Next, in step S1104, the time interval selection unit 113 excludes feature values ​​(interval mean values) that are outside the preset tolerance range from the feature values ​​(interval mean values) of each time interval 1 to n of the input variable actual time series data derived in step S1103 (see Figures 3A and 3B).

[0105] Next, in step S1105, the time interval selection unit 113 scales the feature values ​​(interval mean values) of the actual time interval data of the input variables in multiple time intervals 1 to n, excluding the feature values ​​that were excluded in step S1104, for each input variable (heat input, coke temperature error, furnace temperature error). Next, in step S1106, the time interval selection unit 113 derives evaluation values ​​(interval data evaluation values) for the feature values ​​(interval mean values) of the input variable actual time interval data scaled in step S1105.

[0106] If the parameter selection conditions are determined for each operator, the time interval selection unit 113 derives the interval data evaluation value using equation (3). Also, if the parameter selection conditions are determined for each operator, the time interval selection unit 113 derives the distance of time interval m from the overall average as the interval data evaluation value. As mentioned above, the distance of time interval m from the overall average corresponds to the distance between the overall average values ​​322a and 322b (black circles) of the input variable actual time interval data and the in-interval average values ​​(white circles) of the input variable actual time interval data within the dashed lines 321a and 321b in Figure 3B.

[0107] Next, in step S1107, the time interval selection unit 113 selects a time interval that has input variable time interval data satisfying the parameter selection conditions, for each parameter selection condition. If the parameter selection conditions are operator-independent, the time interval selection unit 113 sorts the interval data evaluation values ​​of multiple time intervals 1 to n and selects a time interval that has the interval data evaluation value showing the highest evaluation value, for each parameter selection condition (see Figure 7). If the parameter selection conditions are determined for each operator, the time interval selection unit 113 derives the time interval that minimizes the interval data evaluation value (distance from the average of all time intervals m), for each parameter selection condition (for each operator A and B) (see Figure 9).

[0108] Next, in step S1108, the parameter candidate derivation unit 114 derives candidate weight coefficients w1 to w3 for each parameter selection condition using the input variable actual time interval data included in the time interval selected by the time interval selection unit 113.

[0109] Finally, in step S1109, the storage unit 115 stores the candidate weight coefficients w1 to w3 (weight coefficients w1 to w3 for each parameter condition) derived by the parameter candidate derivation unit 114 in the storage medium inside the parameter creation device 110.

[0110] [Processing method] Next, an example of a processing method performed using the processing unit 120 will be explained with reference to the flowchart in Figure 12. The flowchart in Figure 12 starts after the processing according to the flowchart in Figure 11 has been completed.

[0111] In step S1201, the condition acquisition unit 121 acquires the selection conditions (parameter selection conditions) for the weight coefficients w1 to w3 to be applied to the evaluation function J of equation (1). Next, in step S1202, the parameter selection unit 122 selects from the candidate weight coefficients w1 to w3 (weight coefficients w1 to w3 for each parameter condition) created by the parameter creation device 110 the weight coefficients w1 to w3 that satisfy the parameter selection conditions obtained in step S1201, and uses them as the weight coefficients w1 to w3 in equation (1).

[0112] Next, in step S1203, the target furnace temperature determination unit 125 sets the weight coefficients w1 to w3 selected in step S1202 as the weight coefficients w1 to w3 of the evaluation function J in equation (1).

[0113] Next, in step S1204, the target furnace temperature determination unit 125 takes the candidate target furnace temperature trajectory Tr_ref(t) output to the prediction value derivation unit 124 and the predicted value of the input heat Q(t), the predicted value of the furnace group temperature Tr(t), and the coke temperature Tc(t) derived (determined) by the prediction value derivation unit 124 for the candidate target furnace temperature trajectory Tr_ref(t), as well as the predicted value of the input heat Q(t), the predicted value of the furnace group temperature Tr(t), and the coke temperature Tc(t) e Using the predicted values ​​of (1) and (2), the value of the evaluation function J (fit function) in equation (1) when the constraint equation in equation (2) is satisfied is derived. The target furnace temperature determination unit 125 repeatedly performs the derivation of the value of the evaluation function J in equation (1), the determination of whether or not the convergence condition is satisfied, and the regeneration of the candidate group of target furnace temperature orbit Tr_ref until the convergence condition is satisfied. When the convergence condition is satisfied, the target furnace temperature determination unit 125 determines the target furnace temperature orbit Tr_ref(t) determined by the design variables when the value of the evaluation function J is minimized among the multiple candidate target furnace temperature orbit Tr_ref(t) derived, as the optimal solution for the target furnace temperature orbit Tr_ref(t).

[0114] Finally, in step S1205, the control unit 126 generates and outputs a control signal to set the amount of heat input to the combustion chamber to the amount of heat corresponding to the difference between the optimal solution of the target furnace temperature trajectory determined by the target furnace temperature determination unit 125 and the actual value of the furnace bed temperature.

[0115] [summary] As described above, in this embodiment, the processing unit 120 acquires parameter selection conditions, selects pre-configured parameters that satisfy the acquired parameter selection conditions from among the candidates of pre-configured parameters stored in the storage medium, and applies the selected pre-configured parameters to the input / output model to derive the output variables. Therefore, the pre-configured parameters can be changed adaptively. The pre-configured parameters may be, for example, weight coefficients for multiple evaluation indices included in the evaluation function used to solve the optimization problem. Alternatively, a set of multiple pre-configured parameters may be selected. In this case, multiple pre-configured parameters can be changed adaptively at once.

[0116] Furthermore, in this embodiment, the processing unit 120 obtains the parameter selection conditions specified by the operator from among a plurality of pre-set parameter selection conditions. Therefore, the parameter selection conditions can be obtained without performing complex calculations.

[0117] Furthermore, in this embodiment, the processing unit 120 derives output variables using an input / output model for the control of the equipment that manufactures the product. In this case, the parameter selection conditions include selection conditions determined based on the product manufacturing conditions. Therefore, pre-set parameters can be set according to the product manufacturing conditions. In this case, the manufacturing conditions may include manufacturing conditions determined using multiple manufacturing policies. Therefore, pre-set parameters can be set that can balance multiple manufacturing policies.

[0118] Here, the product may be coke produced in the coke manufacturing process. Operations in the coke manufacturing process have relied on manual operation based on the tacit knowledge of the operator. On the other hand, as described in Patent Document 1, algorithms for controlling the coke manufacturing process have been constructed in order to automate at least a part of the operations in the coke manufacturing process. However, even if such an algorithm is constructed, if the criteria for setting the pre-set parameters are left to the operator's tacit knowledge, the control performance of the algorithm may depend on the operator's skill. In contrast, in this embodiment, pre-set parameters according to various manufacturing conditions are prepared, and by selecting the pre-set parameters that satisfy the manufacturing conditions and setting them in the input / output model, the degree of dependence on the operator's tacit knowledge can be reduced, and the control performance when automating operations can be improved.

[0119] Furthermore, in industries such as the steel industry, where the same type of product (coke in this embodiment) is manufactured at various steel mills, the pre-set parameters may be shared in real time at each steel mill. Figure 13 shows an example of a system for sharing pre-set parameters at multiple locations. Figure 13 illustrates a case where the parameter creation device 110 and multiple processing devices 120a to 120d are connected to each other via a communication network 130 so as to be able to communicate with one another. Each of the multiple processing devices 120a to 120d is equipped with the parts of the processing device 120 described above. The multiple processing devices 120a to 120d are installed, for example, at each steel mill. The parameter creation device 110 may send candidate pre-set parameters to the processing device that made the request for pre-set parameters in response to the request for pre-set parameters from the processing devices 120a to 120d. Alternatively, the multiple processing devices 120a to 120d may obtain candidate pre-set parameters without communicating with the parameter creation device 110. For example, multiple processing units 120a to 120d may acquire candidate pre-configured parameters stored in an external storage device of the parameter creation device 110. In this case, the candidate pre-configured parameters may be stored on a server owned by the manufacturer or in the cloud.

[0120] Furthermore, in this embodiment, the parameter creation device 110 acquires the actual time-series data of the input variable, divides it into actual time-interval data of the input variable for each time interval 1 to n, and selects a time interval from the multiple time intervals 1 to n that contains actual time-interval data of the input variable that satisfies the parameter selection conditions. Then, the parameter creation device 110 uses the actual time-interval data of the input variable included in the selected time interval to derive candidate pre-set parameters and stores them in the storage medium. Therefore, pre-set parameters that satisfy the selection conditions of the parameter can be derived using actual time-interval data of the input variable that satisfies the parameter selection conditions. Thus, pre-set parameters that satisfy the selection conditions of the parameter can be derived regardless of the parameter selection conditions, and pre-set parameters corresponding to various parameter selection conditions can be derived.

[0121] Furthermore, in this embodiment, the parameter creation device 110 derives characteristic values ​​of the input variable actual time interval data for each time interval 1 to n, and based on the derived characteristic values ​​for each time interval 1 to n, selects a time interval from the multiple time intervals 1 to n that contains input variable actual time interval data that satisfies the parameter selection conditions. Therefore, input variable actual time interval data having characteristics corresponding to the parameter selection conditions can be easily extracted.

[0122] Furthermore, in this embodiment, the parameter creation device 110 selects a time interval that satisfies the parameter selection conditions based on the feature values ​​of each time interval 1 to n, excluding feature values ​​that are outside the preset tolerance range. Therefore, it is possible to suppress the derivation of preset parameters with values ​​unintended by the operator.

[0123] Furthermore, in this embodiment, the parameter creation device 110 scales the feature values ​​for each time interval 1 to n for each input variable, and uses the scaled feature values ​​for each time interval 1 to n as the feature values ​​for multiple intervals 1 to n. Therefore, even if the units of multiple input variables are different, it is easy to treat the feature values ​​of the actual time interval data of these multiple input variables equally (i.e., as if they had the same units).

[0124] Furthermore, in this embodiment, the parameter creation device 110 derives evaluation values ​​for each time interval 1 to n to evaluate whether the characteristic values ​​of the input variable actual time interval data satisfy the parameter selection conditions, and selects a time interval containing the input variable actual time interval data that satisfies the parameter selection conditions based on the derived evaluation values. Therefore, the input variable actual time interval data that satisfies the parameter selection conditions can be extracted based on quantitative indicators.

[0125] Furthermore, in this embodiment, the parameter creation device 110 derives candidate pre-set parameters by performing inverse reinforcement learning using the input variable actual time interval data included in the time interval selected as described above. Therefore, the values ​​of the pre-set parameters can be set to values ​​suitable for the input / output model.

[0126] (Other embodiments) Furthermore, the embodiments of this disclosure described above can be realized by a computer executing a program. A computer-readable recording medium on which the program is recorded, and computer program products such as the program itself, can also be applied as embodiments of this disclosure. Examples of recording media that can be used include flexible disks, hard disks, optical disks, magneto-optical disks, CD-ROMs, magnetic tapes, non-volatile memory cards, ROMs, and the like. Furthermore, the embodiments of this disclosure described above are merely examples of concrete implementations of this disclosure, and the technical scope of this disclosure should not be interpreted as being limited by them. In other words, this disclosure can be implemented in various ways without departing from its technical concept or its main features.

[0127] Furthermore, the disclosure of the above embodiments is as follows, for example. [Disclosure 1] A processing device that derives an output variable using an input / output model that takes an input variable as input and outputs an output variable, A condition acquisition unit that acquires selection conditions for pre-set parameters that are set in advance for the input / output model in order to perform the aforementioned derivation, A parameter selection unit selects a pre-configured parameter that satisfies the aforementioned selection conditions from among a plurality of pre-configured parameter candidates stored in the storage medium; An output variable derivation unit that applies the pre-set parameters selected in the parameter selection unit to the input / output model and performs the derivation; A processing device equipped with the following features. [Disclosure 2] The aforementioned input / output model includes an evaluation function used to solve the optimization problem, The processing apparatus according to Disclosure 1, wherein the pre-configured parameters include weight coefficients for a plurality of evaluation indicators included in the evaluation function. [Disclosure 3] The processing apparatus according to disclosure 1 or 2, wherein the parameter selection unit selects a set of multiple pre-configured parameters as pre-configured parameters to be applied to the input / output model. [Disclosure 4] The processing apparatus according to any one of disclosures 1 to 3, wherein the condition acquisition unit acquires a selection condition specified by the operator from among a plurality of pre-set selection conditions. [Disclosure 5] The aforementioned derivation is performed for the control of the equipment used to manufacture the product. The apparatus according to any one of disclosures 1 to 4, wherein the selection conditions include selection conditions determined based on the manufacturing conditions of the product. [Disclosure 6] The apparatus according to disclosure 5, wherein the manufacturing conditions include manufacturing conditions determined using multiple manufacturing policies. [Disclosure 7] A parameter creation device for deriving candidate pre-set parameters described in any one of disclosures 1 to 6, A performance acquisition unit acquires time-series data of the actual values ​​of at least one variable among the input and output variables that can be measured, A splitting unit that divides the aforementioned time-series data into data for each of multiple time intervals, A time interval selection unit selects a time interval from the plurality of time intervals that has data satisfying the selection criteria, based on the data for each of the plurality of time intervals. A parameter candidate derivation unit derives candidate parameters for the pre-set parameters using time-series data included in the time interval selected by the time interval selection unit, A parameter creation device equipped with the following features. [Disclosure 8] The parameter creation device according to disclosure 7, wherein the time interval selection unit derives characteristic values ​​of the time series data for each of the multiple time intervals, and selects a time interval from the multiple time intervals that has time series data satisfying the selection conditions based on the characteristic values ​​of each of the multiple time intervals derived. [Disclosure 9] The parameter creation device according to Disclosure 8, wherein the time interval selection unit selects a time interval having time series data that satisfies the selection conditions based on feature values ​​from among the feature values ​​for each of the plurality of time intervals, excluding feature values ​​that are outside a preset tolerance range. [Disclosure 10] The parameter creation device according to disclosure 8 or 9, wherein the time interval selection unit scales the feature values ​​for each of the multiple time intervals for each of the input variables, and uses the feature values ​​scaled for each of the multiple time intervals as the feature values ​​for each of the multiple intervals. [Disclosure 11] The parameter creation device according to any one of disclosures 8 to 10, wherein the time interval selection unit derives an evaluation value for each of the multiple time intervals for evaluating whether the characteristic value of the time series data for each of the multiple time intervals is a value that satisfies the selection condition, and selects a time interval having time series data that satisfies the selection condition based on the derived evaluation value. [Disclosure 12] There are multiple selection criteria, The aforementioned time interval selection unit selects data that satisfies the selection conditions for each of the multiple selection conditions, The parameter creation apparatus according to any one of disclosures 7 to 11, wherein the parameter candidate derivation unit derives candidate parameters for each of the multiple selection conditions. [Disclosure 13] The parameter creation apparatus according to any one of disclosures 7 to 12, wherein the parameter candidate derivation unit derives candidate parameters for the pre-set parameters by performing inverse reinforcement learning using data included in the time interval selected by the time interval selection unit. [Disclosure 14] The aforementioned input / output model includes an evaluation function used to solve the optimization problem, The aforementioned pre-configured parameters include weight coefficients for multiple evaluation metrics included in the evaluation function, The performance acquisition unit is a parameter creation device according to any one of disclosures 7 to 13, which acquires time-series data of the performance values ​​of the multiple evaluation indicators. [Disclosure 15] The parameter creation apparatus according to any one of disclosures 7 to 14, wherein the parameter candidate derivation unit derives candidate sets of pre-set parameters using time-series data included in the time interval selected by the time interval selection unit. [Disclosure 16] The results of the above derivation are used to control the equipment used to manufacture the product. The selection criteria are determined based on the manufacturing conditions of the product, and the parameter creation apparatus is as described in any one of disclosures 7 to 15. [Disclosure 17] The parameter creation apparatus according to disclosure 16, wherein the manufacturing conditions include manufacturing conditions determined using multiple manufacturing policies. [Disclosure 18] A method for deriving an output variable using an input / output model that takes an input variable as input and outputs an output variable, A condition acquisition step to acquire selection conditions for pre-set parameters that are set in advance for the input / output model in order to perform the derivation, A parameter selection step in which a pre-configured parameter that satisfies the aforementioned selection conditions is selected from a plurality of pre-configured parameter candidates that are stored in advance on a storage medium, An output variable derivation step is performed by applying the pre-set parameters selected in the parameter selection step to the input / output model and performing the derivation; A processing method comprising: [Disclosure 19] A parameter creation method for deriving candidate pre-configured parameters as described in Disclosure 18, A performance acquisition step involves acquiring time-series data of the actual values ​​of at least one variable among the input and output variables that can be measured, A division step of dividing the aforementioned time series data into data for each of multiple time intervals, A time interval selection step, based on the data for each of the plurality of time intervals, selects a time interval from the plurality of time intervals that has data satisfying the selection criteria, A parameter candidate derivation step is performed to derive candidate parameters for the pre-set parameters using time-series data included in the time interval selected in the time interval selection step, A method for creating parameters that include the following features. [Disclosure 20] A program for causing a computer to function as a part of the processing apparatus described in any one of disclosures 1 to 6. [Disclosure 21] A program for causing a computer to function as a component of the parameter creation device described in any one of disclosures 7 to 17. [Explanation of Symbols]

[0128] 110 Parameter creation device 111 Performance Acquisition Department 112 Division 113 Time zone selection section 114 Parameter Candidate Derivation Unit 115 Storage section 116 Parameter output section 120 Processing Units 121 Condition Acquisition Unit 122 Parameter Selection Section 123 Data Acquisition Unit 124 Predicted Value Derivation Unit 125 Target furnace temperature determination unit (variable derivation unit) 126 Control Unit 130 Communication Networks 201 Actual time-series data of heat input 202 Actual Time Series Data of Coke Temperature Error 203 Actual Time Series Data of Furnace Temperature Error 311a~311c Mean values ​​within the interval of the actual interval data for the input variables 321a~321c Dashed lines indicating the average value within the interval of the actual interval data for each operator's input variables. 322a~322b Average value of all interval data for the input variable at actual time

Claims

1. A processing device that derives an output variable using an input / output model that takes an input variable as input and outputs an output variable, A condition acquisition unit that acquires selection conditions for pre-set parameters that are set in advance for the input / output model in order to perform the aforementioned derivation, A parameter selection unit selects a pre-configured parameter that satisfies the aforementioned selection conditions from among a plurality of pre-configured parameter candidates stored in the storage medium, An output variable derivation unit that applies the pre-set parameters selected in the parameter selection unit to the input / output model and performs the derivation; A processing device equipped with the following features.

2. The aforementioned input / output model includes an evaluation function used to solve the optimization problem, The processing apparatus according to claim 1, wherein the pre-configured parameters include weight coefficients for a plurality of evaluation indicators included in the evaluation function.

3. The processing apparatus according to claim 1 or 2, wherein the parameter selection unit selects a set of multiple pre-configured parameters as pre-configured parameters to be applied to the input / output model.

4. The processing apparatus according to claim 1 or 2, wherein the condition acquisition unit acquires a selection condition specified by an operator from a plurality of pre-set selection conditions.

5. The aforementioned derivation is performed for the control of the equipment used to manufacture the product. The apparatus according to claim 1 or 2, wherein the selection conditions include selection conditions determined based on the manufacturing conditions of the product.

6. The apparatus according to claim 5, wherein the manufacturing conditions include manufacturing conditions determined using a plurality of manufacturing policies.

7. A parameter creation device for deriving candidate pre-set parameters according to claim 1 or 2, A performance acquisition unit acquires time-series data of actual values ​​of at least one variable among the input and output variables that can be measured, A splitting unit that divides the aforementioned time-series data into data for each of multiple time intervals, A time interval selection unit selects a time interval from the plurality of time intervals that has data satisfying the selection criteria, based on the data for each of the plurality of time intervals. A parameter candidate derivation unit derives candidate parameters for the pre-set parameters using time-series data included in the time interval selected by the time interval selection unit, A parameter creation device equipped with the following features.

8. The parameter creation device according to claim 7, wherein the time interval selection unit derives characteristic values ​​of the time series data for each of the multiple time intervals, and selects a time interval from the multiple time intervals that has time series data satisfying the selection conditions based on the characteristic values ​​of each of the multiple time intervals derived.

9. The parameter creation device according to claim 8, wherein the time interval selection unit selects a time interval having time series data that satisfies the selection conditions based on the feature values ​​of the plurality of time intervals, excluding feature values ​​that are outside a preset tolerance range.

10. The parameter creation device according to claim 8, wherein the time interval selection unit scales the feature values ​​for each of the multiple time intervals for each of the input variables, and uses the feature values ​​scaled for each of the multiple time intervals as the feature values ​​for each of the multiple intervals.

11. The parameter creation device according to claim 8, wherein the time interval selection unit derives an evaluation value for each of the multiple time intervals for evaluating whether the characteristic value of the time series data for each of the multiple time intervals is a value that satisfies the selection condition, and selects a time interval having time series data that satisfies the selection condition based on the derived evaluation value.

12. There are multiple selection criteria, The aforementioned time interval selection unit selects data that satisfies the selection conditions for each of the multiple selection conditions, The parameter creation device according to claim 7, wherein the parameter candidate derivation unit derives candidate parameters for each of the plurality of selection conditions.

13. The parameter creation device according to claim 7, wherein the parameter candidate derivation unit derives candidate parameters for the pre-set parameters by performing inverse reinforcement learning using data included in the time interval selected by the time interval selection unit.

14. The aforementioned input / output model includes an evaluation function used to solve the optimization problem, The aforementioned pre-configured parameters include weight coefficients for multiple evaluation metrics included in the evaluation function, The parameter creation device according to claim 7, wherein the performance acquisition unit acquires time-series data of the performance values ​​of the plurality of evaluation indicators.

15. The parameter creation apparatus according to claim 7, wherein the parameter candidate derivation unit derives candidate sets of pre-set parameters using time-series data included in the time interval selected by the time interval selection unit.

16. The results of the above derivation are used to control the equipment used to manufacture the product. The parameter creation apparatus according to claim 7, wherein the selection conditions are determined based on the manufacturing conditions of the product.

17. The parameter creation apparatus according to claim 16, wherein the manufacturing conditions include manufacturing conditions determined using a plurality of manufacturing policies.

18. A method for deriving an output variable using an input / output model that takes an input variable as input and outputs an output variable, A condition acquisition step to acquire selection conditions for pre-set parameters that are set in advance for the input / output model in order to perform the derivation, A parameter selection step in which a pre-configured parameter that satisfies the aforementioned selection conditions is selected from a plurality of pre-configured parameter candidates that are stored in advance on a storage medium, An output variable derivation step is performed by applying the pre-set parameters selected in the parameter selection step to the input / output model and performing the derivation; A processing method comprising:

19. A parameter creation method for deriving candidate pre-set parameters as described in claim 18, A performance acquisition step involves acquiring time-series data of the actual values ​​of at least one variable among the input and output variables that can be measured, A division step of dividing the aforementioned time series data into data for each of multiple time intervals, A time interval selection step, based on the data for each of the multiple time intervals, selects a time interval from the multiple time intervals that has data satisfying the selection criteria. A parameter candidate derivation step is performed to derive candidate parameters for the pre-set parameters using time-series data included in the time interval selected in the time interval selection step, A method for creating parameters that include the following features.

20. A program for causing a computer to function as each part of the processing apparatus according to claim 1 or 2.

21. A program for causing a computer to function as each part of the parameter creation device described in claim 7.

Images