JP7734571B2 - Process model construction system and process model construction method - Google Patents

Description

The present invention relates to a process model construction system and a process model construction method.

In process operations, there are times when the optimal operating conditions are known from past operating records, but the operating method to reproduce those operating conditions is unknown. This is because various disturbances, such as reaction heat generated within controlled objects such as valves and other control devices, the characteristics of deteriorated control devices, and external environmental conditions that cannot be obtained as input, vary with each operation, and the operating method to achieve the same operating conditions varies each time.

One existing technology that addresses this issue is MPC (Model Predictive Control). MPC models the characteristics of a controlled object, enabling prediction of the controlled variable relative to the manipulated variable of the controlled object, and thereby enabling back-calculation of the manipulated variable that satisfies the desired control waveform. An example of such existing technology is Patent Document 1. In Patent Document 1, a first-principles model is used to simulate a batch process, and this first-principles model can be used to construct a multiple-input/multiple-output control routine to control the batch process. The first-principles model can generate estimates of batch variables that cannot be measured or are not measured during the operation of the actual batch process. An example of such a variable may be the rate of change of a component of the batch process (e.g., production rate, cell growth rate, etc.). It is stated that the first-principles model and the constructed multiple-input/multiple-output control routine can be used to facilitate control of the batch process.

JP 2012-64245 A

In the aforementioned Patent Document 1, by adding a mechanism (software sensor) that uses a first-principles model to estimate difficult-to-measure factors present in a batch process, appropriate operating methods are clarified based on the estimated values of the first-principles model, and the estimated values are used as the control object to shorten the batch cycle and improve yield. However, controlling control objects with different characteristics requires manually constructing different first-principles models, making practical application difficult in multi-item production sites. For example, if there are multiple combinations of the characteristics of the product manufactured by the control object and the characteristics of the control equipment to be controlled, a process model must be constructed for each combination. In particular, in multi-item production sites such as batch processes, there are many combinations of product types and equipment, resulting in a huge number of process models to be constructed. Building a huge number of process models is difficult, and post-construction maintenance becomes cumbersome, resulting in a heavy operational burden. For this reason, there is a need for technology that can build process models that impose less construction and operational burden than conventional methods, even when the control object has multiple characteristics.

One aspect of the present invention aims to provide technology for constructing process models that impose less construction and operational burden than conventional methods, even when the controlled object has multiple characteristics.

The process model construction system of the present invention is a process model construction system that constructs a process model that models the characteristics of a controlled object using a computer, and is configured as a process model construction system characterized by having: a first processing unit that acquires the characteristics from past operation performance data that indicates operating performance according to the characteristics of the controlled object; and a second processing unit that calculates characteristic parameters for each characteristic to correct the process model so that actual values of the controlled object process for the characteristics included in the past operation performance data and estimated values of the controlled object process estimated by the process model constructed using a predetermined theoretical formula satisfy predetermined conditions.

According to one aspect of the present invention, it is possible to construct a process model that imposes less construction and operational burden than conventional methods, even when the controlled object has multiple characteristics.

FIG. 1 illustrates an example of a configuration of a process model construction system according to a first embodiment. FIG. 1 is a diagram illustrating an outline of a computer. FIG. 10 is a diagram showing an example of past operation performance data. FIG. 2 is a diagram illustrating an example of a process model and an example of a relationship between the process model and characteristics. FIG. 10 is a diagram illustrating an example of a characteristic parameter group. FIG. 10 is a diagram illustrating an example of characteristic information. FIG. 10 is a diagram illustrating an example of a flowchart showing a processing procedure of a model construction process performed by a model construction unit in the first embodiment. FIG. 8 is a diagram illustrating an example of a flowchart showing a processing procedure of a characteristic parameter calculation process performed in step S706 shown in FIG. 7. FIG. 8 is a diagram illustrating an example of a flowchart showing a processing procedure of a characteristic parameter calculation process performed in step S706 shown in FIG. 7. FIG. 8C is a diagram showing an example of a flowchart illustrating the processing procedure for model identification processing of a process model recorded in a model identification data group in the characteristic parameter calculation processing shown in FIGS. 8A and 8B. FIG. 8C is a diagram showing an example of a flowchart illustrating the processing procedure for model identification processing of a process model recorded in a model identification data group in the characteristic parameter calculation processing shown in FIGS. 8A and 8B. FIG. 10 is a diagram illustrating an example of a flowchart showing a processing procedure of a simulation process performed by a simulation unit in the first embodiment. FIG. 10 is a diagram illustrating an example of a configuration of a process model construction system according to a second embodiment. FIG. 10 is a diagram illustrating an example of a configuration of a process model building system according to a third embodiment. FIG. 10 is a diagram illustrating an example of additional characteristic information. FIG. 10 is a diagram illustrating an example of a configuration of a process model construction system according to a fourth embodiment. FIG. 10 is a diagram illustrating an example of model identification information. FIG. 13 is a diagram illustrating an example of a configuration of a process model construction system according to a fifth embodiment. FIG. 13 is a flowchart illustrating an example of a processing procedure of a simulation process performed by a simulation unit in the fifth embodiment. FIG. 20 is a diagram illustrating an example of a configuration of a process model construction system according to a sixth embodiment. FIG. 20 is a diagram illustrating an example of a flowchart showing the processing procedure of a model construction process performed by a model construction unit in the sixth embodiment. FIG. 20 is a diagram illustrating an example of a flowchart showing the processing procedure of the characteristic model calculation processing performed in S1706 shown in FIG. 17; FIG. 20 is a diagram illustrating an example of a flowchart showing the processing procedure of the characteristic model calculation processing performed in S1706 shown in FIG. 17; FIG. 10 is a diagram showing an example of a screen (model construction screen) displayed on a display of a computer constituting the process model construction system when a model construction unit executes a model construction process. FIG. 10 is a diagram showing an example of a screen (simulation execution screen) displayed on a display of a computer constituting the process model building system when a simulation unit executes a simulation process. FIG. 10 is a diagram showing an example of a screen (feedback execution screen) displayed on a display of a computer constituting a process model building system when a control device input calculation unit back-calculates an operation method of a controlled object that satisfies optimal operating conditions.

Embodiments of the present invention will be described below with reference to the drawings. The following description and drawings are examples for explaining the present invention, and some details have been omitted or simplified as appropriate for clarity of explanation. The present invention can also be implemented in various other forms. Unless otherwise specified, each component may be singular or plural.

The position, size, shape, range, etc. of each component shown in the drawings may not represent the actual position, size, shape, range, etc. in order to facilitate understanding of the invention. Therefore, the present invention is not necessarily limited to the position, size, shape, range, etc. disclosed in the drawings.

In the following explanation, various types of information may be described using terms such as "database," "table," and "list," but the various types of information may also be expressed in other data structures. To indicate that the information is not dependent on the data structure, "XX table," "XX list," etc. may be referred to as "XX information." When describing identification information, terms such as "identification information," "identifier," "name," "ID," and "number" are used, and these terms are interchangeable.

When there are multiple components with the same or similar functions, they may be described using the same reference numeral with different subscripts. However, when there is no need to distinguish between these multiple components, the subscripts may be omitted.

In addition, the following description may describe processing performed by executing a program, but the program is executed by a processor (e.g., a CPU (Central Processing Unit), GPU (Graphics Processing Unit)) to perform the specified processing using storage resources (e.g., memory) and/or interface devices (e.g., communication ports) as appropriate, so the processor may be the subject of the processing. Similarly, the subject of processing performed by executing a program may be a controller, device, system, computer, or node having a processor. The subject of processing performed by executing a program may be any computing unit, and may include a dedicated circuit that performs specific processing (e.g., an FPGA (Field-Programmable Gate Array) or ASIC (Application Specific Integrated Circuit)).

A program may be installed on a device such as a computer from a program source. The program source may be, for example, a program distribution server or a computer-readable storage medium. If the program source is a program distribution server, the program distribution server may include a processor and storage resources for storing the program to be distributed, and the processor of the program distribution server may distribute the program to be distributed to other computers. Also, in the following description, two or more programs may be realized as one program, and one program may be realized as two or more programs.

The following describes the configuration of one embodiment of the process model construction system and process model construction method according to this embodiment.

FIG. 1 is a diagram showing an example of the configuration of a process model construction system according to the first embodiment. As shown in FIG. 1, the process model construction system 1000 according to this embodiment includes a model construction unit 1001, a simulation unit 1002, a characteristic parameter group 1003, and a process model 1004.

The process model construction system 1000 constructs a process model suitable for a control object with many combinations of characteristics by setting the parameters of the process model constructed using a theoretical formula according to the characteristics of the process model based on past operational performance data of the control object. This enables process simulation using a small number of process models without having to create a process model for each combination of characteristics. The control object is the object controlled by this system, such as control equipment that performs process operation. In the following, the control object is described assuming equipment such as reactors used in plants, but it may also be applied to various equipment and devices other than plants that perform process operation. Furthermore, the characteristics of the control object are described assuming the type of equipment and the variety of products manufactured by the control object, but other information that represents the characteristics of the control object may also be defined as characteristics.

Characteristics are elements that affect the behavior of a process model, such as the type of product manufactured by the controlled object through process operation, or the equipment that operates that process. A characteristic type, for example, represents the type of elements contained in a certain type of characteristic. A characteristic combination, for example, represents a combination of multiple types of characteristics.

The model construction unit 1001 calculates characteristic parameters (described below) that correct the parameters of the process model for controlling the control object that performs process operation based on the past operation performance data 101 and the prepared process model 1004, and stores the calculated characteristic parameters in the characteristic parameter group 1003.

The simulation unit 1002 corrects the parameters of the process model 1004 with the correction parameters stored in the characteristic parameter group 1003 based on the input characteristic information 102. The simulation unit 1002 simulates the behavior of the process performed by the controlled object based on the simulation input information 103, and stores the results in the simulation result 104. The simulation input information 103 is information that defines the method and conditions for performing the simulation (e.g., operating conditions desired by the user), and includes, for example, reference values that represent the setting values set for the controlled object at each time.

The characteristic parameter group 1003 stores one or more characteristic parameters. The characteristic parameter group 1003 is a table that stores a set of correction parameters for correcting a process model that indicates the characteristics of the controlled object. Here, an example is shown in which two characteristic parameters are stored in the characteristic parameter group 1003 as the characteristics of the controlled object: a characteristic parameter that indicates the type of product manufactured by the controlled object, and a characteristic parameter that indicates the equipment that is the controlled object. As mentioned above, the number and types of characteristic parameters may be determined appropriately depending on the environment in which the system is applied.

The process model 1004 is a model of the characteristics of the controlled object, and is constructed using a predetermined theoretical formula. The process model 1004 can be expressed using various mathematical formulas. However, the following description will be given using a model expressed as y = ^ax² + bx + c as an example. In this case, a, b, and c are model parameters, x is the model's input value, and y is the model's output value. Specific functions of the process model construction system 1000 will be described later using flowcharts and the like. Furthermore, although this example illustrates a case in which the process model 1004 is expressed by one logical formula related to a mathematical model, the process model 1004 may be composed of multiple logical formulas, and further, one or more of these logical formulas may have multiple inputs or outputs.

The process model construction system 1000 shown in Figure 1 can be realized, for example, by a general computer 1600, as shown in Figure 2 (schematic computer diagram), equipped with a CPU 1601, memory 1602, an external storage device 1603 such as an HDD (Hard Disk Drive), a reading/writing device 1607 for reading and writing information from/to a portable storage medium 1608 such as a CD (Compact Disk) or USB memory, an input device 1606 such as a scanner, keyboard, or mouse, an output device 1605 such as a display, a communication device 1604 such as a NIC (Network Interface Card) for connecting to a communication network, and an internal communication line (referred to as a system bus) 1609 such as a system bus that connects these devices.

Various data (e.g., characteristic parameter group 1003, process model 1004) stored in the process model construction system 1000 or used for processing can be realized by the CPU 1601 reading and using it from memory 1602 or external storage device 1603. Furthermore, each functional unit (e.g., model construction unit 1001, simulation unit 1020) possessed by each system or device can be realized by the CPU 1601 loading a specific program stored in the external storage device 1603 into memory 1602 and executing it.

The above-mentioned predetermined program may be stored (downloaded) into the external storage device 1603 from the storage medium 1608 via the reading/writing device 1607 or from the network via the communication device 1604, and then loaded into the memory 1602 and executed by the CPU 1601. Alternatively, the program may be loaded directly into the memory 1602 from the storage medium 1608 via the reading/writing device 1607 or from the network via the communication device 1604, and then executed by the CPU 1601.

The following describes an example in which the process model construction system 1000 is configured by a single computer, but similar functions may be realized by distributing all or part of these functions across one or more computers, such as a cloud, and communicating with each other via a network. The specific processing performed by each component of the process model construction system 1000 will be described later using flowcharts.

Figure 3 shows an example of past operational performance data 101. The past operational performance data 101 is data indicating past operational performance obtained from a controlled object performing process operation and is input to the model construction unit 1001. As shown in Figure 3, the past operational performance data 101 is stored in association with a data ID for identifying the data, the product type and equipment indicating characteristics, and operational performance data. The operational performance data is, for example, operational and measurement information for the controlled object constituting the plant, and stores, for example, time-series data including reference values, manipulated variables, and controlled variables. The reference value is information indicating the set value (in this example, the temperature set value) set for the controlled object (e.g., a reactor) at each time. The manipulated variable is information indicating the operating method of the controlled object calculated by the controlled object equipment based on the reference value to achieve control using the controlled variable. In this example, we assume that hot water flows on the wall of the controlled reactor, and the temperature is changed by changing the opening and closing degree of the valve through which the hot water flows. The controlled variable is the operating condition of the control object, and is, for example, information indicating the actual temperature inside the reactor that is the control object.

In Figure 3, for example, the past operational performance data identified by data ID "0" includes operational performance data 301 for a controlled object with characteristics represented by product type "I" and equipment "A." The reference value, manipulated variable, and controlled variable are stored in chronological order at predetermined intervals (e.g., every minute) as reference value α0 [°C]: [20, 20, 50, 50, ..., 20], manipulated variable β0 [%]: [15, 18, 80, 70, ..., 0], and controlled variable γ0 [°C]: [15, 18, 37, 40, ..., 30]. For example, the reference value α0 for equipment "A" that produces product type "I" stores temperature values such as 20°C, 20°C, 50°C, 50°C, ..., 20°C for every minute.

FIG. 4 is a diagram showing an example of a process model 1004 and an example of the relationship between the process model and characteristics. The process model is expressed by a model constructed using a predetermined theoretical formula as described above. In FIG. 4, a model expressed by y = ax ² + bx + c is shown as an example of the process model. In this example, the initial values of the parameters a, b, and c of the process model are given as 2, 1, and 0, respectively. Furthermore, by the processing described below, the correction amounts of the process model parameters are calculated as +0.2, +0, and +0 for the product type, respectively, and as +0, −0.1, and +0 for the equipment, respectively. As a result, the corrected parameters a, b, and c of the process model are calculated as 1.2, −0.1, and 0, respectively.

Figure 5 shows an example of the characteristic parameter group 1003. The characteristic parameter group 1003 stores characteristic parameters corresponding to the characteristic. In this example, characteristic parameters are stored for each characteristic and each characteristic type. As shown in Figure 5, the characteristic parameter group 1003 stores the type of characteristic of the controlled object (product type in this example) in association with characteristic parameters that are correction parameters for the parameters of the process model that indicate that characteristic. In Figure 5, for example, it can be seen that the characteristic parameters for correcting the parameters of the process model indicated by product type "I" are stored as a: 1.2, b: -0.1, and c: 0, respectively.

5 illustrates an example of a table storing characteristic parameters for product types, and the table stores as many characteristic parameters as there are characteristics in the characteristic parameter group 1003. Also, in FIG. 5, the process model is assumed to be a model expressed as y=ax ² +bx+c, and therefore the description is based on the assumption that the parameters are composed of three parameters: a, b, and c. However, as described above, when the process model is expressed in another way, the number of characteristic parameters stored corresponds to the form of the expression.

Figure 6 is a diagram showing an example of characteristic information 102. Characteristic information 102 is information indicating the characteristics of the controlled object and the characteristic types included in those characteristics. Characteristic information 102 is input when simulation unit 1020 executes a simulation. As shown in Figure 6, characteristic information 102 stores characteristics (e.g., variety, equipment, etc.) and the characteristic types included in each characteristic (A, B, A, B, etc.) in association with each other. Figure 6 shows, for example, that A, B, etc. are defined as characteristic types for variety.

Figure 7 is a diagram showing an example of a flowchart illustrating the processing steps of the model construction process performed by the model construction unit 1001 in Example 1. This process is performed when a user issues an instruction or operation to start the process. The model construction process is a process for calculating characteristic parameters based on the past operational performance data 101, and is mainly a process for preparing for the calculation of the characteristic parameters.

In S701, the model construction unit 1001 acquires the process model 1004 and past operational performance data 101 (S701).

In S702, the model construction unit 1001 creates a process model P using arbitrarily set initial values of the process model (S702). For example, the model construction unit 1001 creates a process model P in which initial values a=2, b=1, and c=0 are set for the parameters of the process model expressed as y=ax ² +bx+c.

In S703, the model construction unit 1001 acquires a characteristic list L from the past operation performance data 101 (S703). For example, the model construction unit 1001 acquires, as the characteristic list L, a list [product type, equipment] that associates the product types and equipment defined in the past operation performance data 101.

In S704, the model construction unit 1001 extracts characteristic l from the characteristic list L acquired in S703 without restoring it, and acquires a characteristic type list K for characteristic l (S704). For example, the model construction unit 1001 acquires l = variety from the characteristic list L described above, and acquires a characteristic type list K = [A, B] for the variety.

In S705, the model construction unit 1001 extracts, without restoration, characteristic type k from the characteristic type list K obtained in S704 (S705). For example, the model construction unit 1001 obtains characteristic type k = I from the characteristic type list K = [I, II] described above.

In S706, the model construction unit 1001 calculates the characteristic parameters for characteristic type k acquired in S705 and stores them in the characteristic parameter group 1003 (S706). For example, the model construction unit 1001 calculates the characteristic parameters for type I. Specific methods for calculating the characteristic parameters will be described later using Figures 8A and 8B.

In S707, the model construction unit 1001 determines whether the characteristic type list K is an empty set (S707). For example, if the model construction unit 1001 is calculating characteristic parameters for the characteristic type I of a variety, it determines that the characteristic type list K = [II] exists and therefore determines that it is not an empty set (S707; No), and returns to S705. The model construction unit 1001 then executes S705 and S706 for the characteristic type list K = [II]. At this time, processing has been executed for each of the characteristic type lists K = [I, II]. Therefore, the model construction unit 1001 determines that the characteristic type list K is an empty set (S707; Yes), and proceeds to S708.

In S708, the model construction unit 1001 determines whether the characteristic list L is an empty set (S708). For example, if the model construction unit 1001 is processing characteristic list L = [product type], it determines that the characteristic list L = [equipment] exists and therefore is not an empty set (S708; No), and returns to S704. The model construction unit 1001 then executes S704, S705, and S706 for l = [equipment]. At this time, processing has been performed for each of the characteristic lists L = [product type, equipment]. Therefore, the model construction unit 1001 determines that the characteristic list L is an empty set (S708; Yes), and ends processing.

Figures 8A and 8B are an example of a flowchart showing the processing steps for the characteristic parameter calculation process performed in S706 shown in Figure 7. The characteristic parameter calculation process is a process for calculating characteristic parameters. A specific explanation will be given below using the past operational performance data 101 shown in Figure 3. In the explanation using Figures 8A and 8B, a control target with two characteristics including the two characteristic types described above will be used as an example, and therefore specific explanations will be given in the order of Loop 1 to Loop 4.

In Loop 1, the characteristic list L = [Equipment] and characteristic type list K = [Ro] have not been processed, and characteristic l = product type and characteristic type k = I will be processed.

In S801, the model construction unit 1001 extracts partial data D for characteristic l = characteristic type k from the past operation performance data 101 (S801). For example, the model construction unit 1001 extracts data for product type I from the past operation performance data 101 and creates partial data D consisting of data IDs (#) of 0 and 1. In this case, from the past operation performance data 101 shown in Figure 3, the records "0, I, A, [α0, β0, γ0...]" and "1, I, B, [α1, β1, γ1...]", whose data IDs (#) are identified by 0 and 1, respectively, are extracted as partial data D.

In S802, the model construction unit 1001 sets the loop variable i = 0 (S802).

In S803, the model construction unit 1001 creates an empty list M of model identification data groups (S803). For example, the model construction unit 1001 creates a model identification data group M = [ ]. The model identification data group is a collection of data for identifying characteristic parameters of a process model that are corrected in accordance with past operational performance data 101 for each characteristic and characteristic type. The model identification data group will be described later using Figures 9A and 9B.

In S804, the model construction unit 1001 selects the i-th data D[i] from the partial data D (S804). In this case, since i = 0, the model construction unit 1001 selects the record "0, I, A, [α0, β0, γ0...]" with a data ID (#) of 0 from the partial data D.

In S805, the model construction unit 1001 determines whether characteristic parameters other than characteristic l of characteristic type k have been calculated for data D[i] (S805). At this point, characteristic parameters other than the type of product, i.e., characteristic parameters for equipment, have not been calculated. Therefore, the model construction unit 1001 determines that characteristic parameters other than characteristic l of characteristic type k have not been calculated for data D[i] (S805; No), and proceeds to S809.

In S806, the model construction unit 1001 creates a process model m by correcting the process model P using each calculated characteristic parameter (S806). In this case, since the determination in S805 is No as described above, the model construction unit 1001 skips this process.

In S807, [process model m, data d[i]] is added to the model identification data group M (S807). In this case, since the determination in S805 is No as described above, the model construction unit 1001 skips this process.

In S808, the model construction unit 1001 executes i = i + 1 (S808).

In S809, the model construction unit 1001 determines whether i >= the number of data D (S809). In this case, i = 1 and the number of data D is 2, so the model construction unit 1001 determines that i >= the number of data (S809; No) and returns to S804. Thereafter, the model construction unit 1001 skips the processing of S807 and S808. In the next loop, i = 2 and i >= 2, so the model construction unit 1001 determines that i >= the number of data (S809; Yes) and proceeds to S810.

In S810, the model construction unit 1001 determines whether the list of model identification data groups M is empty (S810). Here, the model construction unit 1001 determines that the list of model identification data groups M is still empty (S810; Yes) and proceeds to S811.

In S811, the model construction unit 1001 adds [process model P, data D] to the model identification data group M (S811). For example, the model construction unit 1001 creates a model identification data group M = [[process model P, data D]] in which the records "0, I, A, [α0, β0, γ0...]" and "1, I, B, [α1, β1, γ1...]" selected in the loop from S804 to S809 are associated with the process model P created in S702 using the initial values.

In S812, the model construction unit 1001 calculates correction parameters C that minimize the objective function for each model and data pair in the model identification data group M (S812). For example, the model construction unit 1001 calculates correction parameters C for the process model P that minimize the difference between the measured control variables γ0 and γ1 of each data D and the estimated control variables of the process model P. The correction parameters C are parameters for correcting the parameters a, b, and c of the process model 1004 shown in FIG. 4, such as C = {a: +1, b = -1, c = +3}.

In S813, the model construction unit 1001 stores the correction parameters calculated in S812 as characteristic parameters for characteristic l and characteristic type k (S813). For example, the model construction unit 1001 records the characteristic parameters for characteristic type "I" of characteristic "Variety" = {a: +1, b = -1, c = +3}. Once the processing up to this point is complete, the process returns to S705, where characteristic parameters are calculated for a different characteristic type "II" for the same characteristic "Variety" and stored in the characteristic parameter group 1003 (S706, Figures 8A and 8B).

In the following Loop 2, we will explain the case where characteristic list L = [Device] and characteristic type list K = [] have not been processed, and characteristic l = product type and characteristic type k = ro are to be processed.

In S801, the model construction unit 1001 extracts data D for characteristic l = characteristic type k from the past operation performance data 101 (S801). For example, the model construction unit 1001 extracts data for product type B from the past operation performance data 101 and creates partial data D with a data ID (#) of 2. In this case, from the past operation performance data 101 shown in FIG. 3, the record "2, B, B, [α2, β2, γ2...]" identified by the data ID (#) of 2 is extracted as partial data D. In this case, the model construction unit 1001 reassigns the data ID (#) to identify the record of partial data D, and creates partial data D as the record "0, B, B, [α2, β2, γ2...]".

In S802 and S803, the model construction unit 1001 sets the loop variable i = 0, as in the case of product type I, and creates an empty list M of model identification data groups.

In S804, the model construction unit 1001 selects the i-th data D[i] from the partial data D (S804). In this case, since i = 0, the model construction unit 1001 selects the record "0, B, B, [α2, β2, γ2...]" with a data ID (#) of 0 from the partial data D.

In S805, the model construction unit 1001 determines whether characteristic parameters for data D[i] other than characteristic l of characteristic type k have been calculated (S805). Here, as in the case of type I, characteristic parameters for characteristics other than the type, i.e., characteristics of equipment, have not yet been calculated, so the process proceeds to S809.

In S806 and S807, as in the case of variety I, the determination in S805 is No as described above, so the model construction unit 1001 skips this process.

In S808, the model construction unit 1001 executes i = i + 1 (S808).

In S809, the model construction unit 1001 determines whether i >= the number of data D (S809). In this case, i = 1 and the number of data D is 1, so the model construction unit 1001 determines that i >= the number of data (S809; Yes) and proceeds to S810.

In S810, the model construction unit 1001 determines whether the list of model identification data groups M is empty (S810). Here, the model construction unit 1001 determines that the list of model identification data groups M is still empty (S810; Yes) and proceeds to S811.

In S811, the model construction unit 1001 adds [process model P, data D] to the model identification data group M (S811). For example, the model construction unit 1001 creates a model identification data group M = [[process model P, data D]] in which the process model P created in S702 using the initial values is associated with the records "0, B, B, [α2, β2, γ2...]" selected in the loop from S804 to S809.

In S812, the model construction unit 1001 calculates correction parameters C that minimize the objective function for each model and data pair in the model identification data group M (S812). For example, the model construction unit 1001 calculates correction parameters C for the process model P that minimize the difference between the measured control variable γ2 of data D and the estimated control variable of the process model P. The correction parameters C are parameters for correcting the parameters a, b, and c of the process model 1004 shown in FIG. 4, such as C = {a: +2, b = +1, c = +10}.

In S813, the model construction unit 1001 stores the correction parameters calculated in S812 as characteristic parameters for characteristic l and characteristic type k. (S813). For example, the model construction unit 1001 records the characteristic parameters for characteristic type "L" of characteristic "Variety" = {a: +2, b = +1, c = +10}. When the processing up to this point is complete, the process returns to S704, where characteristic parameters are calculated for characteristic type "A" of a different characteristic "Device" and stored in the characteristic parameter group 1003 (S706, Figures 8A and 8B).

In the following loop 3, the characteristic list L = [] and characteristic type list K = [B] have not been processed, and characteristic l = device and characteristic type k = A will be processed.

In S801, the model construction unit 1001 extracts data D for characteristic l = characteristic type k from the past operation performance data 101 (S801). For example, the model construction unit 1001 extracts data for equipment A from the past operation performance data 101 and creates partial data D consisting of a data ID (#) of 0. In this case, from the past operation performance data 101 shown in Figure 3, the record "0, I, A, [α0, β0, γ0...]" identified by a data ID (#) of 0 is extracted as partial data D.

In S802 and S803, the model construction unit 1001 sets the loop variable i = 0, as in the cases of types A and B, and creates an empty list M of the model identification data group.

In S804, the model construction unit 1001 selects the i-th data D[i] from the partial data D (S804). Here, since i = 0, the model construction unit 1001 selects the record "0, I, A, [α0, β0, γ0...]" with a data ID (#) of 0 from the partial data D.

In S805, the model construction unit 1001 determines whether characteristic parameters other than characteristic l of characteristic type k have been calculated for data D[i] (S805). In this case, characteristic parameters other than those for the equipment, i.e., the characteristic for the variety, have been calculated, so the process proceeds to S806.

In S806, the model construction unit 1001 creates a process model m by correcting the process model P using each calculated characteristic parameter (S806). For example, the model construction unit 1001 corrects the process model P using the characteristic parameters whose characteristic is product type and whose characteristic type is I, which were stored in S813 of loop 1, to create the process model m.

In S807, the model construction unit 1001 adds [process model m, data d[i]] to the model identification data group M (S807). For example, the model construction unit 1001 creates a model identification data group M = [[process model m, data d[0] (#0)]] in which the record "0, I, A, [α0, β0, γ0...]" selected in S804 is associated with the process model m created in S806. Through the processing of S807, the process model m corrected with the calculated characteristic parameters is associated with the data D[i] currently being processed (record "0, I, A, [α0, β0, γ0...]") and added to the list of model identification data groups M.

In S808, the model construction unit 1001 executes i = i + 1 (S808).

In S809, the model construction unit 1001 determines whether i >= the number of data D (S809). In this case, i = 1 and the number of data D is 1, so the model construction unit 1001 determines that i >= the number of data (S809; Yes) and proceeds to S810.

In S810, the model construction unit 1001 determines whether the list of model identification data groups M is empty (S810). Here, the model construction unit 1001 created model identification data for process model m in S807 and added this model identification data to the list of model identification data groups M. Therefore, the model construction unit 1001 determines that the list of model identification data groups M is not empty (S810; No) and proceeds to S812.

In S811, since the determination in S810 is No as described above, the model construction unit 1001 skips this process.

In S812, the model construction unit 1001 calculates correction parameters C that minimize the objective function for each model and data pair in the model identification data group M (S812). For example, the model construction unit 1001 calculates correction parameters C for process model m that minimize the difference between the measured control variable γ0 of data d[0] and the estimated control variable of process model m. The correction parameters C are parameters for correcting the parameters a, b, and c of the process model 1004 shown in FIG. 4, such as C = {a: -1, b = +1, c = +1}.

In S813, the model construction unit 1001 stores the correction parameters calculated in S812 as characteristic parameters for characteristic l and characteristic type k (S813). For example, the model construction unit 1001 records the characteristic parameters for characteristic type "A" of characteristic "equipment" = {a: -1, b = +1, c = +1}. Once the processing up to this point is complete, the process returns to S704, where characteristic parameters are calculated for a different characteristic type "B" for the same characteristic "equipment" and stored in the characteristic parameter group 1003 (S706, Figures 8A and 8B).

In the following Loop 4, we will explain the case where the characteristic list L = [] and the characteristic type list K = [] have not been processed, and characteristic l = device and characteristic type k = B are to be processed.

In S801, the model construction unit 1001 extracts data D for characteristic l = characteristic type k from the past operation performance data 101 (S801). For example, the model construction unit 1001 extracts data for equipment B from the past operation performance data 101 and creates partial data D consisting of data IDs (#) of 1 and 2. At this time, from the past operation performance data 101 shown in FIG. 3, records "1, I, B, [α1, β1, γ1...]" and "2, B, B, [α2, β2, γ2...]" identified by data IDs (#) of 1 and 2 are extracted as partial data D. At this time, the model construction unit 1001 reassigns the data IDs (#) to identify the records of partial data D, and creates partial data D as records "0, I, B, [α1, β1, γ1...]" and "1, B, B, [α2, β2, γ2...]".

In S802 and S803, the model construction unit 1001 sets the loop variable i = 0, as in the cases of product type A, B, and device A, and creates an empty list M of the model identification data group.

In S804, the model construction unit 1001 selects the i-th data D[i] from the partial data D (S804). In this case, since i = 0, the model construction unit 1001 selects the record "0, I, B, [α1, β1, γ1...]" with a data ID (#) of 0 from the partial data D.

In S805, the model construction unit 1001 determines whether characteristic parameters other than characteristic l of characteristic type k have been calculated for data D[i] (S805). In this case, characteristic parameters other than those for the equipment, i.e., the characteristic for the variety, have been calculated, so the process proceeds to S806.

In S806, the model construction unit 1001 creates a process model m by correcting the process model P using each calculated characteristic parameter (S806). For example, the model construction unit 1001 corrects the process model P using the characteristic parameters whose characteristic is product type and whose characteristic type is I, which were stored in S813 of loop 1, to create the process model m.

In S807, the model construction unit 1001 adds [process model m, data d[i]] to the model identification data group M (S807). For example, the model construction unit 1001 creates a model identification data group M = [[process model m, data d[0] (#0)]] in which the record "0, I, B, [α1, β1, γ1...]" selected in S804 is associated with the process model m created in S806. By the processing of S807, the process model m corrected with the calculated characteristic parameters is added to the list of model identification data groups M in association with the data D[i] (record "0, I, B, [α1, β1, γ1...]") currently being processed.

In S808, the model construction unit 1001 executes i = i + 1 (S808).

In S809, the model construction unit 1001 determines whether i >= the number of data D (S809). In this case, i = 1 and the number of data D is 2, so the model construction unit 1001 determines that i >= the number of data (S809; No) and returns to S804. Thereafter, the model construction unit 1001 skips the processing of S806 and S807. In the next loop, i = 2 and i >= 2, so the model construction unit 1001 determines that i >= the number of data (S809; Yes) and proceeds to S810. Upon completing the next loop, the model construction unit 1001 creates a model identification data group M = [[process model m, data d[1] (#1)]] in which the record "2, b, B, [α2, β2, γ2...]" selected in S804 of the next loop is associated with the process model m created in S806. Finally, the model identification data group M [[process model m(a), data d[0](#0)]], [[process model m(b), data d[1](#1)]] is created.

In S810, the model construction unit 1001 determines whether the list of model identification data groups M is empty (S810). Here, the model construction unit 1001 created model identification data for process model m in S807 and added this model identification data to the list of model identification data groups M. Therefore, the model construction unit 1001 determines that the list of model identification data groups M is not empty (S810; No) and proceeds to S812.

In S811, since the determination in S810 is No as described above, the model construction unit 1001 skips this process.

In S812, the model construction unit 1001 calculates correction parameters C that minimize the objective function for each model and data pair in the model identification data group M (S812). For example, the model construction unit 1001 calculates correction parameters C for process model m that minimize the difference between the measured control variable γ1 of data d[0] and the estimated control variable of process model m(a), and the difference between the measured control variable γ2 of data d[1] and the estimated control variable of process model m(b). The correction parameters C are parameters for correcting the parameters a, b, and c of the process model 1004 shown in FIG. 4, such as C = {a: +0, b = +0, c = +1}.

In S813, the model construction unit 1001 stores the correction parameters calculated in S812 as characteristic parameters for characteristic l and characteristic type k. (S813). For example, the model construction unit 1001 records the characteristic parameters for characteristic type "B" of characteristic "equipment" = {a: +0, b = +0, c = +1}. When the processing up to this point is complete, the characteristic parameter calculation processing shown in Figures 8A and 8B has been performed for all characteristics and all characteristic types. Therefore, the process proceeds to S707 of the model construction processing shown in Figure 7.

In S707, since processing has already been performed for each of the characteristic type lists K = [A, B], the model construction unit 1001 determines that the characteristic type list K is an empty set (S707; Yes) and proceeds to S708.

In S708, since processing has already been performed for each of the characteristic list L = [product type, equipment], the model construction unit 1001 determines that the characteristic list L is an empty set (S708; Yes) and terminates processing.

Figures 9A and 9B are diagrams showing an example of a flowchart illustrating the processing steps for model identification processing of process models recorded in the model identification data group in the characteristic parameter calculation processing shown in Figures 8A and 8B. This processing is performed when called from the characteristic parameter calculation processing. For example, it is called within the processing of S812 of the characteristic parameter calculation processing. The model identification processing is processing for performing identification on pairs of process models and data recorded in the model identification data group in the characteristic parameter calculation processing.

8A and 8B, the model identification data group M stores multiple pairs of data (e.g., [process model P, data D], [process model m, data d[i]]) in which a process model and data are associated with each other. In this example, the model identification data group M stores [[process model m(a), data d[0] (#0)]] and [[process model m(b), data d[1] (#1)]]. The model identification data group M stores a process model corrected using the characteristic parameters for each calculated characteristic and the actual values of the characteristics of the controlled object included in the past operation performance data 101 used when constructing the pre-correction process model. Data D and data d[i] are partial data of the past operation performance data 101, and as explained above, the actual values of the reference value α, manipulated variable β, and controlled variable γ are recorded.

In the process described below, for each data pair, the model construction unit 1001 first corrects the process model with characteristic parameters. Next, the model construction unit 1001 calculates the difference between the data (e.g., controlled variable) corresponding to the corrected process model and the output estimate of the process model (e.g., the controlled variable estimate obtained by inputting a reference value/manipulated variable) using an arbitrary objective function. As the arbitrary objective function, for example, RMSE (Root Mean Squared Error) can be used.

The model construction unit 1001 then finally calculates the difference between the process model and each actual value for each pair of data, and calculates the average value of the differences calculated for each pair of data. If the calculated average value meets a predetermined criterion, such as being sufficiently small, the model construction unit 1001 terminates the process and returns it as the final characteristic parameter. On the other hand, if the above-mentioned arbitrary criterion is not met, the model construction unit 1001 starts over from determining the characteristic parameter. Specific explanations are provided below using the flowchart.

In S901, the model construction unit 1001 selects the candidate characteristic parameters and characteristic parameters calculated in S812 based on an arbitrary optimization algorithm (S901).

In S902, the model construction unit 1001 sets the loop variable j = 0 (S902).

In S903, the model construction unit 1001 creates an empty list V that stores the results of the objective function (S903). For example, the model construction unit 1001 creates an objective function result list V = [ ].

In S904, the model construction unit 1001 obtains the process model m and data d from the jth list M[j] recorded in the model identification data group M (S904).

In S905, the model construction unit 1001 corrects the process model m acquired in S904 with the characteristic parameters selected in S901 to acquire the process model m' (S905).

In S906, the model construction unit 1001 calculates the value v of an arbitrary objective function using the process model m' corrected in S905 and the data d corresponding to the process model m obtained in S904, and adds the value v to the objective function result list V (S906).

In S907, the model construction unit 1001 executes j = j + 1 (S907).

In S908, the model construction unit 1001 determines whether the length of the model identification data group M is greater than or equal to j (S908). In other words, the model construction unit 1001 determines whether j has reached the number of records stored as the model identification data group M.

If the model construction unit 1001 determines that j has not reached the number of records stored as the model identification data group M (S908; No), it returns to S904 and repeats the subsequent processing. On the other hand, if the model construction unit 1001 determines that j has reached the number of records stored as the model identification data group M (S908; Yes), it proceeds to S909.

In S909, the model construction unit 1001 calculates the average value a of the values v of any objective function recorded in the objective function result list V in S906, and evaluates the calculated average value a by, for example, comparing it with a predetermined standard (S909).

In S910, the model construction unit 1001 determines whether the calculated average value a satisfies a predetermined criterion (S910). If the model construction unit 1001 determines that the calculated average value a does not satisfy the predetermined criterion (S910; No), the process returns to S901 and the characteristic parameters are reselected.

In S911, if the model construction unit 1001 determines that the calculated average value a meets the predetermined criteria (S910; Yes), it determines that the characteristic parameter selected in S901 is the final characteristic parameter and outputs it (S911).

As explained above, once the processing in Figures 7 to 9B is complete, a new process model can be constructed in which the parameters of the process model constructed using theoretical formulas are corrected for each characteristic based on past operational performance data. This makes it possible to represent the dynamic characteristics of the controlled object without having to create a process model for each combination of characteristics. Therefore, even for controlled objects with a huge number of characteristic combinations, process simulation can be performed using a small number of process models.

Figure 10 is a diagram showing an example of a flowchart illustrating the processing procedure of the simulation processing performed by the simulation unit 1002 in Example 1. This processing is performed when a user issues an instruction or operates to start the processing. The simulation processing is a process of performing a process simulation using the process model constructed in the model construction processing performed by the model construction unit 1001.

In S1001, the simulation unit 1002 acquires characteristic parameters that match the input characteristic information 102 from the characteristic parameter group 1003 stored in Figures 7 to 9B (S1001).

In S1002, the simulation unit 1002 corrects the process model P using the acquired characteristic parameters and creates a process model P' (S1002).

In S1003, the simulation unit 1002 inputs the simulation input information 103 into the process model P' created in S1002, and executes a process simulation of the control object using the process model P' (S1003).

In S1004, the simulation unit 1002 returns the results of the process simulation (S1004).

In this manner, in this embodiment, a process model construction system that constructs a process model that models the characteristics of a control object using a computer includes a first processing unit (e.g., model construction unit 1001, Figure 7) that acquires the characteristics (e.g., product type, equipment, etc.) from past operational performance data (e.g., past operational performance data 101) that indicates operational performance according to the characteristics of the control object, and a second processing unit (e.g., model construction unit 1001, Figure 8) that calculates characteristic parameters (e.g., characteristic parameters stored in the characteristic parameter group 1003 shown in Figure 5) for correcting the process model for each characteristic so that the actual value of the process of the control object (e.g., the controlled variable of the controlled object) for the characteristic included in the past operational performance data and the estimated value of the process of the control object estimated by the process model constructed using a predetermined theoretical formula (e.g., the theoretical formula shown in Figure 4) satisfy a predetermined condition (e.g., the condition that the difference between the actually measured controlled variable and the estimated controlled variable of the process model is minimized). This makes it possible to build process models that impose less construction and operational burden than conventional methods, even when the controlled object has multiple characteristics, and also reduces the number of process models that need to be built to perform process simulation.

The system also includes a third processing unit (e.g., simulation unit 1002) that creates a corrected process model using the characteristic parameters of the same characteristics as characteristic information (e.g., characteristic information 102) that indicates the characteristics of the controlled object and the characteristic types included in the characteristics and is input to the process model construction system, and simulates the behavior of the process of the controlled object using the created process model. This makes it possible to perform process simulation using a small number of process models constructed by the model construction unit 1001, even for controlled objects with an enormous number of combinations of characteristics and characteristic types.

The above describes Example 1 in detail, but the following modifications are possible depending on the environment in which the system is used:

For example, the model construction unit 1001 may correct the parameters of the process model in various ways.

In the above-described embodiment, the model construction unit 1001 calculated characteristic parameters for correcting the parameters of the process model constructed using the theoretical formula for each characteristic of the controlled object based on past operational performance data. However, the model construction unit 1001 may calculate statistical values based on the calculated characteristic parameters and correct the parameters of the process model constructed using the theoretical formula using statistical characteristic parameters based on the calculated statistical values. Examples of the statistical values include statistical values obtained by the model construction unit 1001 performing processes such as correcting the characteristic parameters stored in the characteristic parameter group 1003 using a weighted linear sum, correcting the characteristic parameters using a weighted multiplication, or correcting using a correction factor for the initial parameters of the process model. The weighting and correction factor may be determined in advance by the user.

In this way, the second processing unit (e.g., model construction unit 1001, FIG. 8) calculates statistical values based on the characteristic parameters calculated for each characteristic, and corrects the process model constructed using the predetermined theoretical formula using the statistical characteristic parameters based on the statistical values. This allows the parameters of the process model to be corrected using the characteristic parameters corrected using each of the above-mentioned methods, as specified in advance by the user.

In addition, the initial values of the process model parameters may be identified and determined from past operational performance data.

In the above-described embodiment, the initial values of the process model parameters were described as being given in advance. However, for example, the model construction unit 1001 may set the initial values to the process model parameters that have been corrected using the characteristic parameters obtained by the model identification process shown in Figures 9A and 9B, which have been identified from past operational performance data.

In this way, the second processing unit (e.g., model construction unit 1001, FIG. 8) creates model identification data that associates the process model corrected with the characteristic parameters with actual values included in the past operational performance data used to construct the uncorrected process model, and performs model identification processing (e.g., the model identification processing shown in FIGS. 9A and 9B) that identifies pairs of the process model corrected with the characteristic parameters included in the created model identification data and the actual values. The first processing unit sets the parameters of the process model corrected with the characteristic parameters obtained by the model identification processing as initial values of the parameters of the process model. This makes it possible to automatically set initial values of the parameters of the process model based on the past operational performance data, without the user having to specify them in advance.

Alternatively, characteristic parameters may be manually specified by the user.

In the above-described embodiment, the model construction unit 1001 calculated characteristic parameters for correcting the parameters of the process model constructed using a theoretical formula for each characteristic of the controlled object based on past operational performance data. However, the model construction unit 1001 may also store characteristic parameters specified by the user in the characteristic parameter group. Alternatively, the model construction unit 1001 may store some of the calculated characteristic parameters (for example, {a: +0, b = +0} of characteristic parameters = {a: +0, b = +0, c = +1}) and characteristic parameters specified by the user (for example, {c = +1} of characteristic parameters = {a: +0, b = +0, c = +1}) in the characteristic parameter group.

In this way, the process model construction system 1000 has an input unit (e.g., input device 1606) that accepts input of the characteristic parameters, and the second processing unit (e.g., model construction unit 1001, Figure 8) sets at least some of the characteristic parameters to the parameters input from the input unit and performs the calculations for parameters other than the set parameters. This makes it possible to directly store characteristic parameters that reflect the user's intentions.

As described above, this system can be applied to various plants that perform process operations. In this case, the first processing unit (e.g., model construction unit 1001, FIG. 7) acquires a first device and a first product type that indicate the characteristics of the controlled object (e.g., reactor characteristics) from the past operational performance data. The second processing unit (e.g., model construction unit 1001, FIG. 8) calculates, for each of the first device and the first product type, characteristic parameters such that the actual values of the controlled object process (e.g., reactor control variables) and the estimated values of the controlled object process for the first device and the first product type satisfy a predetermined condition (e.g., the condition that the difference between the measured control variables for the reactor and the estimated control variables of the process model is minimized). The system corrects the process model used to generate the first product type in the first device based at least on the calculated characteristic parameters. This enables the construction of process models that impose less construction and operational burden than conventional methods in various plants that perform process operations, and also reduces the number of process models constructed for process simulation.

In Example 2, we will explain a case where the results of a simulation performed by the simulation unit 1002 are used to back-calculate an operating method for a controlled object that satisfies optimal operating conditions, and the results are fed back to the controlled object.

11 is a diagram showing an example of the configuration of a process model construction system according to the second embodiment. As shown in FIG. 11, the process model construction system 2000 according to this embodiment further includes a control device input calculation unit 2001 in addition to the configuration of the process model construction system 1000 according to the first embodiment. The process model construction system 2000 also accepts input of desired operating condition data 106 for the controlled object, specified by the user. As in the first embodiment, the process model construction system 2000 shown in FIG. 11 can be realized by, for example, a general computer 1600 as hardware, as shown in FIG. 2 (schematic diagram of a computer). Below, the same components as those in the process model construction system 1000 according to the first embodiment are designated by the same reference numerals, and their description will be omitted.

The control device input calculation unit 2001 is a processing unit that reverse-calculates and determines input values for the control object 105 that satisfy the desired operating condition data 106 for a given control object. The control device input calculation unit 2001 uses an arbitrary optimization algorithm to calculate input values for the control object 105 through trial and error, performing an optimization calculation to find input values that match the desired operating conditions. For example, the control device input calculation unit 2001 sequentially re-executes the optimization calculation using feedback results of the controlled variables obtained from the control object 105, and calculates optimal input values that match the operating status of the control object 105. Examples of input values for the control object 105 include time-series data of manipulated variables, reference values, and controlled variables (or data sets combining some or all of these). Furthermore, examples of the desired operating condition data 106 include exemplary operational performance data selected by the user from the operational performance data 301 of the past operational performance data 101 shown in Figure 3.

As such, this embodiment includes a fourth processing unit (e.g., control device input calculation unit 2001) that uses a predetermined algorithm (e.g., an arbitrary optimization algorithm) and feedback results of the control variables for the control object obtained from the control object to calculate input values for the control object (e.g., time-series data of manipulated variables, reference values, and controlled variables (or a data set combining some or all of these)) and performs optimization calculations to find input values that match the desired operating conditions for the control object (e.g., operating condition data 106). That is, the control device input calculation unit 2001 recalculates the input values for the control object that satisfy the desired operating conditions each time based on the feedback results obtained from the control object, thereby calculating input values that satisfy the optimal operating conditions for the current control object. Therefore, as with Example 1, it is possible to back-calculate manipulated variables that satisfy the optimal operating conditions even when there are many types and combinations of characteristics.

In Example 3, we will explain how to estimate characteristic parameters for unknown characteristics (e.g., product type or equipment) and perform process simulation.

12A is a diagram showing an example of the configuration of a process model construction system according to Example 3. As shown in FIG. 12A, the process model construction system 3000 according to this example includes, in addition to the configuration of the process model construction system 1000 according to Example 1, a characteristic parameter model construction unit 3001, a simulation unit 3002 that is different from those in Examples 1 and 2, and a characteristic parameter model 3003. The process model construction system 3000 also accepts input of additional characteristic information 107. As in Examples 1 and 2, the process model construction system 3000 shown in FIG. 12A can be realized by, for example, a general computer 1600 as hardware, as shown in FIG. 2 (schematic diagram of a computer). Below, the same components as those in the process model construction system 1000 and the process model construction system 2000 according to Examples 1 and 2 are designated by the same reference numerals, and their description will be omitted.

The characteristic parameter model construction unit 3001 is a processing unit that uses the characteristic parameter group 1003 and the characteristic additional information 107 to construct a machine learning model that estimates characteristic parameters from the characteristic additional information 107. For example, when the factors (e.g., the ratio of raw materials) that affect a certain characteristic (e.g., the variety manufactured by the controlled object) are known, the characteristic parameter model construction unit 3001 acquires the variety and the ratio of raw materials as the characteristic additional information 107. The characteristic parameter model construction unit 3001 uses the acquired characteristic additional information 107 as explanatory variables and the characteristic parameters as objective variables to construct a characteristic parameter model that serves as a parameter estimator using any machine learning or other method. This makes it possible to estimate characteristic parameters even for unknown characteristic types and estimate the dynamic characteristics of a process model.

The additional characteristic information 107 is data that is referenced when constructing the characteristic parameter model described above or when estimating characteristic parameters for unknown characteristics.

Figure 12B is a diagram showing an example of characteristic addition information 107. As shown in Figure 12B, characteristic addition information 107 stores an unknown characteristic (e.g., an unknown product variety produced by the controlled object) in association with feature information corresponding to that characteristic (e.g., the raw materials of the product variety produced by the controlled object). Figure 12B shows, for example, that the breakdown of the raw materials for product variety I produced by the controlled object is 30% raw material 1, ... and 20% raw material k. While the example given here is of raw materials of the product variety, various information that represents the characteristics of the characteristics, such as equipment setting values, may be defined as characteristic addition information.

When correcting the process model with characteristic parameters that match the characteristics of the characteristic information 102, the simulation unit 3002 estimates characteristic parameters for unknown characteristic types (such as varieties with no past operating history) using the characteristic addition information 107 and the characteristic parameter model described above. For example, in S1001 shown in FIG. 10, the simulation unit 3002 inputs the input characteristic information 102 into the characteristic parameter model constructed using the characteristic addition information 107, and obtains estimated values of characteristic parameters that match the characteristic information 102. Furthermore, in S1002, the simulation unit 3002 corrects the process model using the obtained estimated values of the characteristic parameters, and thereafter performs simulation in the same manner as in S1003 and S1004.

In this way, this embodiment has a fifth processing unit (e.g., characteristic parameter model construction unit 3001) that constructs a characteristic parameter model that estimates the characteristic parameters for the characteristic by machine learning using characteristic information (e.g., characteristic additional information 107) that corresponds the characteristic of the controlled object (e.g., an unknown variety manufactured by the controlled object) with factors that affect the characteristic (e.g., characteristic information such as the raw materials of the variety manufactured by the controlled object) as explanatory variables and the characteristic parameters that correct the process model of the controlled object for the characteristic as objective variables, and the third processing unit (e.g., simulation unit 3002) simulates the behavior of the process of the controlled object using the process model corrected using the estimated value of the characteristic parameter obtained by inputting the characteristic information into the characteristic parameter model. That is, the characteristic parameter model construction unit 3001 constructs a characteristic parameter model using the characteristic additional information 107 (for example, if the characteristic is a variety, the raw material ratio of that variety, etc.) and the characteristic parameter group 1003, and the simulation unit 3002 corrects the process model using estimated values of the characteristic parameters obtained by inputting the characteristic information 102 into the constructed characteristic parameter model, and the simulation unit 3002 performs a simulation using the corrected process model. Therefore, it is possible to estimate characteristic parameters and perform process simulation even for unknown varieties or equipment. In other words, it becomes possible to estimate dynamic characteristics and perform process simulation even for unknown types of characteristics or combinations of characteristics.

In Example 4, for control targets with a small amount of past operational performance data, there are fewer combinations of characteristics and characteristic types, and certain combinations may result in inconsistencies in the characteristic parameters. Below, we will explain how to resolve inconsistencies in the characteristic parameters when they arise.

FIG. 13A is a diagram showing an example of the configuration of a process model construction system according to Example 4. As shown in FIG. 13A, in addition to the configuration of the process model construction system 1000 according to Example 1, the process model construction system 4000 according to this example further includes model identification information 4001 and a model construction unit 4002 that differs from Examples 1 to 3. As in Examples 1 to 3, the process model construction system 4000 shown in FIG. 13A can be realized by, for example, a general computer 1600 as shown in FIG. 2 (schematic diagram of a computer) in terms of hardware. In the following, the same components as those in the process model construction system 1000, process model construction system 2000, and process model construction system 3000 according to Examples 1 to 3 are designated by the same reference numerals, and their description will be omitted.

The model identification information 4001 is information referenced by the model construction unit 4002 when performing model identification as shown in Figures 9A and 9B, and is a collection of data for identifying a characteristic model of a process model that is corrected according to the past operational performance data 101 for each characteristic and characteristic type. In Example 1, characteristic parameters were calculated for all parameters of the process model as the characteristic parameter group 1003 shown in Figure 5, but in this example, only the parameters of the process model that correspond to the characteristic or characteristic type are stored as the characteristic parameters for that characteristic or characteristic type.

Figure 13B is a diagram showing an example of model identification information 4001. As shown in Figure 13B, model identification information 4001 stores one or more characteristic parameters of a process model corresponding to a characteristic or characteristic type. In Figure 13B, for example, correction parameters {a: +1.2, b = -0.1}, which are characteristic parameters for type A, and correction parameters {c = 0}, which are characteristic parameters for type B, are stored in association with the type of each characteristic type. Here, of the parameters a, b, and c that make up the characteristic parameters, the characteristic parameter for type A is associated with a and b, and the characteristic parameter for type B is associated with c and stored.

Next, we will explain the model identification process in this embodiment. However, in Example 4, only the step corresponding to S901 differs from the model identification process shown in Example 1 (Figures 9A and 9B). Therefore, the flowchart will be omitted and only the parts that differ from S901 will be explained.

In S901 of the model identification process shown in Figures 9A and 9B, the model construction unit 4002 selects, from the characteristic parameters stored in S813, characteristic parameters that correspond to the characteristics and characteristic types defined in the model identification information 4001, based on an arbitrary optimization algorithm. The characteristic parameters that correspond to the characteristics and characteristic types can be selected by referring to the model identification information 4001. The model identification information 4001 may be set in advance by the user by referring to the characteristic parameters stored in S813.

When the model construction unit 4002 selects the characteristic parameters determined according to the characteristics and characteristic type as described above, it executes the processing from S902 onwards. At this time, in S905, the model construction unit 4002 acquires a process model m' by correcting the process model m acquired in S904 with the characteristic parameters according to the characteristics and characteristic type selected in the step corresponding to S901 in the fourth embodiment described above.

As such, in this embodiment, the process model construction system 4000 has model identification information (e.g., model identification information 4001) storing the characteristic parameters of the process model corresponding to one or more characteristics, and the second processing unit (e.g., model construction unit 4002) creates model identification data that associates the process model corrected using the characteristic parameters calculated for each characteristic with actual values included in the past operation performance data 101 used to construct the uncorrected process model, and executes a model identification process (e.g., the model identification process shown in FIGS. 9A and 9B) that performs identification on pairs of the process model corrected using the characteristic parameters included in the created model identification data and the actual values, and in the model identification process, selects the characteristic parameters defined in the model identification information based on a predetermined algorithm (e.g., an arbitrary optimization algorithm), and corrects the process model using the selected characteristic parameters.

In other words, in this embodiment, in addition to the configuration of embodiment 1, characteristic parameters corresponding to the characteristics and characteristic types defined in the model identification information 4001 are used to correct only the parameters of the process model corresponding to those characteristic parameters. As a result, even when there is little data on combinations of characteristics and characteristic types and inconsistencies occur in the characteristic parameters for certain combinations, it is possible to obtain output values of the process model with high accuracy and suppress increases in prediction errors in the process model.

In the above example, characteristic parameters corresponding to the characteristics and characteristic types are defined as model identification information 4001. However, the model construction unit 4002 may set a target value for the objective function value v calculated during model identification in S906, for example, by using a percentage error for the objective function, and setting it so that identification is performed up to a percentage error of N% during model identification of product characteristics, and a percentage error of 0% is the target during model identification of equipment characteristics. This makes it possible to obtain the output value of the process model with high accuracy, as above, without defining model identification information 4001.

In Example 5, we explain how to perform a simulation and correct the process model when a discrepancy occurs between the behavior of the actual process and the behavior of the process model.

Figure 14 is a diagram showing an example of the configuration of a process model construction system in Example 5. As shown in Figure 14, the process model construction system 5000 according to this example has a simulation unit 5001 that is different from the configuration of the process model construction system 1000 in Example 1 and that differs from Examples 1 to 4. In addition, the process model construction system 5000 receives as input the feedback results of the control amount obtained from the control target 1401.

The process model construction system 5000 shown in FIG. 14 can be realized, as in the cases of Examples 1 to 4, by a general-purpose computer 1600 as hardware, as shown in FIG. 2 (schematic diagram of a computer). Below, the same components as those in the process model construction system 1000, process model construction system 2000, process model construction system 3000, and process model construction system 4000 in Examples 1 to 4 are designated by the same reference numerals, and their description will be omitted.

The simulation unit 5001 compares the simulation results with the control results of the controlled object 1401 (e.g., the control amount obtained from the controlled object 1401) and determines whether the difference exceeds a certain value temporarily (e.g., 10 seconds) or cumulatively (e.g., 1 month). If the simulation unit 5001 determines that the difference exceeds a certain value temporarily or cumulatively, it corrects and recalculates the simulation results and saves the results in the simulation results 104. For example, the simulation unit 5001 can correct the simulation results by identifying process model parameters that minimize the difference between the control results of the controlled object 1401 and the simulation results, or by reselecting characteristic parameters from a group of characteristic parameters that satisfy such conditions.

Figure 15 is a diagram showing an example of a flowchart illustrating the processing steps of the simulation processing performed by the simulation unit 5001 in Example 5. This processing is performed when a user issues an instruction or operates to start the processing. As in Example 1, the simulation processing is a process of performing a process simulation using the process model constructed in the model construction processing performed by the model construction unit 1001.

Since the processes in S1501 and S1502 are the same as those in S1001 and S1002 shown in Figure 10, only S1503 and onwards will be explained.

In S1503, the simulation unit 5001 inputs the simulation input information 103 into the process model P' created in S1502, executes a simulation of the control object using the process model P', and obtains the results S of the executed simulation (S1503).

In S1504, the simulation unit 5001 acquires the actual measurement value sequence T from the control object 1401 (S1504). The actual measurement value sequence T is time-series data included in the operational performance data, such as the actual measurement control variables γ0, γ1, γ2, etc.

In S1505, the simulation unit 5001 determines whether the difference between the simulation result S and the actual measurement value sequence T is equal to or greater than an arbitrary standard that serves as a predetermined threshold (S1505). In S1505, if the simulation unit 5001 determines that the difference between the simulation result S and the actual measurement value sequence T is not equal to or greater than an arbitrary standard that serves as a predetermined threshold (S1505; No), the process proceeds to S1507. On the other hand, in S1505, if the simulation unit 5001 determines that the difference between the simulation result S and the actual measurement value sequence T is equal to or greater than an arbitrary standard that serves as a predetermined threshold (S1505; Yes), the process proceeds to S1506.

In S1506, the simulation unit 5001 corrects the characteristic parameters of the process model used in S1502 so as to minimize the difference between the results of the simulation of the control object 1401 using the process model and the actual measurement value sequence T of the control object 1401, and then performs the simulation again (S1506).

In S1507, the simulation unit 5001 returns the results of the simulation or the results of a re-run simulation (S1507).

In this way, in this embodiment, if the difference between the simulation result and the control result of the control object exceeds a predetermined threshold, the third processing unit (e.g., simulation unit 5001) corrects the characteristic parameters of the process model to reduce the difference, and re-executes the simulation.

In other words, the simulation unit 5001 checks the difference between the results of the process simulation and the actual controlled variable output by the controlled object, and if this difference exceeds a certain amount, corrects the characteristic parameters of the process model so that the waveforms of the simulation results up to that point and the actual controlled variable become closer. This makes it possible to perform process simulation using a small number of process models even for controlled objects with a huge number of combinations of characteristics and characteristic types, and if a deviation occurs between the actual measured values of the controlled object after a certain period of simulation, the characteristic parameters of the process model can be re-identified based on the actual measured values obtained from the controlled object.

In the example described above, the simulation unit 5001 corrected and optimized the characteristic parameters to minimize the difference between the results of the simulation of the control object 1401 using the process model and the actual measurement value sequence T of the control object 1401. However, the simulation may be performed again by correcting any characteristic parameter selected by the user from the group of characteristic parameters.

In Example 6, we will explain the use of machine learning models for process models and characteristic parameters.

Figure 16 is a diagram showing an example of the configuration of a process model construction system in Example 6. As shown in Figure 16, the process model construction system 6000 according to this example has a model construction unit 6001 and a simulation unit 6002 that are different from those in Examples 1 to 5, compared to the configuration of the process model construction system 1000 in Example 1, and has a characteristic model group 6003 instead of the characteristic parameter group 1003.

The process model construction system 6000 shown in FIG. 16 can be realized, as in the cases of Examples 1 to 5, by a general-purpose computer 1600 as hardware, as shown in FIG. 2 (schematic diagram of a computer). Below, the same components as those in the process model construction system 1000, process model construction system 2000, process model construction system 3000, process model construction system 4000, and process model construction system 5000 in Examples 1 to 5 are designated by the same reference numerals, and their description will be omitted.

The model construction unit 6001 is a processing unit that constructs the process model 1004 and the characteristic model group 6003 using machine learning based on the past operational performance data 101.

The simulation unit 6002 is a processing unit that calls the process model 1004 and the characteristic model group 6003 and executes a simulation based on the input characteristic information 102 and simulation input information 103. The characteristic model group 6003 is a collection of characteristic models constructed for each characteristic based on the difference between the process model constructed by machine learning and past operational performance data.

Figure 17 is a diagram showing an example of a flowchart illustrating the processing steps of the model construction process performed by the model construction unit 6001 in Example 6. This process is performed when a user issues an instruction or operation to start the process. The model construction process is a process for calculating a characteristic model based on the past operation performance data 101, and is primarily a process for preparing to construct the characteristic model. In the following, the process model will be described as being average value data of the past operation performance data 101.

In S1701, the model construction unit 6001 acquires the process model 1004 and past operational performance data 101 (S1701).

In S1702, the model construction unit 6001 constructs a process model P from the average values of the past operation performance data 101 (S1702). For example, the model construction unit 6001 calculates the average values of the reference values, manipulated variables, and controlled variables, which are time-series data included in the operation performance data accumulated as the past operation performance data 101, and creates this calculated average value data as the process model P. This process model P is an initial machine learning model for constructing a machine learning model, which will be described later using Figures 18A and 18B.

In S1703, the model construction unit 6001 obtains the characteristic list L from the past operational performance data 101, as in Example 1.

In S1704, as in Example 1, the model construction unit 6001 extracts characteristic l without restoration from characteristic list L obtained in S1703, and obtains characteristic type list K for characteristic l (S1703, S1704).

In S1705, the model construction unit 6001 extracts, without restoration, characteristic type k from the characteristic type list K obtained in S1704, as in the first embodiment (S1705).

In S1706, the model construction unit 6001 calculates a characteristic model for characteristic type k acquired in S1705 and stores it in the characteristic model group 6003 (S1706). For example, the model construction unit 6001 calculates a characteristic model for variety I. A specific method for calculating a characteristic model will be described later using Figures 18A and 18B.

In S1707, the model construction unit 6001 determines whether the characteristic type list K is an empty set (S1707). For example, if the model construction unit 6001 is calculating a characteristic model for the characteristic type I of a variety, it determines that the characteristic type list K = [II] exists and therefore determines that it is not an empty set (S1707; No), and returns to S1705. The model construction unit 6001 then executes S1705 and S1706 for the characteristic type list K = [II]. At this time, as in Example 1, processing has been performed for each of the characteristic type lists K = [I, II]. Therefore, the model construction unit 6001 determines that the characteristic type list K is an empty set (S1707; Yes), and proceeds to S1708.

In S1708, the model construction unit 6001 determines whether the characteristic list L is an empty set (S1708). For example, if the model construction unit 6001 is processing characteristic list L = [product type], it determines that the characteristic list L = [equipment] exists and therefore is not an empty set (S1708; No), and returns to S1704. The model construction unit 6001 then executes S1704, S1705, and S1706 for l = [equipment]. At this time, as in the first embodiment, processing has been performed for each of the characteristic lists L = [product type, equipment]. Therefore, the model construction unit 6001 determines that the characteristic list L is an empty set (S708; Yes), and terminates processing.

Figures 18A and 18B are examples of flowcharts showing the processing steps of the characteristic model calculation process performed in S1706 shown in Figure 17. The characteristic model calculation process is a process for constructing a characteristic model.

In the characteristic model calculation process, the model construction unit 6001 extracts data D whose characteristic type matches the characteristics of the characteristic model to be constructed from the past operation performance data 101. As explained below, data D and data d[i] are partial data of the past operation performance data 101, as in Example 1, and as explained above, the actual values of the reference value α, manipulated variable β, and controlled variable γ are recorded.

If characteristic models for other characteristics have not been constructed, the model construction unit 6001 creates differential data between data D and the output estimate of process model P (e.g., a controlled variable estimate obtained by inputting a reference value/manipulated variable), and constructs a machine learning model using this as the objective variable. As an arbitrary objective function, for example, RMSE can be used. If the operational performance data included in the past operational performance data 101 is time series data of controlled variables, the process model P is average time series data of the controlled variables. The model construction unit 6001 creates differential data d between the time series data of each controlled variable in data D and the average time series data, and constructs a time series prediction model for the set of data d. For explanatory variables, the actual controlled variables may be obtained from the controlled object, and the first window of data may be provided, or data D may be expanded and dummy data for one window of data may be inserted. Alternatively, additional information may be provided in the past operational performance data 101.

On the other hand, if another characteristic model has already been constructed, the model construction unit 6001 subtracts the output of the constructed model when calculating the difference between each data D and the process model P, constructs a characteristic model using that data D, and stores the constructed characteristic model in the characteristic model group 6003.

The following is a specific explanation using the past operational performance data 101 shown in Figure 3. In the explanation using Figures 18A and 18B, as in Example 1, a controlled object with two characteristics including two different characteristic types will be used as an example, and specific explanations will be given in the order of Loop 1 to Loop 4. As mentioned above, the explanation will be based on the assumption that the process model is a model identified using all of the past operational performance data 101, or time-series data of the average value of the controlled variable constructed from the average value of the controlled variable at each time.

In Loop 1, the characteristic list L = [Equipment] and characteristic type list K = [Ro] have not been processed, and characteristic l = product type and characteristic type k = I will be processed.

In S1801, the model construction unit 6001 extracts partial data D for characteristic l = characteristic type k from the past operation performance data 101, as in Example 1 (S801). For example, from the past operation performance data 101 shown in Figure 3, the records "0, I, A, [α0, β0, γ0...]" and "1, I, B, [α1, β1, γ1...]", whose data IDs (#) are identified by 0 and 1, respectively, are extracted as partial data D.

In S1802 and S1803, the model construction unit 6001 sets the loop variable i = 0, as in Example 1, and creates an empty list M of the model identification data group (S1802, S1803). The model identification data group is subjected to the same processing as the model identification processing shown in Figures 9A and 9B.

In S1804, the model construction unit 6001 selects the i-th data D[i] from the partial data D, as in the first embodiment (S1804). Here, since i = 0, the model construction unit 1001 selects the record "0, I, A, [α0, β0, γ0...]" with a data ID (#) of 0 from the partial data D.

In S1805, the model construction unit 6001 determines whether a characteristic model other than characteristic l of characteristic type k has been constructed for data D[i] (S1805). At this point, the model construction unit 1001 determines that a characteristic model other than characteristic l of characteristic type k has not been constructed for data D[i] (S1805; No), and proceeds to S1809.

In S1806, the model construction unit 6001 creates differential data g between the value obtained by adding the output of each constructed characteristic model to the output of the process model P and the data d[i] (S1806). In this case, since the determination in S1805 is No as described above, the model construction unit 6001 skips this process.

In S1807, the differential data g is added to the model identification data group (S1807). In this case, since the determination in S805 is No as described above, the model construction unit 6001 skips this process.

In S1808, the model construction unit 6001 executes i = i + 1 (S1808).

In S1809, the model construction unit 6001 determines whether i >= the number of data D (S1809). In this case, i = 1 and the number of data D is 2, so the model construction unit 6001 determines that i >= the number of data (S1809; No) and returns to S1804. Thereafter, the model construction unit 1001 skips the processing of S807 and S808. In the next loop, i = 2 and i >= 2, so as in Example 1, the model construction unit 6001 determines that i >= the number of data (S1809; Yes) and proceeds to S1810.

In S1810, the model construction unit 6001 determines whether the list of model identification data groups M is empty (S1810). Here, the model construction unit 6001 determines that the list of model identification data groups M is still empty (S1810; Yes) and proceeds to S1811.

In S1811, the model construction unit 6001 adds difference data g between the output of the process model P and the control amount of data D to the model identification data group M (S1811). For example, the model construction unit 6001 creates a model identification data group M = [data g], which is the difference between the output value of the process model P using the initial value created in S1702 and each of the records "0, I, A, [α0, β0, γ0...]" and "1, I, B, [α1, β1, γ1...]" selected in the loop from S1804 to S1809.

In S1812, the model construction unit 6001 constructs a regression model s using machine learning for the data set of the model identification data group (S1812). For example, the model construction unit 6001 constructs a regression model s using the reference value/operation amount of data D as the explanatory variable and the control amount of data D as the objective variable.

In S1813, the model construction unit 6001 stores the regression model s calculated in S1812 as a characteristic model for characteristic l and characteristic type k (S1813). For example, the model construction unit 6001 records a characteristic model for characteristic type "I" of characteristic "Variety". Once the processing up to this point is complete, the process returns to S1705, where a characteristic model is constructed for a different characteristic type "II" for the same characteristic "Variety" and stored in the characteristic model group 6003 (S1706, Figures 18A and 18B).

In the following Loop 2, we will explain the case where characteristic list L = [Device] and characteristic type list K = [] have not been processed, and characteristic l = product type and characteristic type k = ro are to be processed.

In S1801, the model construction unit 6001 extracts data D for characteristic l = characteristic type k from the past operation performance data 101 (S1801). For example, the model construction unit 6001 extracts data for product type B from the past operation performance data 101 and creates partial data D with a data ID (#) of 2. In this case, from the past operation performance data 101 shown in Figure 3, the record "2, B, B, [α2, β2, γ2...]" identified by the data ID (#) of 2 is extracted as partial data D. In this case, the model construction unit 1001 reassigns the IDs and creates partial data D as the record "0, B, B, [α2, β2, γ2...]" as in Example 1.

In S1802 and S1803, the model construction unit 6001 sets the loop variable i = 0, as in the case of product type I, and creates an empty list M of model identification data groups.

In S1804, the model construction unit 6001 selects the i-th data D[i] from the partial data D (S1804). In this case, since i = 0, the model construction unit 1001 selects the record "0, B, B, [α2, β2, γ2...]" with a data ID (#) of 0 from the partial data D.

In S1805, the model construction unit 6001 determines whether characteristic models have been constructed for data D[i] other than characteristic l of characteristic type k (S1805). Here, as in the case of product type I, characteristic models have not yet been constructed for characteristics other than the product type, i.e., for equipment characteristics, so the process proceeds to S1809.

In S1806 and S1807, as in the case of variety I, the determination in S1805 is No as described above, so the model construction unit 6001 skips this process.

In S1808, the model construction unit 6001 executes i = i + 1 (S1808).

In S1809, the model construction unit 6001 determines whether i >= the number of data D (S1809). In this case, i = 1 and the number of data D is 1, so the model construction unit 6001 determines that i >= the number of data (S1809; Yes) and proceeds to S1810.

In S1810, the model construction unit 1001 determines whether the list of model identification data groups M is empty (S1810). Here, the model construction unit 6001 determines that the list of model identification data groups M is still empty (S1810; Yes) and proceeds to S1811.

In S1811, the model construction unit 1001 adds difference data g between the output value of the process model P and the control amount of data D to the model identification data group M (S1811). For example, the model construction unit 6001 creates a model identification data group M = [data g], which is the difference between the output value of the process model P using the initial value created in S1702 and each of the records "0, B, B, [α2, β2, γ2...]" selected in the loop from S1804 to S1809.

In S1812, the model construction unit 6001 constructs a regression model s using machine learning for the data set of the model identification data group M (S1812). For example, the model construction unit 6001 constructs a regression model s using the reference value/operation amount of data D as the explanatory variable and the control amount of data D as the objective variable.

In S1813, the model construction unit 6001 stores the regression model s calculated in S1812 as a characteristic model for characteristic l and characteristic type k (S1813). For example, the model construction unit 1001 records a characteristic model for characteristic type "B" of characteristic "Variety". Once the processing up to this point is complete, the process returns to S1704, where a characteristic model is constructed for characteristic type "A" of a different characteristic "Equipment" and stored in the characteristic model group 6003 (S1706, Figures 18A and 18B).

In the following loop 3, the characteristic list L = [] and characteristic type list K = [B] have not been processed, and characteristic l = device and characteristic type k = A will be processed.

In S1801, the model construction unit 6001 extracts data D for characteristic l = characteristic type k from the past operation performance data 101 (S1801). For example, the model construction unit 6001 extracts data for equipment A from the past operation performance data 101 and creates partial data D consisting of a data ID (#) of 0. In this case, from the past operation performance data 101 shown in Figure 3, the record "0, I, A, [α0, β0, γ0...]" identified by a data ID (#) of 0 is extracted as partial data D.

In S1802 and S1803, the model construction unit 6001 sets the loop variable i = 0, as in the cases of types A and B, and creates an empty list M of the model identification data group.

In S1804, the model construction unit 6001 selects the i-th data D[i] from the partial data D (S1804). In this case, since i = 0, the model construction unit 6001 selects the record "0, I, A, [α0, β0, γ0...]" with a data ID (#) of 0 from the partial data D.

In S1805, the model construction unit 6001 determines whether characteristic models have been constructed for data D[i] other than characteristic l of characteristic type k (S1805). In this case, characteristic models have been constructed for characteristics other than equipment, i.e., the characteristic is for the product type, so the process proceeds to S1806.

In S1806, the model construction unit 6001 creates difference data g between the value obtained by adding the output of each constructed characteristic model to the output of process model P and data d[i] (S1806). For example, the model construction unit 6001 creates new difference data g as the difference between the control amount of data d[i] and the sum of the output of the characteristic model whose characteristic is product type and characteristic type is I, which was stored in S1813 of loop 1, and the output of process model P.

In S1807, the model construction unit 6001 adds the above-mentioned difference data g to the model identification data group M (S1807). For example, the model construction unit 6001 sets the model identification data group M = [g]. By the processing of S1807, new difference data g, which is the difference between the constructed characteristic model and data d[i], is associated with the currently processed data D[i] (record "0, I, A, [α0, β0, γ0...]") and added to the list of model identification data group M.

In S1808, the model construction unit 6001 executes i = i + 1 (S1808).

In S1809, the model construction unit 6001 determines whether i >= the number of data D (S1809). In this case, i = 1 and the number of data D is 1, so the model construction unit 6001 determines that i >= the number of data (S1809; Yes) and proceeds to S1810.

In S1810, the model construction unit 6001 determines whether the list of model identification data group M is empty (S1810). Here, the model construction unit 6001 added the new differential data g to the list of model identification data group M in S1807. Therefore, the model construction unit 6001 determines that the list of model identification data group M is not empty (S1810; No) and proceeds to S1812.

In S1811, since the determination in S1810 is No as described above, the model construction unit 6001 skips this process.

In S1812, the model construction unit 6001 constructs a regression model s using machine learning for the data set of the model identification data group (S1812). For example, the model construction unit 6001 constructs a regression model s using the reference value/operation amount of the new difference data g as the explanatory variable and the control amount of the new difference data g as the objective variable.

In S1813, the model construction unit 6001 stores the regression model s calculated in S1812 as a characteristic model for characteristic l and characteristic type k (S1813). For example, the model construction unit 1001 records a characteristic model for characteristic type "A" of characteristic "equipment." Once the processing up to this point is complete, the process returns to S1704, where a characteristic model is constructed for a different characteristic type "B" for the same characteristic "equipment" and stored in the characteristic model group 6003 (S1706, Figures 18A and 18B).

In the following Loop 4, we will explain the case where the characteristic list L = [] and the characteristic type list K = [] have not been processed, and characteristic l = device and characteristic type k = B are to be processed.

In S1801, the model construction unit 6001 extracts data D for characteristic l = characteristic type k from the past operation performance data 101 (S1801). For example, the model construction unit 6001 extracts data for equipment B from the past operation performance data 101 and creates partial data D consisting of data IDs (#) of 1 and 2. At this time, from the past operation performance data 101 shown in FIG. 3, records "1, A, B, [α1, β1, γ1...]" and "2, B, B, [α2, β2, γ2...]" identified by data IDs (#) of 1 and 2 are extracted as partial data D. At this time, the model construction unit 6001 reassigns IDs, as in the case of Example 1, and creates partial data D as records "0, A, B, [α1, β1, γ1...]" and "1, B, B, [α2, β2, γ2...]".

In S1802 and S1803, the model construction unit 6001 sets the loop variable i = 0, as in the cases of product type A, B, and device A, and creates an empty list M of the model identification data group.

In S1804, the model construction unit 6001 selects the i-th data D[i] from the partial data D (S1804). In this case, since i = 0, the model construction unit 1001 selects the record "0, I, B, [α1, β1, γ1...]" with a data ID (#) of 0 from the partial data D.

In S1805, the model construction unit 6001 determines whether characteristic models have been constructed for data D[i] other than characteristic l of characteristic type k (S1805). In this case, characteristic models have been constructed for characteristics other than equipment, i.e., the characteristic is for the product type, so the process proceeds to S1806.

In S1806, the model construction unit 6001 creates difference data g between the value obtained by adding the output of each constructed characteristic model to the output of process model P and data d[i] (S1806). For example, the model construction unit 6001 creates new difference data g as the difference between the control amount of data d[0] and the sum of the output of the characteristic model whose characteristic is product type and characteristic type is I, which was stored in S1813 of loop 1, and the output of process model P.

In S1807, the model construction unit 6001 adds data g to the model identification data group M (S1807). For example, the model construction unit 6001 sets the model identification data group M = [g(I)]. Through the processing of S1807, new differential data g, which is the difference between the constructed characteristic model and data d[0], is associated with the currently processed data D[0] (record "0, I, B, [α1, β1, γ1...]") and added to the list of model identification data group M.

In S1808, the model construction unit 6001 executes i = i + 1 (S1808).

In S1809, the model construction unit 6001 determines whether i >= the number of data D (S1809). In this case, i = 1 and the number of data D is 2, so the model construction unit 1001 determines that i >= the number of data (S1809; No) and returns to S1804. Thereafter, the model construction unit 6001 skips the processing of S1807 and S1808. In the next loop, i = 2 and i >= 2, so the model construction unit 6001 determines that i >= the number of data (S1809; Yes) and proceeds to S1810. After completing the next loop, the model construction unit 6001 creates data g by adding the output of the characteristic model for product type B to the output of the process model P created in S1806 and calculating the difference between the control amount of data d[1] and data g. Finally, a model identification data group M = [g(A), g(B)] is created.

In S1810, the model construction unit 6001 determines whether the list of model identification data group M is empty (S1810). Here, the model construction unit 6001 added the new differential data g(i) to the list of model identification data group M in S1807. Therefore, the model construction unit 6001 determines that the list of model identification data group M is not empty (S1810; No) and proceeds to S1812.

In S1811, since the determination in S1810 is No as described above, the model construction unit 6001 skips this process.

In S1812, the model construction unit 6001 constructs a model s using machine learning on the data set of the model identification data group (S1812). For example, the model construction unit 6001 constructs a regression model s using the reference values/operation quantities of the new difference data g(a) and g(b) as explanatory variables and the control quantities of the new difference data g(a) and g(b) as objective variables.

In S1813, the model construction unit 6001 stores the regression model s calculated in S1812 as the characteristic model for characteristic l and characteristic type k (S1813). For example, the model construction unit 1001 records the characteristic model for characteristic type "B" of characteristic "equipment." When the processing up to this point is complete, the characteristic model construction process shown in Figures 18A and 18B has been performed for all characteristics and all characteristic types. Therefore, the process proceeds to S1707 of the model construction process shown in Figure 17.

In S1707, since processing has already been performed for each of the characteristic type lists K = [A, B], the model construction unit 6001 determines that the characteristic type list K is an empty set (S1707; Yes) and proceeds to S1708.

In S1708, since processing has already been performed for each of the characteristic list L = [product type, equipment], the model construction unit 6001 determines that the characteristic list L is an empty set (S1708; Yes) and terminates processing.

As described above, in this embodiment, a process model construction system that constructs a process model that models the characteristics of a control object using a computer includes a first processing unit (e.g., model construction unit 6001, FIG. 17) that acquires the characteristics (e.g., product type, equipment, etc.) from past operational performance data (e.g., past operational performance data 101) that indicates operational performance according to the characteristics of the control object, and a second processing unit (e.g., model construction unit 6001, FIG. 18) that calculates, for each characteristic, a characteristic model (e.g., regression model s) that corrects the process model based on the difference between the actual value of the process of the control object (e.g., the controlled variable of the control object) for the characteristic included in the past operational performance data and the estimated value of the control object estimated by the process model constructed using predetermined machine learning (e.g., any machine learning such as linear regression). In other words, a characteristic model for each characteristic is constructed based on the process model constructed using machine learning and the difference between the process model and the past operational performance data, and the calculation results of the process model constructed using machine learning are corrected using the calculation results of the constructed characteristic model. Therefore, as in the first embodiment, even when a control object has multiple characteristics, it is possible to construct a process model that imposes less construction and operation load than conventional methods, and it is also possible to reduce the number of process models constructed for performing process simulation. Furthermore, as in the first embodiment, even for control objects with an enormous number of combinations of characteristics and characteristic types, it is possible to perform process simulation using a small number of process models constructed by the model construction unit 6001.

In this example, the case where the entire process model is replaced with a machine learning model has been described, but some of the mathematical formulas in the process model may also be replaced with a machine learning model. Furthermore, in this example, the differential data is an example of a controlled variable, but it is not limited to this and other measured values or values that can be inferred from measured values may also be used. Furthermore, in this example, the case where differential data is prepared as training data when constructing an equipment or product type model has been described, but instead, appropriate training data (such as corresponding measured values specified by the user) may be prepared for one or more characteristics or characteristic types (for example, equipment or product type) as shown in Example 4.

In Example 7, the input/output interface in the model construction process shown in Example 1 will be described. As already explained, in the model construction process, the model construction unit 1001 receives past operation performance data 101 as input data, and reads out the process model 1004 and the initial values of the parameters of the process model 1004. Then, in the model construction process, the model construction unit 1001 outputs characteristic parameters for each characteristic.

Figure 19 shows an example of a screen (model construction screen) that the model construction unit 1001 displays on the display of a computer that constitutes the process model construction system 1000 (e.g., the output device 1605 of the computer 1600 shown in Figure 2) when executing the model construction process.

As shown in FIG. 19, the model construction unit 1001 displays a model construction screen W1901 that includes input past operational performance data 101, a process model 1004, and initial values for the process model parameters, which are input data for the model construction process. The user checks this input data on the model construction screen and instructs execution of the model construction process by, for example, pressing an execute button (not shown) on the screen.

Also, as shown in FIG. 19, the model construction unit 1001 displays a model construction screen W1902 that includes a characteristic parameter group 1003, which is the output data of the model construction process. The user can check the output data on the model construction screen and confirm the characteristic parameters for each characteristic and characteristic type.

As described above, this embodiment includes a model construction unit (e.g., model construction unit 1001) that displays on a computer screen, as before and after processing, input information including past operational performance data (e.g., past operational performance data 101) indicating operational performance according to the characteristics of the controlled object (e.g., product type, equipment, etc.), a process model (e.g., a process model represented by the theoretical formula shown in FIG. 4) for modeling the characteristics of the controlled object obtained from the past operational performance data, and output information including characteristic parameters (e.g., characteristic parameters stored in characteristic parameter group 1003 shown in FIG. 5) for correcting the process model calculated for each characteristic of the controlled object so that the operational performance and the estimated value of the controlled object estimated by the process model satisfy a predetermined condition (e.g., the condition that the difference between the actual control amount and the estimated control amount of the process model is minimized). In other words, when the model construction unit performs the model construction process, the model construction screen described above is displayed. Therefore, the user can quickly grasp the calculated characteristic parameter values for each characteristic and characteristic type for the input process model.

In this embodiment, a screen is used as an interface when accepting the input data from the user or when presenting the output data to the user. However, this is not limited to a screen. If the input device 1606 includes an audio input/output device such as a microphone or speaker, other user-recognizable interfaces may be used, such as accepting the input data or presenting the output data via audio through the device.

In Example 8, the input/output interface in the simulation process shown in Example 1 will be described. As already explained, in the simulation process, the simulation unit 1002 receives, as input data, characteristics such as the product type and equipment, and the characteristic types included therein, as characteristic information 102. It also receives time-series data on the manipulated variable of the controlled object as simulation input information 103, which is the desired operating condition. Then, in the simulation process, the simulation unit 1002 outputs, for example, time-series data on the controlled variable as the result of the simulation.

Figure 20 shows an example of a screen (simulation execution screen) that the simulation unit 1002 displays on the display of a computer that constitutes the process model creation system 1000 (e.g., the output device 1605 of the computer 1600 shown in Figure 2) when executing a simulation process.

As shown in FIG. 20, the simulation unit 1002 displays a simulation execution screen W2001 that includes simulation input information 103, which is input data for the simulation process, and characteristic information 102. The user checks this input data on the simulation execution screen and instructs execution of the simulation process by, for example, pressing an execution button (not shown) on the screen.

Also, as shown in FIG. 20, the simulation unit 1002 displays a model construction screen W2002 containing the simulation results, which are the output data of the simulation process. The user can check the output data on the simulation execution screen and confirm the simulation results, such as the time series changes in the controlled variables.

As described above, in this embodiment, the third processing unit (e.g., simulation unit 1002) creates input information including the characteristic information (e.g., characteristic information 102) and simulation input information (e.g., simulation input information 103) defining the desired operating conditions for the simulation, creates a process model corrected using the characteristic parameters with the same characteristics as the characteristic information, inputs the simulation input information into the created process model, and executes the simulation. The output information includes the simulation results (e.g., time-series data of the controlled variable), and displays these output information on the computer screen as the pre- and post-processing states. That is, when the simulation unit 1002 performs simulation processing, it displays the simulation execution screen described above. Therefore, the user can quickly grasp the results of a simulation using a process model corrected using characteristic parameters obtained from the input characteristic information. In this example, time-series data of the manipulated variable of the controlled object is provided, but a reference value may also be provided.

In this embodiment, a screen is used as an interface when accepting the above input data from the user or when presenting output data to the user. However, as in Example 7, other types of interfaces that are recognizable to the user may also be used.

In Example 9, an input/output interface is described for the case where an operating method (e.g., manipulated variable) for a controlled object that satisfies the optimal operating conditions shown in Example 2 is calculated backward, and the result is fed back to the controlled object. As already explained, in Example 2, the control device input calculation unit 2001 receives, as input data, desired operating condition data 106 for the controlled object, and characteristics of the controlled object such as product type and equipment as characteristic information 102, as well as the characteristic types included therein. The control device input calculation unit 2001 calculates, for example, an input (manipulated variable in this example) that matches the optimal operating conditions, which are the desired operating condition data 106, with the results of a simulation performed by the simulation unit 1002, and outputs the time-series data of the manipulated variable as the input value of the controlled object.

Figure 21 shows an example of a screen (feedback execution screen) displayed on the display of a computer constituting the process model creation system 1000 (e.g., output device 1605 of computer 1600 shown in Figure 2) when the control equipment input calculation unit 2001 back-calculates an operating method for the controlled object that satisfies the optimal operating conditions.

As shown in FIG. 21, the control device input calculation unit 2001 displays a feedback execution screen W2101 that includes the desired operating condition data 106, which is the input data for performing the above-mentioned feedback, and characteristic information 102. The user checks this input data on the feedback execution screen and, by pressing an execute button (not shown) on the screen, instructs the reverse calculation of an operating method for the controlled object that satisfies the optimal operating conditions.

Also, as shown in FIG. 21, the control device input calculation unit 2001 displays a feedback execution screen W2102 containing the feedback results, which are the output data of the above-mentioned inverse calculation. The user can check the output data on the feedback execution screen and confirm the feedback results, such as changes in the manipulated variable over time.

In this way, in this embodiment, the fourth processing unit (e.g., control device input calculation unit 2001) displays on the computer screen, as the state before and after processing, input information including desired operating conditions for the controlled object (e.g., operating condition data 106), characteristic information indicating the characteristics of the controlled object and the characteristic types included in the characteristics (e.g., characteristic information 102), and output information including input values for the controlled object (e.g., time series data of manipulated variables) that match the results of simulating the behavior of the process of the controlled object using a process model corrected using the characteristic parameters of the same characteristics as the characteristic information.

In other words, the control device input calculation unit 2001 displays the feedback execution screen described above when it back-calculates the operating method of the controlled object that satisfies the optimal operating conditions and feeds back the results to the controlled object. Therefore, the user can grasp at a glance the feedback results of control using input values to the controlled object that satisfy the optimal operating conditions for the controlled object at the current time.

In this embodiment, an example has been given of a case where a screen is used as an interface when accepting the above input data from the user or when presenting output data to the user. However, as in Examples 7 and 8, other types of interfaces that are recognizable to the user may also be used. While this example describes the case of back-calculating an operation amount, it may also be applied in the same way to back-calculating a reference value.

The present invention is not limited to the above-described embodiments, and in the implementation stage, the components may be modified and embodied within the scope of the gist of the invention, or multiple components disclosed in the above-described embodiments may be combined as appropriate.

1000 to 6000 Process model construction system 1001 Model construction unit 1002 Simulation unit 1003 Characteristic parameter group 1004 Process model 2001 Control device input calculation unit 3001 Characteristic parameter model construction unit 3002 Simulation unit 4001 Model identification information 4002 Model construction unit 6001 Model construction unit 6002 Simulation unit

Claims (15)

A process model construction system that constructs a process model that models characteristics of a control target by a computer, comprising:
a first processing unit that acquires a plurality of different types of characteristics from past operation performance data that indicates operation performance according to the characteristics of the control object, which are elements that affect the behavior of the process model; and
a second processing unit that calculates, for each characteristic, a characteristic parameter for correcting the process model such that an actual value of the process to be controlled for the characteristic included in the past operation result data and an estimated value of the process to be controlled estimated by the process model constructed using a predetermined theoretical formula satisfy a predetermined condition;
the second processing unit corrects the process model based on the characteristic parameters associated with combinations of a plurality of different types of the characteristics;
A process model construction system comprising: a third processing unit that creates a corrected process model using the characteristic parameters of the same characteristics as the characteristics included in characteristic information that indicates the characteristics of the controlled object and characteristic types included in the characteristics and that is input to the process model creation system, and that simulates the behavior of the controlled object process using the created process model;
2. The process model construction system according to claim 1, further comprising: a fourth processing unit that calculates an input value for the controlled object using a predetermined algorithm and a feedback result of a control amount for the controlled object obtained from the controlled object, and performs an optimization calculation to find an input value that matches a desired operating condition for the controlled object;
2. The process model construction system according to claim 1, further comprising: a fifth processing unit that constructs a characteristic parameter model for estimating the characteristic parameter of the characteristic by machine learning using characteristic additional information that associates the characteristic of the controlled object with feature information that is an element that affects the characteristic and corresponds to the characteristic for estimating the characteristic as an explanatory variable, and the characteristic parameter that corrects a process model of the controlled object for the characteristic as an objective variable;
the third processing unit simulates the behavior of the process to be controlled by using the process model corrected using the estimated value of the characteristic parameter obtained by inputting the characteristic additional information into the characteristic parameter model.
3. The process model construction system according to claim 2. the process model construction system has model identification information in which the characteristic parameters of the process model corresponding to one or more characteristics are stored;
the second processing unit creates model identification data that associates the process model corrected by the characteristic parameters with actual values included in the past operation result data used when constructing the process model before correction, and performs a model identification process that performs identification on a pair of the process model corrected by the characteristic parameters included in the created model identification data and the actual values;
In the model identification process, the characteristic parameters defined in the model identification information are selected based on a predetermined algorithm, and the process model is corrected with the selected characteristic parameters.
2. The process model construction system according to claim 1. the second processing unit calculates a statistical value based on the characteristic parameter calculated for each of the characteristics, and corrects the process model constructed using the predetermined theoretical formula by using the statistical characteristic parameter based on the calculated statistical value.
2. The process model construction system according to claim 1. the second processing unit creates model identification data that associates the process model corrected by the characteristic parameters with actual values included in the past operation result data used when constructing the process model before correction, and performs a model identification process that performs identification on a pair of the process model corrected by the characteristic parameters included in the created model identification data and the actual values;
the first processing unit sets the parameters of the process model corrected by the characteristic parameters obtained by the model identification processing as initial values of the parameters of the process model.
2. The process model construction system according to claim 1. the process model construction system has an input unit that accepts input of the characteristic parameters,
the second processing unit sets at least some of the characteristic parameters to parameters input from the input unit, and performs the calculation for parameters other than the set parameters;
2. The process model construction system according to claim 1. when a difference between the result of the simulation and the control result of the control object exceeds a predetermined threshold, the third processing unit corrects the characteristic parameters of the process model so as to reduce the difference, and re-executes the simulation.
3. The process model construction system according to claim 2. A process model construction system that constructs a process model that models characteristics of a control target by a computer, comprising:
a first processing unit that acquires a plurality of different types of characteristics from past operation performance data that indicates operation performance according to the characteristics of the control object, which are elements that affect the behavior of the process model; and
a second processing unit that calculates, for each characteristic, a characteristic model obtained by correcting the process model based on a difference between an actual value of the process of the controlled object for the characteristic included in the past operation result data and an estimated value of the controlled object estimated by the process model constructed by predetermined machine learning,
the second processing unit corrects the process model based on the characteristic model associated with a combination of a plurality of different types of the characteristics;
A process model construction system comprising: Input information including past operational performance data indicating operation performance according to characteristics of a control object, which is an element that affects the behavior of the process model, and process models for modeling the characteristics of a plurality of different types of the control object obtained from the past operational performance data;
output information including characteristic parameters for correcting the process model, which are calculated for each characteristic of the controlled object so that the operation record and the estimated value of the controlled object estimated by the process model satisfy predetermined conditions , the output information including the characteristic parameters related to a combination of a plurality of different types of the controlled object, and the output information including the characteristic parameters related to a combination of a plurality of different types of the controlled object, are displayed on a computer screen as states before and after processing.
A process model construction system comprising: the third processing unit creates a corrected process model using input information including the characteristic information and simulation input information defining desired operating conditions for performing a simulation, and the characteristic parameters having the same characteristics as the characteristic information, and inputs the simulation input information into the created process model to execute the simulation, and displays output information including the results of the simulation obtained by executing the simulation on a screen of the computer as states before and after processing.
3. The process model construction system according to claim 2. the fourth processing unit displays, on a screen of the computer, as states before and after processing, input information including desired operating conditions for the controlled object, characteristic information indicating characteristics of the controlled object and characteristic types included in the characteristics, and output information including input values to the controlled object that match results of simulating the behavior of the process of the controlled object using a process model corrected using the characteristic parameters of the same characteristics as the characteristic information.
4. The process model construction system according to claim 3. the first processing unit acquires, from the past operation performance data, a first device and a first product type that indicate characteristics of the control target;
the second processing unit calculates, for each of the first equipment and the first product type, the characteristic parameters such that an actual value of the process to be controlled and an estimated value of the process to be controlled for the first equipment and the first product type satisfy a predetermined condition, and corrects the process model when generating the first product type in the first equipment based on at least the calculated characteristic parameters;
2. The process model construction system according to claim 1. A process model construction method for constructing a process model that models characteristics of a control target by a computer, comprising the steps of:
acquiring a plurality of different types of characteristics from past operation performance data showing operation performance according to the characteristics of the control object, which are elements that affect the behavior of the process model;
calculating, for each characteristic, a characteristic parameter for correcting the process model such that an actual value of the process to be controlled for the characteristic included in the past operation record data and an estimated value of the process to be controlled estimated by the process model constructed using a predetermined theoretical formula satisfy a predetermined condition;
correcting the process model based on the characteristic parameters associated with a plurality of different types of combinations of the characteristics;
A process model construction method comprising: