Estimation device, model generation device, estimation method, model generation method, and program

Abstract

An estimation device (200) identifies the process characteristics of process data to be estimated. The estimation device (200) estimates output data from input data included in the process data to be estimated, using, from among a set of estimation models, an estimation model corresponding to the identified process characteristics of the process data to be estimated.

Description

Estimation device, model generation device, estimation method, model generation method, and program

[0001] This application claims priority to Japanese Patent Application No. 2024-110020, filed on July 9, 2024, the contents of which are incorporated herein by reference in their entirety.

[0002] Patent Document 1 discloses estimating the reliability of power supply using a regression model suitable for the state of the power system.

[0003] JP 2019-88150 A

[0004] However, there are processes, such as steelmaking processes, in which measured values ​​of temperature, pressure, etc. vary significantly depending on the product type, even when the same equipment is used. Thus, even in processes in which the process data can take on various values ​​even when the operating state is not intentionally changed, a method is desired that can accurately estimate the operating state of the process.

[0005] The present disclosure has been made in consideration of the above problems, and aims to accurately estimate the operating state of a process in which process data can take on various values ​​even when the operating state is not intentionally changed.

[0006] According to an aspect of the present disclosure, there is provided an estimation device including: a process characteristic identification unit that identifies process characteristics of process data to be estimated that are acquired in a process having a plurality of process characteristics; and an estimation unit that estimates output data from input data of the process data to be estimated using an estimation model that corresponds to the identified process characteristic of the process data to be estimated, from a set of estimation models that obtain output data from input data selected from process data generated for each of the process characteristics possessed by the process.

[0007] The process characteristic identification unit may identify the process characteristic of the process data to be estimated by identifying to which class the process data to be estimated belongs, out of classes corresponding to process characteristics obtained by clustering process data previously acquired in the process.

[0008] The process characteristic identification unit may factorize a matrix representing the acquired process data of the estimation target with a feature matrix representing an operating state of the process of the set of estimation models to obtain a coefficient matrix, and identify the process characteristics of the process data of the estimation target based on the coefficient matrix.

[0009] According to another aspect of the present invention, there is provided a model generation device including: a data classification processing unit that classifies a plurality of process data acquired in a process having a plurality of process characteristics into a plurality of classes based on the process characteristics; a model generation unit that generates, for each of the plurality of classes, estimation models that obtain output data from input data selected from the process data; and a search unit that evaluates a set of estimation models based on an estimation error of each of the generated estimation models and searches for a set of estimation models for the process based on the evaluation result.

[0010] The data classification processing unit may factorize a matrix representing a plurality of process data into two non-negative matrices, a feature matrix representing the operating state of the process and a coefficient matrix representing the process characteristics, based on the number of categories to be classified, and classify the plurality of process data into a plurality of classes based on the coefficient matrix.

[0011] The search unit may calculate an evaluation value for evaluating the set of estimation models, where the evaluation value is the sum of estimation errors of output data obtained by a plurality of estimation models for each class.

[0012] The search unit may calculate an evaluation value for evaluating a set of estimation models, and while the minimum evaluation value is being updated, the data classification processing unit reclassifies the process data, the model generation unit regenerates estimation models for each of the reclassified classes, the search unit recalculates an evaluation value for evaluating the regenerated set of estimation models, and when the minimum evaluation value is no longer being updated, the search unit may set the set of estimation models for which the evaluation value is calculated as the set of estimation models.

[0013] The model generating unit may generate, as the estimation model, a regression equation that represents the relationship between input data and output data included in the process data.

[0014] According to another aspect of the present disclosure, there is provided an estimation method in which a computer executes processing including: a process characteristic identification step of identifying process characteristics of process data to be estimated that are acquired in a process having a plurality of process characteristics; and an estimation step of estimating output data from input data of the process data to be estimated using an estimation model corresponding to the identified process characteristic of the process data to be estimated, from a set of estimation models that obtain output data from input data selected from process data generated for each of the process characteristics possessed by the process.

[0015] According to another aspect of the present disclosure, there is provided a model generation method in which a computer executes processing including: a data classification processing step of classifying a plurality of process data acquired in a process having a plurality of process characteristics into a plurality of classes based on the process characteristics; a model generation step of generating, for each of the plurality of classes, estimation models that obtain output data from input data selected from the process data; and a search step of evaluating a set of estimation models based on an estimation error of each of the generated estimation models and searching for a set of estimation models for the process based on the evaluation result.

[0016] According to another aspect of the present disclosure, there is provided a program that causes a computer to function as an estimation device, the program including: a process characteristic identification unit that identifies process characteristics of process data to be estimated that are acquired in a process having a plurality of process characteristics; and an estimation unit that estimates output data from input data of the process data to be estimated using an estimation model that corresponds to the identified process characteristic of the process data to be estimated, from a set of estimation models that obtain output data from input data selected from process data generated for each of the process characteristics possessed by the process.

[0017] According to another aspect of the present disclosure, there is provided a program that causes a computer to function as a model generation device, the program including: a data classification processing unit that classifies a plurality of pieces of process data acquired in a process having a plurality of process characteristics into a plurality of classes based on the process characteristics; a model generation unit that generates, for each of the plurality of classes, estimation models that obtain output data from input data selected from the process data; and a search unit that evaluates a set of estimation models based on an estimation error of each of the generated estimation models, and searches for a set of estimation models for the process based on the evaluation result.

[0018] As described above, according to the present disclosure, it is possible to accurately estimate the operational state of a process in which process data can take various values ​​even when the operational state is not intentionally changed.

[0019] FIG. 1 is a conceptual diagram illustrating an example of the relationship between the operational state of a process and process data. FIG. 2 is a block diagram illustrating an example of the configuration of an estimation system. FIG. 3 is a flowchart illustrating an example of an outline of a model generation method. FIG. 4 is a flowchart illustrating an example of a model generation method using NMF. FIG. 5 is a diagram illustrating an example of the relationship between an input data matrix Y and an output data matrix P. FIG. 6 is a diagram illustrating an example of the relationship between the input data matrix Y and the matrix product of a coefficient matrix Φ and an operational state matrix X. FIG. 7 is a diagram illustrating an example of the correspondence relationship between input data stored in the input data matrix Y and classes. FIG. 8 is a flowchart illustrating an example of an outline of an estimation method. FIG. 9 is a flowchart illustrating an example of an estimation method for estimating output data using an estimation model generated using NMF. FIG. 10 is a diagram illustrating an example of numerical values ​​for explaining an example of a model generation method. FIG. 11 is a graph illustrating an example of characteristics of data classified into two classes in the numerical example shown in FIG. 9. FIG. 12 is a diagram illustrating an example of a method for generating estimation models (regression equations) for two classes in the numerical example shown in FIG. 9. FIG. 13 is a diagram illustrating an example of a method for estimating output data in the numerical example shown in FIG. 9. FIG. 14 is a block diagram showing an example of the hardware configuration of an information processing device that functions as a model generating device or an estimation device.

[0020] An embodiment of the present disclosure will be described below with reference to the drawings. In this specification and the drawings, "multiple values ​​compared with each other being the same" may include not only being strictly the same, but also being different within the scope of the present disclosure. For example, if the difference between multiple calculated values ​​is within the range of computer calculation error, the multiple calculated values ​​may be considered to be the same. Furthermore, if the difference between multiple measured values ​​is within the range of error caused by the measurement accuracy of a measuring instrument, the multiple measured values ​​may be considered to be the same. Furthermore, in this specification and the drawings, components having substantially the same functional configuration are designated by the same reference numerals, and redundant description will be omitted.

[0021] [1. System Configuration] The estimation system according to this embodiment is a system for estimating data representing the operational state of a process. The process in this embodiment has multiple process characteristics. The data representing the operational state of the process may be process data. The process data refers to data acquired during the operation of the process. The process data may include, for example, at least one of data indicating the operating conditions of equipment used for the operation of the process, data indicating the type of product manufactured in the process, and measurement data measured using a sensor or the like installed in the process. During the operation of the process, various data are acquired at multiple times. For example, various data are continuously acquired at each time during the operation of the process. Data acquired during the operation of the process in this manner is called process data. Furthermore, the process characteristics refer to the trends in the process data acquired during the operation of the process. More specifically, the process characteristics refer to the trends in the values ​​of the process data continuously acquired at each time during the operation of the process. For example, assume that during the operation of the process, values ​​of five pieces of data (data a, b, c, d, and e) are continuously acquired every second as process data. In this case, the tendency that appears in the combination of the values ​​of the data a to e at each time is an example of a process characteristic.

[0022] In processes where the operating conditions change continuously, there may be multiple trends in process data obtained from a normally operating process, even though humans do not intentionally switch the operating conditions during daily operations. For example, in processes where the operating conditions change continuously, such as processes for manufacturing steel products, the values ​​of process data may vary significantly depending on the product type, even when the same equipment is used. For example, in a casting process using continuous casting equipment, the way in which abnormalities in the solidification state of the solidified shell occur varies depending on the electromagnetic stirring conditions and the type of powder, even though humans do not intentionally switch the operating conditions. Such a process is an example of a process with multiple process characteristics. The steel products described above may be final products or intermediate products.

[0023] 1 is a diagram conceptually illustrating an example of the relationship between the operational state of a process and process data. In FIG. 1, a case where a process has three process characteristics is illustrated. The number of process characteristics that a process has is not limited to three. For example, the number of process characteristics that a process has may be two or more, three or more, or four or more.

[0024] 1 illustrates an example in which, by using information representing the characteristics of each of a plurality of process data PD, the operational state of a process is identified in a space spanned by a minimum number of basis vectors (directional vectors) that rationally explain the relationship between data items included in the plurality of process data. FIG. 1 illustrates an example in which the minimum dimension is three. In this case, each process data PD is classified into one of three classes c1, c2, and c3 based on the information representing the characteristics of each of the plurality of process data PD. The three classes c1, c2, and c3 correspond to one of three process characteristics. The class is information that identifies the classification of data (a class is also referred to as a cluster).

[0025] The information representing the characteristic may be, for example, information indicating the relationship between multiple items of data included in the process data, which is determined independently of time. Specifically, the information representing the characteristic may be a basis vector of an N-dimensional vector space (N is an integer equal to or greater than 2) in which each piece of process data can be expressed by a single vector. In FIG. 1, the basis vector is a directional vector a １ , a ２ , a ３ The above-mentioned tendency of the process data may be, for example, information representing the characteristic. In this way, the process characteristic may be expressed by, for example, a value that quantitatively indexes the tendency of the process data.

[0026] Searching for multiple process characteristics of a process corresponds to searching for optimal basis vectors as the aforementioned basis vectors, for example. In this case, in a vector space defined by N basis vectors, the number (=N) and directions of the basis vectors are searched for so that the direction of a vector representing as many process data as possible is as close as possible to the direction of any one of the N basis vectors. By using the basis vectors searched in this way, the operational state of the process is expressed by a minimum-dimensional polyhedron that reasonably explains the relationship between multiple items of data included in the multiple process data. In a specific example described below, the number (=N) of basis vectors corresponds to the number N of process data categories. Furthermore, the direction of the basis vector corresponds to the operational state matrix.

[0027] As described above, process data acquired in a process having multiple process characteristics is classified into one of multiple classes based on the trend of the process data values. The estimation system classifies the multiple process data acquired in the process into multiple classes based on the process characteristics. The estimation system then generates a set of optimal estimation models for each of the multiple classes. The set of estimation models includes multiple estimation models. Each of the multiple estimation models is a model that obtains output data from input data included in the process data. The trend of the process data values ​​used to create the set of estimation models may be represented, for example, by using an operational state matrix. In this case, the operational state matrix may be used as a feature matrix of the set of estimation models (a matrix that represents the features of the set of estimation models in terms of the operational state of the process). More specifically, for example, the operational state matrix (feature matrix) may be used as a matrix that represents the operational state of the process at the time of creating the set of estimation models. The operational state of the process at the time of creating the set of estimation models is, for example, a state determined based on the process data used to create the set of estimation models.

[0028] The data item to be used as input data may be selected from multiple items of process data. In this case, the estimation system identifies process characteristics of the process data (input data) from which output data is to be estimated. Then, the estimation system estimates the output data using an estimation model corresponding to the identified process characteristics. The output data of the estimation model is, for example, data to be estimated or predicted. The output data of the estimation model may be, for example, data of an unobservable state quantity or data of an observable state quantity. The data item to be used as output data may be selected from multiple items of process data. The data item to be used as output data may be an item different from the items of the process data. In this way, the output data may be process data or may not be process data. In addition, the output data of the estimation model may be, for example, future process data. In this way, the output data of the estimation model may be data at the time of estimation or may be future data. When the output data of the estimation model includes an unobservable state quantity, a calculated value may be used as the output data used in generating the estimation model.

[0029] In this way, the estimation system generates an optimal estimation model corresponding to any one of multiple process characteristics of a process for each of the multiple process characteristics. In this way, the estimation system generates a set including multiple optimal estimation models as a set of optimal estimation models. Then, the estimation system selects an estimation model corresponding to the process characteristics of the process data to be estimated from the multiple optimal estimation models included in the set of optimal estimation models. The estimation system estimates output data using the selected estimation model. This allows the estimation system to accurately estimate data (e.g., process data) indicating the operating state of a process having multiple process characteristics.

[0030] The configuration of the estimation system 1 according to this embodiment will be described below with reference to Fig. 2. Fig. 2 is a block diagram showing an example of the configuration of the estimation system 1 according to this embodiment. As shown in Fig. 2, the estimation system 1 according to this embodiment includes a model generation device 100, an estimation device 200, and a model storage unit 300.

[0031] [1-1. Model Generation Device] The model generation device 100 generates an estimation model for estimating data (e.g., process data) indicating the operating state of a process 10. As shown in FIG. 1 , the model generation device 100 according to this embodiment includes a data classification processing unit 110, a model generation unit 120, and a search unit 130.

[0032] The data classification processing unit 110 classifies multiple pieces of process data acquired in the process 10 into multiple classes based on process characteristics. The process data acquired in the process 10 is stored in the process data storage unit 30. The data classification processing unit 110 acquires process data used to generate an estimation model from the process data storage unit 30. The data classification processing unit 110 may acquire multiple pieces of process data generated within a predetermined period, or may acquire a predetermined number of process data.

[0033] The data classification processing unit 110 classifies multiple pieces of process data acquired in the process 10 into multiple classes based on process characteristics. A method for classifying the multiple pieces of process data is, for example, clustering. The clustering method may be, for example, non-negative matrix factorization (hereinafter also referred to as "NMF"), k-means, or the like. The data classification processing unit 110 outputs the results of classifying the multiple pieces of process data into multiple classes based on the process characteristics to the model generation unit 120.

[0034] The model generation unit 120 generates estimation models for each of the multiple classes using the process data classified by the data classification processing unit 110. The estimation models may be machine learning models constructed using, for example, a regression equation, a neural network, a decision tree, or the like. The machine learning models may also be unsupervised learning models. The model generation unit 120 generates a set of estimation models. The set of estimation models includes multiple estimation models, each corresponding to a process characteristic of the process 10. In the set of estimation models, there may be a one-to-one correspondence between the estimation models and the process characteristics of the process 10. In other words, the set of estimation models includes multiple estimation models, one or more of which are generated for each of the multiple process characteristics that the process 10 can have.

[0035] The output data is data indicating the operating status of the process 10. The output data may be quantitative data (digitized data) or qualitative data (non-digitized data). When the output data is qualitative data, the estimation model may be an estimation model that outputs, as output data, qualitative data indicating, for example, any of "very large," "large," "normal," "small," and "very small." An estimation model that outputs qualitative data as output data may be a decision tree. The output data may have one or more items.

[0036] The search unit 130 evaluates the set of estimation models generated by the model generation unit 120 and searches for a set of estimation models for the process 10 based on the evaluation results. The search unit 130 evaluates the set of estimation models based on estimation errors of output data obtained by the estimation models and searches for a set of estimation models for the process 10 that can accurately estimate output data based on the evaluation results. The search unit 130 stores the set of estimation models for the process 10 in the model storage unit 300. The set of estimation models for the process 10 may be, for example, a set of estimation models that satisfies a condition to be adopted as an optimal solution in an algorithm that searches for a set of estimation models for the process 10. The condition is determined based on the evaluation results of the set of estimation models. For example, the condition may be determined using the number of evaluations of the set of estimation models. Specifically, for example, the condition may be a condition that the number of evaluations of the set of estimation models is a predetermined value. Furthermore, the condition may be determined using an evaluation value of the set of estimation models. Specifically, for example, the condition may be that the absolute value of the difference between the current evaluation value and the previous evaluation value of the set of estimation models is equal to or less than a predetermined value. Alternatively, the condition may be that the evaluation value of the set of estimation models is equal to or less than a predetermined value. A set of estimation models for the process 10 that satisfies the condition is stored in the model storage unit 300 as a set of optimal estimation models for the process 10. In this manner, it is preferable to use the set of optimal estimation models for the process 10 as the set of estimation models for the process 10. However, for example, as long as a set of estimation models that satisfies a predetermined condition based on the results of evaluating the set of estimation models is searched for as the set of estimation models for the process 10, the condition does not need to be a condition for adoption as an optimal solution in the algorithm that searches for the set of estimation models for the process 10.

[0037] [1-2. Estimation Device] The estimation device 200 estimates output data from input data included in process data to be estimated by using an estimation model. As shown in FIG. 2 , the estimation device 200 according to this embodiment includes a process data acquisition unit 210, a process characteristic identification unit 220, and an estimation unit 230.

[0038] The process data acquiring unit 210 acquires process data to be estimated. The process data to be estimated is, for example, process data acquired in the process 10. The process data acquiring unit 210 may acquire process data input from a device installed in the process 10 as the process data to be estimated. The process data acquiring unit 210 may also acquire process data input from the terminal 500 as the process data to be estimated. The process data acquiring unit 210 outputs the acquired process data to be estimated to the process characteristic identifying unit 220.

[0039] The process characteristic identifying unit 220 identifies a process characteristic of the process data to be estimated. When the process data to be estimated is input from the process data acquiring unit 210, the process characteristic identifying unit 220 identifies to which class the process data to be estimated belongs among multiple classes generated by classifying multiple pieces of process data in the model generating device 100. The process characteristic identifying unit 220 assumes that the identified class corresponds to the process characteristic of the process data to be estimated, and outputs the identified class to the estimating unit 230.

[0040] The estimation unit 230 estimates output data from input data included in the process data to be estimated using an estimation model. The estimation unit 230 acquires an estimation model corresponding to the process characteristic identified by the process characteristic identifying unit 220 from a plurality of estimation models included in a set of estimation models stored in the model storage unit 300. The estimation unit 230 according to this embodiment acquires an estimation model corresponding to a class to which the process data to be estimated belongs. The estimation unit 230 estimates output data using the estimation model acquired from the model storage unit 300. The estimation unit 230 outputs the estimated output data to, for example, the terminal 500.

[0041] The terminal 500 is, for example, an information processing device. The terminal 500 and the estimation device 200 are connected so as to be able to communicate with each other. The terminal 500 performs, for example, acquiring information from an external device, performing information processing, storing information, and outputting information to an external device. Outputting information includes, for example, displaying the information on a computer display and transmitting the information to an external device. The terminal 500 may be, for example, a personal computer, a tablet terminal, or the like. The user can check the output data estimated by the estimation device 200 via the terminal 500.

[0042] An example of the configuration of the estimation system 1 according to this embodiment has been described above. Note that FIG. 1 illustrates an example in which the model generation device 100, the estimation device 200, and the model storage unit 300 are configured by different devices. However, the present disclosure is not limited to such an example. For example, at least two of the model generation device 100, the estimation device 200, and the model storage unit 300 may be configured by a single device. Alternatively, the model generation device 100 may be configured by two or more devices. Similarly, the estimation device 200 may be configured by two or more devices.

[0043] Furthermore, a program for realizing each function of the above-described model generating device 100 and estimation device 200 may be created. The program may be a computer product or may be implemented in a computer or the like. Furthermore, a computer-readable storage medium on which such a program is stored may be provided. The storage medium may be, for example, a magnetic disk, an optical disk, a magneto-optical disk, a flash memory, or the like. Furthermore, the program does not have to be provided in a state where it is stored in a storage medium. For example, the program may be distributed via a network.

[0044] 2. Model Generation Method and Estimation Method In the estimation system 1, for example, a model generation device 100 realized by a computer executes a model generation method that generates an estimation model. Also, in the estimation system 1, for example, an estimation device 200 realized by a computer executes an estimation method that estimates the operating state of a process using the estimation model. An example of the model generation method and the estimation method according to this embodiment will be described below.

[0045] [2-1. Model Generation Method] (2-1-1. Overview) Fig. 3 is a flowchart showing an example of an overview of the model generation method according to this embodiment. As shown in Fig. 3, in the model generation method according to this embodiment, first, the data classification processing unit 110 classifies a plurality of process data acquired in a process having a plurality of process characteristics into a plurality of classes based on the process characteristics (S11: data classification processing step).

[0046] Next, the model generation unit 120 generates an estimation model for each of the classes classified in step S11 (S13: model generation step). The model generation unit 120 regards the multiple estimation models generated for each class in this way as a set of estimation models. As described above, an estimation model is a model that obtains output data from input data included in the process data.

[0047] The search unit 130 then evaluates the set of estimation models generated in step S13 and, based on the evaluation results, searches for a set of estimation models (a set of estimation models for the process) for estimating data (e.g., process data) indicating the operating state of a process having multiple process characteristics (S15: search step). The search unit 130 evaluates the set of estimation models based on estimation errors of output data obtained by each of the multiple estimation models generated in step S13 and, based on the evaluation results, searches for a set of estimation models for the process. The estimation error may be expressed as an error between output data obtained from the estimation model by inputting input data included in the process data to the estimation model and actual output data. The estimation error may be, for example, a value obtained by subtracting the actual output data from the output data obtained by the estimation model. Alternatively, the estimation error may be, for example, a value obtained by subtracting the output data obtained by the estimation model from the actual output data. Note that the actual output data used when evaluating the set of estimation models is the output data used when generating the estimation models. For example, the actual output data used when evaluating the set of estimation models is the output data included in the process data used when generating the estimation models.

[0048] In this way, in the model generation method according to the present embodiment, a set of estimation models for a process is generated by generating a plurality of estimation models corresponding to a plurality of process characteristics of the process. As described above, the set of estimation models for a process is preferably a set of optimal estimation models.

[0049] More specifically, in the model generation method according to this embodiment, a plurality of process data are classified into a plurality of classes using nonnegative matrix factorization (NMF) as a clustering technique. Then, in the model generation method according to this embodiment, a multiple regression model (hereinafter simply referred to as a "regression model") is generated as an estimation model by using a plurality of process data classified into the same class. Such regression model generation is performed for each of the plurality of classes. As a result, one regression model is generated for each of the plurality of classes. NMF is an algorithm that decomposes a nonnegative matrix into the product of two nonnegative matrices. In the model generation method according to this embodiment, a matrix storing a plurality of process data (input data) used to generate an estimation model is decomposed into a nonnegative matrix using NMF, thereby classifying the plurality of process data into a plurality of classes. Then, in the model generation method according to this embodiment, a regression model generated by using a plurality of process data classified into the same class is evaluated for each of the plurality of classes, and a set of optimal regression models is generated based on the evaluation results. An example of a model generation method using NMF will be described below.

[0050] (2-1-2. Model Generation Method Using NMF) FIG. 4 is a flowchart showing an example of a model generation method using NMF. As shown in FIG. 5, when using NMF, past process data stored in the process data storage unit 30 is represented by a matrix Y and a matrix P. The matrix Y is a matrix with T rows and M_Y columns (Y∈R （Ｔ，Ｍ＿Ｙ） ). Each row of the matrix Y represents time. Each column of the matrix Y represents an item of input data input to the estimation model. The matrix P has T rows and M_P columns (P∈R （Ｔ，Ｍ＿Ｐ）). Each row of matrix P represents time. Each column of matrix P represents an item of output data output from the estimation model. The values ​​of the input data items in each row of matrix Y and the values ​​of the output data items in each row of matrix P indicate the values ​​of process data acquired at each time. Therefore, the values ​​of the input data items in each row of matrix Y and the values ​​of the output data items in each row of matrix P correspond to each other by time. Hereinafter, matrix Y will also be referred to as "input data matrix Y." Matrix P will also be referred to as "output data matrix P."

[0051] (S101: Setting Initial Values) In the model generation method using NMF, as shown in Fig. 4, first, the data classification processing unit 110 executes initial setting of the initial value and upper limit value of the number of classifications N of the process data, the initial value of the coefficient matrix Φ, and the initial value of the operation state matrix X (S101). Hereinafter, the "number of classifications N of the process data" will also be simply referred to as the "number of classifications N".

[0052] The number of classifications N is the number of classes into which process data is classified. The initial value and upper limit of the number of classifications N may be any value. For example, the initial value of the number of classifications N may be 1 (N ０ For example, the upper limit of the number of classifications N may be 10 (N ｍａｘ =10 is also acceptable).

[0053] For example, the smaller the initial value of the number of categories N, the wider the search range for the set of optimal estimation models, but the longer the time required to search for the set of optimal estimation models.Furthermore, for example, the larger the upper limit value of the number of categories N, the wider the search range for the set of optimal estimation models, but the longer the time required to search for the set of optimal estimation models.The initial value and upper limit value of the number of categories N may be set based on the above considerations.

[0054] As shown in Fig. 6, the coefficient matrix Φ and the operational state matrix X are non-negative matrices generated by non-negative matrix factorization (NMF) of the input data matrix Y. The coefficient matrix Φ is a matrix with T rows and N columns (Φ∈R （Ｔ，Ｎ）). The rows of the coefficient matrix Φ represent time. The columns of the coefficient matrix Φ represent classifications (i.e., classes) of process data. The coefficient matrix Φ is also called a feature matrix. The operational state matrix X is a matrix with N rows and M_Y columns (X∈R （Ｎ，Ｍ＿Ｙ） ). The rows of the operational state matrix X represent the classification of the process data. The columns of the operational state matrix X represent the items of the input data input to the estimation model. The operational state matrix X is also referred to as a feature matrix. The initial values ​​of the coefficient matrix Φ and the operational state matrix X may be set randomly, for example.

[0055] The data classification processing unit 110 also sets the upper limit of the number of trials L as an initial setting. For example, the upper limit of the number of trials L may be 100 (L ｍａｘ (For example, L = 100 may be used.) For example, the larger the upper limit value of the number of trials L, the wider the search range for the set of optimal estimation models becomes, but the longer the time required to search for the set of optimal estimation models becomes. The upper limit value of the number of trials L may be set from this perspective. It is assumed that the initial value of the number of trials L is 1.

[0056] (S103: Obtaining Operational State Matrix) Next, the data classification processing unit 110 executes the NMF algorithm using the initial values ​​set in step S101 to approximate the input data matrix Y to the matrix product (Y≈Φ·X) of the coefficient matrix Φ and the operational state matrix X (S103). The coefficient matrix Φ is generally composed of sparse matrix elements as shown in FIG. 6. Here, each row of the coefficient matrix Φ is represented by a matrix element Φ. ｊ (j=1, . . . , T). The data classification processing unit 110 classifies the value of each item of the input data into a matrix element Φ ｊ The matrix element Φ is used to classify (cluster) the data into multiple classes by relating them with a scalar product. ｊ represents the contribution of the process to each process characteristic at each time. That is, the matrix element Φ ｊEach column of represents a class corresponding to a process characteristic. The operational state matrix X represents the characteristics of the time-invariant process data values, which are independent of time. The NMF algorithm itself can be realized by known techniques, so a detailed description of the NMF algorithm will be omitted here. In the flowchart shown in FIG. 4, the processes using the NMF algorithm are steps S101 and S103. The other processes are different from the processes using the NMF algorithm.

[0057] Matrix element Φ of coefficient matrix Φ calculated by NMF ｊ Each column of represents a class corresponding to a process characteristic. Therefore, the process data at time j is expressed by the matrix element Φ of the coefficient matrix Φ. ｊ It can be said that the matrix element Φ of the coefficient matrix Φ belongs to the class of the column that contains the maximum value. ｊ From this, the class to which the process data at each time belongs can be determined. Then, the process characteristics are automatically identified from the class to which the process data at each time belongs.

[0058] For example, let us assume that the number of classifications N=4 and the input data matrix Y is expressed as the matrix product of the coefficient matrix Φ and the operational state matrix X. Also, let us assume that each row of the input data matrix Y is classified into classes 1 to 4 as shown in FIG. 7. In this case, among the multiple matrix elements included in the coefficient matrix Φ, the column in which the maximum value is stored for each matrix element corresponding to class 1 is the first column. Similarly, among the matrix elements of the coefficient matrix Φ, the columns in which the maximum value is stored for each matrix element corresponding to classes 2, 3, and 4 are the second, third, and fourth columns, respectively. In this way, as a result of NMF, the matrix elements Φ of the coefficient matrix Φ are ｊ In this example, process data in which the same column stores the maximum value is data of the same process characteristic, and therefore are classified into the same class.

[0059] In addition, the matrix element Φ of the coefficient matrix Φ ｊIn the above, if all the columns have the same value, or if there are two or more columns that store maximum values, the plurality of process data may be considered to belong to the same class. In this case, for example, the data classification processing unit 110 may generate one global estimation model by using the plurality of process data, without classifying the plurality of process data into multiple classes.

[0060] In processes where the time period of operation and the operating state of the process are often interrelated, continuous process data in a time series may be classified into the same class as shown in FIG. 7 , but the present disclosure is not limited to such an example. For example, when multiple continuous process data in a time series are classified in chronological order, the classes into which the multiple process data are classified may frequently change. Furthermore, the process data at each time point in the input data matrix Y do not need to be arranged in a time series. Furthermore, the process data in the input data matrix Y and the output data matrix P may be data acquired at the same time intervals (equal intervals) or may be data acquired at different time intervals (unequal intervals). Furthermore, the process data in the input data matrix Y and the output data matrix P may be continuous data or discontinuous data.

[0061] 4, when the process characteristics of the process data are identified in step S103, the model generation unit 120 classifies the input data matrix Y and the output data matrix P into classes (S105). Then, the model generation unit 120 generates an estimation model for each class (S105).

[0062] The model generating unit 120 first classifies the input data matrix Y and the output data matrix P into classes based on the coefficient matrix Φ calculated in step S110. That is, the model generating unit 120 classifies the process data (input data) at each time of the input data matrix Y into the matrix elements Φ of the coefficient matrix Φ. ｊ When the number of classifications is N, the process data (input data) of the input data matrix Y is classified into N classes as a matrix Y １ , Y２ , ..., Y Ｎ (Y = [Y １ ; Y ２ ;…;Y Ｎ ], Y ｉ ∈R （Ｔｉ，Ｍ＿Ｙ） Similarly, the model generating unit 120 converts the process data (output data) at each time of the output data matrix P into matrix elements Φ of the coefficient matrix Φ. ｊ In this case, the model generation unit 120 classifies the model into one of a plurality of classes based on the matrix element Φ of the coefficient matrix Φ. ｊ The process data (output data) at each time of the output data matrix P may be classified into the same class as the class to which the process data (input data) at that time belongs, without referring to the matrix P. When the number of classifications is N, the process data (output data) of the output data matrix P may be classified into the N class-specific matrices P １ , P ２ , ..., P Ｎ (P = [P １ ;P ２ ;...;P Ｎ ], P ｉ ∈R （Ｔｉ，Ｍ＿Ｐ） ). In terms of the time dimension, T = ΣT ｉ is.

[0063] Then, the model generation unit 120 generates the process data (input data) Y ｉ and the process data (output data) P of the output data matrix P ｉ Among them, process data Y classified into the same class ｉ , P ｉ The estimation model is generated based on, for example, the linear regression equation P ｉ = Y ｉ ・m ｉ (i=1,2,...,N,m ｉ ∈R （Ｎ，Ｍ＿Ｐ）) The linear regression equation may be determined by, for example, the least squares method. As a result, a plurality of estimation models corresponding to the plurality of process characteristics are generated as a set of estimation models for estimating data (e.g., process data) indicating the operating state of a single process having a plurality of process characteristics. Note that the least squares method itself can be realized by a known technique, and therefore a detailed description of the least squares method will be omitted here. Also, as mentioned above, the estimation model is not limited to the linear regression equation. If the estimation model is an estimation model other than the linear regression equation (e.g., a neural network), the linear regression equation as the estimation model in the description of the flowchart of FIG. 4 may be replaced with the estimation model other than the linear regression equation (e.g., a neural network).

[0064] (S107-S121: Search for a set of estimation models) Once a set of estimation models is generated, the search unit 130 evaluates the generated set of estimation models and searches for a set of estimation models for the process based on the evaluation results. First, the search unit 130 evaluates the estimation accuracy based on the estimation error of each of the multiple estimation models included in the generated set of estimation models (S107). The estimation error of each estimation model is calculated using, for example, the norm Σ(P ｉ -Y ｉ ・m ｉ ) The search unit 130 calculates the norm Σ(P ｉ -Y ｉ ・m ｉ ) and determines whether the norm has updated to the minimum value. Note that the search unit 130 sets, for example, null (no value) as the norm value to be compared with the norm calculated in the first step S107. This allows the search unit 130 to always determine in the first step S107 that the norm has updated to the minimum value (the search unit 130 can always determine YES in S107). Note that the norm is calculated as Σ(Y ｉ ・m ｉ -P ｉ ) may be expressed as:

[0065] When the search unit 130 determines that the calculated norm has updated to the minimum value (S107: YES), it determines that the estimation accuracy of the N estimation models (the set of estimation models) generated in the current step S105 has improved. Therefore, the set of estimation models that is currently optimal is the set of estimation models generated in the current step S105. However, there is a possibility that a set of estimation models with higher estimation accuracy exists. Therefore, the search unit 130 updates the number of process data categories N, the operation state table column X, and the set of estimation models to the current number of process data categories N, the operation state matrix X, and the set of estimation models (S109). The search unit 130 also updates the norm to the current norm. Then, the search unit 130 proceeds to the processing of step S111. It is assumed that the number of process data categories N, the operation state matrix X, the set of estimation models, and the norm are always updated in the initial step S109.

[0066] Here, the set of current estimation models is the set of estimation models (N estimation models) generated in the current step S105. The number of classifications N of the current process data is the number of estimation models generated in the current step S105. The current operation state matrix X is the operation state matrix X calculated in the current step S103. The current norm is the norm calculated in the current step S107. The current time refers to the most recent time when step S109 is executed.

[0067] On the other hand, if the norm calculated in step S107 does not update the minimum value (S107: NO), the search unit 130 determines that the N estimation models (the set of estimation models) generated in the current step S105 have not improved estimation accuracy compared to the N estimation models (the set of estimation models) generated in the previous step S105. In this case, the search unit 130 proceeds to step S111 without proceeding to step S109.

[0068] When the process up to step S109 is completed, the search unit 130 determines whether the number of trials L reaches the upper limit L ｍａｘ As a result of this determination, it is determined whether the number of trials L is equal to the upper limit L ｍａｘIf not (S111: NO), the search unit 130 changes the coefficient matrix Φ and the operational state matrix X (S113). The coefficient matrix Φ and the operational state matrix X may be changed based on the current value of the norm, may be changed randomly, or may be changed by some other method (the same applies to step S119 described later). Then, the search unit 130 adds 1 to the number of trials L (S115). Thereafter, the processes of steps S103 to S109 are repeatedly executed until a determination of YES is made in step S111.

[0069] The number of trials L is the upper limit L ｍａｘ If it is determined that the number of classifications N of the process data is equal to or greater than the upper limit value N (S111: YES), the search unit 130 ｍａｘ As a result of this determination, it is determined whether the number of classifications N of the process data is equal to the upper limit N ｍａｘ If not (S117: NO), the search unit 130 changes the coefficient matrix Φ and the operational state matrix X (S119). Then, the search unit 130 resets the number of trials L to 1 and adds 1 to the number of classifications N (S121). Thereafter, the processes of steps S103 to S115 are repeatedly executed until a determination of YES is made in step S117.

[0070] The number of classifications N of the process data is the upper limit N ｍａｘ If it is determined that the optimal operation state matrix X, the number of classifications N, and the set of estimation models are the optimal operation state matrix X, the number of classifications N, and the set of estimation models (S117: YES), the search unit 130 stores the currently held operational state matrix X, the number of classifications N, and the set of estimation models in the model storage unit 300 as a set of optimal operation state matrix X, the number of classifications N, and the set of estimation models (S123). The search unit 130 then ends the search for the set of estimation models. Note that the model storage unit 300 stores, as information on the set of optimal estimation models, information necessary for performing estimation using the multiple estimation models included in the set. Furthermore, since the optimal number of classifications N corresponds to the number of estimation models included in the set of optimal estimation models, it does not need to be stored in the model storage unit 300. Furthermore, the optimal operational state matrix X is used by the estimation device 200 to identify the class of process data to be estimated.

[0071] The model generation method according to this embodiment has been described above. According to the model generation method according to this embodiment, multiple pieces of process data acquired in a process are classified into multiple classes based on process characteristics. Then, optimal estimation models for each of these multiple classes are generated as a set of optimal estimation models. This makes it possible to provide estimation models suitable for estimating output data according to the process characteristics of the process data. Therefore, it is possible to estimate output data with high accuracy. Therefore, it is possible to accurately estimate the operating state of a process in which the process data can take on various values ​​even when the operating state is not intentionally changed.

[0072] Generally, the more process data to be classified into classes and the more data items there are, the more effort is required for clustering. On the other hand, by using NMF, process data can be automatically classified into multiple classes. Therefore, the load of clustering can be further reduced.

[0073] [2-2. Estimation Method] (2-2-1. Overview) FIG. 8 is a flowchart showing an example of an overview of the estimation method according to this embodiment. As shown in FIG. 8 , in the estimation method according to this embodiment, first, the process characteristic identifying unit 220 identifies the process characteristic of the process data to be estimated (S21: process characteristic identifying step). The process data to be estimated is acquired in a process having multiple process characteristics. For example, the process characteristic identifying unit 220 may identify the process characteristic of the process data to be estimated by identifying to which class the process data to be estimated belongs, among classes corresponding to the multiple process characteristics. The multiple process characteristics are represented, for example, by multiple classes generated by clustering process data acquired in the process 10. In other words, the multiple process characteristics are, for example, information represented as a result of clustering process data previously acquired in the process 10. The process data previously acquired in the process 10 may be process data acquired in a past operation of the process 10 (an operation temporally preceding the operation from which the process data to be estimated was acquired).

[0074] Next, the estimation unit 230 estimates output data from input data of the process data to be estimated using an estimation model corresponding to the process characteristics of the process data to be estimated identified in step S21 (S23: estimation step).

[0075] As described above, in the process data estimation method according to the present embodiment, the estimation unit 230 estimates output data from input data included in the process data to be estimated by using an estimation model corresponding to the process characteristics of the process data to be estimated. Here, the estimation unit 230 may select the estimation model to be used for estimating the process data from, for example, a set of estimation models generated by the model generation method described above.

[0076] (2-2-2. Estimation Method Using Estimation Model Generated Using NMF) In the flowchart shown in Fig. 4, an estimation model is generated using NMF (see step S103). Therefore, as an example of the estimation method according to this embodiment, an example of an estimation method for estimating output data using an estimation model generated using NMF will be described with reference to Fig. 9. Fig. 9 is a flowchart showing an example of the estimation method for estimating output data using an estimation model generated using NMF.

[0077] 9 , first, the process data acquiring unit 210 acquires process data to be estimated (S200). For example, the process data acquiring unit 210 may acquire process data acquired in the process 10 from a device installed in the process 10 as the process data to be estimated. Furthermore, for example, the process data acquiring unit 210 may acquire the process data acquired in the process 10 via the terminal 500. In this case, the process data acquiring unit 210 may acquire the process data input from the terminal 500 as the process data to be estimated. For example, the process data acquiring unit 210 outputs the process data to be estimated acquired from the process 10 or the terminal 500 to the process characteristic identifying unit 220.

[0078] Next, the process characteristic identifying unit 220 identifies the process characteristic of the process data to be estimated acquired in step S200 (S210). When the process data to be estimated is input from the process data acquiring unit 210, the process characteristic identifying unit 220 identifies to which class the process data to be estimated belongs, among the classes corresponding to the plurality of process characteristics. As described above, the plurality of process characteristics are represented by a plurality of classes generated by clustering the plurality of process data acquired in the process. In this embodiment, the clustering is performed by the model generating device 100 (i.e., each class is generated by the model generating device 100).

[0079] Specifically, the process characteristic specifying unit 220 specifies the process characteristic of the process data to be estimated by approximating the matrix y of the process data to be estimated to the matrix product (y≈φ·X) of the coefficient matrix φ and the operational state matrix X, in the same way as when the input data matrix Y is approximated to the matrix product (Y≈Φ·X) of the coefficient matrix Φ and the operational state matrix X using NMF during generation of the estimation model. Here, the matrix y is a matrix with 1 row and M_Y columns (y∈R （１，Ｍ＿Ｙ） ). Each column of the matrix y stores the value of an item of input data to be input to the estimation model among the values ​​of items of process data to be estimated. In addition, the optimal operation state matrix X stored in the model storage unit 300 in step S170 of FIG. 4 is used as the operation state matrix X. The coefficient matrix φ is a matrix with 1 row and N columns (φ∈R （１，Ｎ） ). The coefficient matrix φ is the matrix element Φ that constitutes the coefficient matrix Φ calculated when generating the specified model. ｊ (j=1, . . . , T).

[0080] The coefficient matrix φ is determined based on the matrix y of the process data to be estimated acquired in step S200 and the optimal operational state matrix X stored in the model storage unit 300 in step S123 of FIG. 6 . Once the coefficient matrix φ is determined, the process characteristic identifying unit 220 can identify a class to which the process data to be estimated belongs from the column storing the maximum value among the multiple columns of the coefficient matrix φ. The process characteristic identifying unit 220 can identify the process characteristics of the process data to be estimated by identifying the class to which the process data to be estimated belongs. The process characteristic identifying unit 220 outputs the identified process characteristics of the process data to be estimated (i.e., the identified class) to the estimation unit 230.

[0081] Then, the estimation unit 230 estimates output data from input data included in the process data to be estimated using an estimation model corresponding to the process characteristic (class) identified by the process characteristic identification unit 220 (S220). A set of optimal estimation models is stored in the model storage unit 300. The estimation unit 230 acquires an estimation model corresponding to the process characteristic (class) of the process data to be estimated identified by the process characteristic identification unit 220 from among a plurality of estimation models included in the set of optimal estimation models stored in the model storage unit 300. For example, when the set of estimation models is expressed by a linear regression formula P ｉ = Y ｉ ・m ｉ (i=1,2,...,N,m ｉ ∈R （Ｎ，Ｍ＿Ｐ） ), and the estimate of the output data is a matrix p(p∈R （１，Ｍ＿Ｐ） In this case, the estimation unit 230 calculates a matrix p of estimated values ​​of output data from a matrix y of process data to be estimated using a linear regression equation of a class i to which the process data to be estimated belongs (p=y·m ｉ The estimation unit 230 outputs the estimated value of the output data to the terminal 500, for example.

[0082] The estimation method according to the present embodiment has been described above. According to the estimation method according to the present embodiment, the process characteristics of the process data to be estimated are identified, and output data is estimated using an estimation model corresponding to the identified process characteristics from among estimation models included in a set of optimal estimation models. In this way, by estimating output data using an estimation model suited to the process characteristics, the estimation accuracy of the output data can be improved. Therefore, even if the operating state is not intentionally changed, the operating state of a process in which the process data can take various values ​​can be accurately estimated.

[0083] Furthermore, in large-scale systems, there are many types of process data acquired in the process, such as equipment operating conditions, product types, and measurements taken using sensors, etc. Therefore, the operational state matrix X represented by these process data is not necessarily uniquely determined. In such large-scale systems, it is significant to obtain an optimal operational state matrix X by searching for it, as in the model generation method according to this embodiment.

[0084] [2-3. Numerical Example] The model generation method and estimation method according to this embodiment will be explained using a simple example of estimating the output value of a small steam turbine from temperature and pressure. In this example, temperature (x1) and pressure (x2) are used as input data items. The output value (p) of the small steam turbine is used as output data item. Table 1 shows the data used to generate the estimation model.

[0085]

[0086] An input data matrix Y and an output data matrix P are generated from the data shown in Table 1. In this example, the input data matrix Y is a matrix with T rows and M_Y columns (Y∈R （Ｔ，Ｍ＿Ｙ） ). Each row of the input data matrix Y represents a unique number (T=5) corresponding to the operation of a small steam turbine. Each column of the input data matrix Y represents a temperature (x1) and a pressure (x2) (M_Y=2). The output data matrix P is a matrix with T rows and M_P columns (P∈R （Ｔ，Ｍ＿Ｐ） ). Each row of the output data matrix P represents a unique number (T=5) corresponding to the operation of a small steam turbine, and each column of the output data matrix P represents an output value (p) (M_P=1).

[0087] For example, assuming the number of classifications is 2 (N=2), the input data matrix Y is approximated by the matrix product (Y≈Φ·X) of the coefficient matrix Φ and the operational state matrix X using NMF. Then, the coefficient matrix Φ and the operational state matrix X are expressed as shown in FIG. 10 . From the coefficient matrix Φ, the five data are classified into two classes: No. 1, No. 2, and No. 5, which have the largest values ​​in the first column of the coefficient matrix Φ, and No. 3 and No. 4, which have the largest values ​​in the second column of the coefficient matrix Φ. Thus, the input data matrix Y is decomposed into a data matrix Y1 of No. 1, No. 2, and No. 5 classified into class 1, and a data matrix Y2 of No. 3 and No. 4 classified into class 2.

[0088] Each row of the operational status matrix X, which is a feature matrix, indicates the directional vector of each class as a data characteristic. That is, as shown in Fig. 11, data belonging to class 1 takes values ​​on a line L1 having a slope determined by a directional vector (2, 1). Data belonging to class 2 takes values ​​on a line L2 having a slope determined by a directional vector (3, 1).

[0089] The data in each row of the output data matrix P is classified into the same class as the class into which the data in that row of the input data matrix Y is classified. That is, as shown in FIG. 12 , the output data matrix P is decomposed into a data matrix P1 storing output data Nos. 1, 2, and 5 classified into class 1, and a data matrix P2 storing output data Nos. 3 and 4 classified into class 2. Then, for each of classes 1 and 2, an estimated model representing the relationship between the input data y and the output data p is generated. Here, a case where the estimated model is a linear regression equation is illustrated. In this case, a regression equation (p = 0.3 × x1 + 0.2 × x2) representing the relationship between the data matrix Y1 and the data matrix P1 is generated as the estimated model corresponding to class 1. Furthermore, a regression equation (p = 0.4 × x1 + 0.5 × x2) representing the relationship between the data matrix Y2 and the data matrix P2 is generated as the estimated model corresponding to class 2.

[0090] Using the estimation model generated as described above, it is assumed that the output value of a small steam turbine is estimated when the small steam turbine is operated under the operating conditions of a temperature of 180°C and a pressure of 90 kPa. As shown in FIG. 13 , the matrix y of the process data to be estimated is expressed as a 1-row, 2-column matrix where x1 = 180 and x2 = 90. The matrix y is approximated by the matrix product (y ≈ φ · X) of the coefficient matrix φ and the operational state matrix X, which is a feature matrix. Then, as shown in FIG. 13 , a 1-row, 2-column matrix where φ1 = 90 and φ2 = 0 is calculated as the coefficient matrix φ. Since the value in the first column of the coefficient matrix φ is the largest, the process data to be estimated belongs to class 1. Therefore, the output value of the small steam turbine is estimated using the estimation model for class 1 (p = 0.3 × x1 + 0.2 × x2). As a result, the output value of the small steam turbine is estimated to be 72 kW (p = 72).

[0091] 3. Hardware Configuration An example of the hardware configuration of the model generation device 100 and the estimation device 200 according to this embodiment will be described with reference to Fig. 14. Fig. 14 is a block diagram showing an example of the hardware configuration of an information processing device 900 that functions as the model generation device 100 or the estimation device 200 according to this embodiment.

[0092] The information processing device 900 includes one or more hardware processors, such as a CPU (Central Processing Unit) 901, and one or more memories, such as a RAM (Random Access Memory) 905 and a ROM (Read Only Memory) 903. The information processing device 900 executes various operations by executing one or more programs stored in the memories by the one or more hardware processors. The information processing device 900 also includes a bus 907, an input I / F 909, an output I / F 911, a storage device 913, a drive 915, a connection port 917, and a communication device 919.

[0093] For example, the CPU 901 functions as an arithmetic processing unit and a control unit. The CPU 901 controls all or part of the operations within the information processing device 900 in accordance with various programs stored in the ROM 903, the RAM 905, the storage device 913, or the removable storage medium 925. The ROM 903 stores programs used by the CPU 901, calculation parameters, etc. The RAM 905 temporarily stores programs used by the CPU 901, or parameters that the CPU 901 changes as appropriate when executing a program. These are interconnected by a bus 907 consisting of an internal bus such as a CPU bus. The bus 907 is connected to an external bus such as a PCI (Peripheral Component Interconnect / Interface) bus or PCI Express (registered trademark) via a bridge.

[0094] In addition, the arithmetic processing unit and control unit may be realized by a PLC (Programmable Logic Controller) other than the CPU 901, or may be realized by dedicated hardware such as an ASIC (Application Specific Integrated Circuit) and an FPGA (Field Programmable Gate Array).

[0095] The input I / F 909 is an interface that accepts input from an input device 921, which is an operating means operated by a user. Examples of the operating means include a mouse, keyboard, touch panel, button, switch, and lever. The input I / F 909 is configured, for example, as an input control circuit that generates an input signal based on information input by a user to the information processing device 900 using the input device 921 and outputs the signal to the CPU 901. The input device 921 may be, for example, a remote control device that uses infrared or other radio waves. Alternatively, an external device 927, such as a PDA (Personal Digital Assistant) that supports operation of the information processing device 900, may serve as the input device 921. A user of the information processing device 900 can operate the input device 921 to input various data to the information processing device 900 and instruct processing operations.

[0096] The output I / F 911 is an interface that outputs information input to the information processing device 900 to an output device 923 that can visually or audibly notify the user. The output device 923 may be, for example, a display device such as a CRT (Cathode Ray Tube) display device, a liquid crystal display device, a plasma display device, an EL (Electronic Luminescence) display device, or a lamp. Alternatively, the output device 923 may be an audio output device such as a speaker or headphones, a printer, a mobile communication terminal, or a facsimile machine. The output I / F 911 instructs the output device 923 to output, for example, processing results obtained in various processes executed by the information processing device 900. Specifically, the output I / F 911 instructs the display device to display the processing results of the information processing device 900 as text or images. The output I / F 911 also instructs the audio output device to convert audio signals, such as audio data, that have been instructed to be played back into analog signals and output them.

[0097] The storage device 913 is one of the storage units of the information processing device 900 and is a device for storing data. The storage device 913 is configured, for example, by a magnetic storage device such as a hard disk drive (HDD), a semiconductor storage device such as a solid state drive (SSD), an optical storage device, or a magneto-optical storage device. The storage device 913 stores programs executed by the CPU 901, various data generated by the execution of the programs, various data obtained from outside, etc.

[0098] The drive 915 is a reader / writer for a storage medium. The drive 915 is built into or externally attached to the information processing device 900. The drive 915 reads information stored in a removable storage medium 925 attached to the information processing device 900 and outputs the information to the RAM 905. The drive 915 can also write information to the removable storage medium 925 attached to the information processing device 900. The removable storage medium 925 is, for example, a magnetic disk, an optical disk, a magneto-optical disk, or a semiconductor memory. Specifically, the removable storage medium 925 may be a CD medium (Compact Disc medium), a DVD medium (Digital Versatile Disc medium), a Blu-ray (registered trademark) medium, a CompactFlash (registered trademark), a flash memory, an SD memory card (Secure Digital memory card), or the like. The removable storage medium 925 may also be, for example, an IC card (Integrated Circuit card) equipped with a contactless IC chip, an electronic device, or the like.

[0099] The connection port 917 is a port for directly connecting a device to the information processing device 900. The connection port 917 is, for example, a Universal Serial Bus (USB) port, an external Serial Advanced Technology Attachment (eSATA), or a Serial Attached Small Computer System Interface (SAS) port. The information processing device 900 can directly acquire various data from an external device 927 connected to the connection port 917 or provide various data to the external device 927. For example, an alarm notification device such as a rotating light for notifying alarm information may be connected to the connection port 917 as the external device 927. Furthermore, a network-attached storage (NAS) may be connected to the connection port 917 as the external device 927. In this case, the NAS may be used as a storage device.

[0100] The communication device 919 is, for example, a communication interface configured with a communication device for connecting to a communication network 929. The communication device 919 is, for example, a communication card for a wired or wireless local area network (LAN), a communication card for Bluetooth (registered trademark), or a communication card for wireless USB (WUSB). The communication device 919 may also be a router for optical communication, a router for asymmetric digital subscriber line (ADSL), or a modem for various communications. The communication device 919 can transmit and receive signals, for example, between the Internet and other communication devices in accordance with a predetermined protocol such as TCP / IP. For example, a computer for operating the information processing device 900 can also be connected to the information processing device 900 via the communication device 919. The communication network 929 connected to the communication device 919 is configured with a network connected by wire or wirelessly. For example, the communication network 929 is the Internet, a home LAN, infrared communication, radio wave communication, satellite communication, or the like.

[0101] The above describes an example of the hardware configuration of the information processing device 900. Each of the aforementioned components may be configured using general-purpose components, or may be configured using hardware specialized for the function of each component. The hardware configuration of the information processing device 900 can be changed as appropriate depending on the technical level at the time of implementing this embodiment.

[0102] Although the preferred embodiments of the present disclosure have been described in detail above, the present disclosure is not limited to such examples. It is clear that a person skilled in the art to which the present disclosure pertains can conceive of various modified or altered examples within the scope of the technical ideas described in the claims, and it is understood that these also naturally fall within the technical scope of the present disclosure.

[0103] The present disclosure can be used, for example, to estimate the operating state of a process.

Claims

1. An estimation device comprising: a process characteristic identification unit that identifies process characteristics of process data to be estimated, the process characteristics being acquired in a process having a plurality of process characteristics; and an estimation unit that estimates output data from input data of the process data to be estimated using an estimation model corresponding to the identified process characteristic of the process data to be estimated, from a set of estimation models that obtain output data from input data selected from process data generated for each of the process characteristics possessed by the process.

2. The estimation device according to claim 1, wherein the process characteristic identification unit identifies the process characteristic of the process data to be estimated by identifying to which class the process data to be estimated belongs, out of classes corresponding to the process characteristics obtained by clustering process data previously acquired in the process.

3. The estimation device according to claim 1 or 2, wherein the process characteristic identification unit factorizes a matrix representing the acquired process data of the estimation target with a feature matrix representing the operating state of the process of the set of estimation models to obtain a coefficient matrix, and identifies the process characteristics of the process data of the estimation target based on the coefficient matrix.

4. A model generation device comprising: a data classification processing unit that classifies multiple process data acquired in a process having multiple process characteristics into multiple classes based on the process characteristics; a model generation unit that generates, for each of the multiple classes, an estimation model that obtains output data from input data selected from the process data; and a search unit that evaluates a set of the estimation models based on an estimation error of each of the generated estimation models, and searches for the set of estimation models for the process based on the evaluation result.

5. The model generating device according to claim 4, wherein the data classification processing unit factorizes a matrix representing the plurality of process data into two non-negative matrices, a feature matrix representing the operating state of the process and a coefficient matrix representing the process characteristics, based on the number of classifications to be classified, and classifies the plurality of process data into the plurality of classes based on the coefficient matrix.

6. The model generation device according to claim 4 or 5, wherein the search unit calculates an evaluation value for evaluating the set of estimation models, and the evaluation value is the sum of estimation errors of the output data obtained by the estimation models for the plurality of classes.

7. The model generation device according to any one of claims 4 to 6, wherein the search unit calculates an evaluation value for evaluating the set of estimation models; while the minimum value of the evaluation value is being updated, the data classification processing unit reclassifies the process data; the model generation unit regenerates the estimation models for each of the reclassified classes; the search unit recalculates an evaluation value for evaluating the regenerated set of estimation models; and when the minimum value of the evaluation value is no longer being updated, the search unit sets the set of estimation models for which the evaluation value is calculated as the set of estimation models.

8. A model generating device according to any one of claims 4 to 7, wherein the model generating unit generates, as the estimation model, a regression equation that represents the relationship between input data and output data included in the process data.

9. An estimation method in which a computer executes processing including: a process characteristic identification step of identifying process characteristics of process data to be estimated obtained in a process having a plurality of process characteristics; and an estimation step of estimating output data from input data of the process data to be estimated using an estimation model corresponding to the identified process characteristic of the process data to be estimated, from a set of estimation models that obtain output data from input data selected from process data generated for each of the process characteristics possessed by the process.

10. A model generation method in which a computer executes processing including: a data classification processing step of classifying multiple pieces of process data acquired in a process having multiple process characteristics into multiple classes based on the process characteristics; a model generation step of generating, for each of the multiple classes, estimation models that obtain output data from input data selected from the process data; and a search step of evaluating a set of estimation models based on the estimation error of each of the generated estimation models and searching for a set of estimation models for the process based on the evaluation result.

11. A program causing a computer to function as an estimation device, comprising: a process characteristic identification unit that identifies process characteristics of process data to be estimated obtained in a process having a plurality of process characteristics; and an estimation unit that estimates output data from input data of the process data to be estimated using an estimation model corresponding to the identified process characteristic of the process data to be estimated, from a set of estimation models that obtain output data from input data selected from process data generated for each of the process characteristics possessed by the process.

12. A program that causes a computer to function as a model generation device, comprising: a data classification processing unit that classifies multiple process data acquired in a process having multiple process characteristics into multiple classes based on the process characteristics; a model generation unit that generates, for each of the multiple classes, an estimation model that obtains output data from input data selected from the process data; and a search unit that evaluates a set of the estimation models based on the estimation error of each of the generated estimation models, and searches for the set of estimation models for the process based on the evaluation result.

Images