JP2023063162A - Prediction model forming method and operating process condition setting method - Google Patents

Abstract

To provide a prediction model forming method for forming a prediction model that can be fed back to an explanatory variable of an operating process.SOLUTION: In a method for forming a prediction model that predicts a predetermined objective variable related to an operating process, the prediction model forming method is provided which comprises: a visualization step of visualizing a relationship between a plurality of explanatory variables for objective variables using a correlation coefficient between the plurality of explanatory variables; a division step of dividing the plurality of explanatory variables into a plurality of clusters using the relationship between the explanatory variables visualized in the visualization step; a selection step of selecting, for each cluster, a representative explanatory variable, which is the explanatory variable representing a cluster, from among the explanatory variables included in the cluster in one or more of the clusters selected from the plurality of clusters; and a prediction model forming step of forming a prediction model that predicts the objective variable using the representative explanatory variable selected in the selection step.SELECTED DRAWING: Figure 1

Description

TECHNICAL FIELD The present invention relates to a method for creating a prediction model and a method for setting operating process conditions.

Considering changes in product manufacturing conditions and manufacturing environment, highly accurate prediction of product quality is becoming more important year by year, coupled with the introduction of IoT in measuring equipment.

For example, Patent Literature 1 describes a quality prediction device that creates a quality prediction model using data of a group of products manufactured in a manufacturing process in the past.

Further, Patent Document 2 discloses a learning method of a blast furnace unloading speed prediction model for predicting the blast furnace unloading speed, which is an important parameter in manufacturing.

JP 2019-074969 A Japanese Patent Application Laid-Open No. 2020-020003

By the way, in the operation process for manufacturing industrial products, data on raw materials, data on the operation process, evaluation values of intermediates, and the like are usually collected as data. When creating a predictive model with final characteristics and important characteristics as objective variables, the predictive model is constructed using the above data as parameters, ie, explanatory variables.

From the viewpoint of manufacturing products of the desired quality with good yield, it is required to be able to feed back from the created prediction model and calculate and set the above-mentioned parameters in the operation process.

However, no method has been proposed for creating a predictive model that can be fed back to calculate parameters in the operational process.

Therefore, in view of the above-described problems of the prior art, it is an object of one aspect of the present invention to provide a prediction model creation method for creating a prediction model that can be fed back to the explanatory variables of the operational process.

In order to solve the above problems, according to one aspect of the present invention,
A method for creating a prediction model that predicts a predetermined objective variable related to an operating process,
a visualization step of visualizing the relationship between a plurality of explanatory variables for the objective variable using a correlation coefficient between the plurality of explanatory variables;
A division step of dividing the plurality of explanatory variables into a plurality of clusters using the relationships between the explanatory variables visualized in the visualization step;
In one or more of the clusters selected from the plurality of clusters, selecting a representative explanatory variable, which is the explanatory variable representing the cluster, from among the explanatory variables included in the cluster for each cluster. process and
and a prediction model creation step of creating a prediction model for predicting the objective variable using the representative explanatory variables selected in the selection step.

ADVANTAGE OF THE INVENTION According to 1 aspect of this invention, the preparation method of the prediction model which produces the prediction model which can be fed back to the explanatory variable of an operating process can be provided.

4 is a flow of a method for creating a predictive model according to one aspect of the present disclosure; 2 is a hardware configuration diagram of a simulation device according to one aspect of the present disclosure; FIG. 1 is a block diagram showing functions of a simulation device according to one aspect of the present disclosure; FIG. 4 is a dendrogram created in the visualization process of Example 1. FIG. 4 is a distribution diagram of measured values and predicted values obtained by the prediction model obtained in Example 1 using the same explanatory variables; FIG.

Modes for carrying out the present invention will be described below, but the present invention is not limited to the following embodiments, and various modifications and variations can be made to the following embodiments without departing from the scope of the present invention. Substitutions can be added.
[Prediction model creation method]
A method for creating a prediction model according to this embodiment will be described.

The inventors of the present invention have studied a prediction model creation method that creates a prediction model that can be fed back to the explanatory variables of the operational process.

In recent years, the operating process has become a multi-step process, and the number of operating process parameters has increased. In such a multi-step operation process, parameters of raw materials, various parameters of the operation process, evaluation values of intermediates, and the like can usually be explanatory variables of the prediction model. However, in the case of multiple processes, there are inevitably many types of parameters, and when a prediction model is created using the parameters as explanatory variables, it is very difficult to obtain appropriate values for each explanatory variable by providing feedback. Become.

For example, it is conceivable to select only parameters related to objective variables, which are final characteristics and important characteristics, as explanatory variables to create a prediction model. However, it is rare that raw material parameters and intermediate evaluation values are uncorrelated. In addition, the parameters of the operating process often change in conjunction with each other, and are quantities that move in correlation with each other. Therefore, it has been difficult to select explanatory variables to be used when creating a prediction model.

Therefore, the inventors of the present invention conducted further studies and completed the method for creating a prediction model of the present invention.

The method for creating a prediction model of the present embodiment is a method for creating a prediction model for predicting a predetermined target variable related to the operation process, and includes the following visualization step S11, division step, like the flow 10 shown in FIG. S12, a selection step S13, and a prediction model creation step S14 can be provided.

The type of objective variable of the prediction model created in the prediction model creation process of the present embodiment is not particularly limited, and examples thereof include the final characteristics required for the product to be manufactured, particularly important characteristics, and the like.

The visualization step can visualize the relationship between a plurality of explanatory variables for the objective variable using correlation coefficients between the plurality of explanatory variables.

The dividing step can divide the multiple explanatory variables into multiple clusters using the relationships between the explanatory variables visualized in the visualization step.

In one or more clusters selected from a plurality of clusters, the selecting step can select, for each cluster, a representative explanatory variable, which is an explanatory variable representing the cluster, from explanatory variables included in the cluster.

The prediction model creation step can create a prediction model for predicting the objective variable using the representative explanatory variables selected in the selection step.
(1) About each process Hereinafter, each process is demonstrated.
(1-1) Visualization step (S11)
In the visualization step, the relationship between a plurality of explanatory variables for the objective variable can be visualized using correlation coefficients between the plurality of explanatory variables.

Although principal component analysis and the like are generally used as a technique for reducing redundancy between explanatory variables, it is preferable to treat the original explanatory variables as they are in order to create a predictive model that can be fed back.

Therefore, in the prediction model creation method of the present embodiment, the relationship between the correlation coefficients can be illustrated, that is, visualized, based on the correlation coefficients between explanatory variables.

In the visualization process, explanatory variables during the operation process can be enumerated for the target variable of the prediction model to be created first. Then, correlation coefficients between all enumerated explanatory variables can be obtained. At this time, a correlation coefficient between the objective variable and the explanatory variable can also be obtained together.

For visualization, the correlation coefficients can be used as they are, but it is preferable to define the distance based on the correlation coefficients between explanatory variables and perform hierarchical clustering. 1-|r| or the like can be used as the distance. In addition, r means the correlation coefficient between each explanatory variable.

The diagram to be created is not limited, and it is preferable to create a diagram that shows the relationship between the explanatory variables, the correlation coefficients between the explanatory variables, and the distance. For example, a dendrogram (tree diagram) can be created.
(1-2) Division step (S12)
In the dividing step, the relationships between the explanatory variables visualized in the visualization step can be used to divide the plurality of explanatory variables into a plurality of clusters, ie, groups.

At this time, the number of clusters to be divided and conditions for division are not particularly limited. Also, it may be divided into a plurality of clusters while confirming the physical meaning.

The number of explanatory variables included in one cluster is not particularly limited, and each cluster can include any number of explanatory variables. Also, the number of explanatory variables included may differ between clusters.
(1-3) Selection step (S13)
In the selection step, in one or more clusters selected from a plurality of clusters, a representative explanatory variable, which is an explanatory variable representing the cluster, can be selected from the explanatory variables included in each cluster.

As mentioned above, in recent years, the operating process has become multi-step, so there is a risk that the number of explanatory variables will be extremely large when creating a predictive model for the operating process. For this reason, it was sometimes difficult to feed back the predictive model and calculate explanatory variables for obtaining desired objective variables.

Therefore, in the prediction model creation method of the present embodiment, in the visualization step and the division step, the explanatory variables with high relevance are classified into clusters, and in the selected one or more clusters, representative explanations from each cluster A variable can be selected as a representative explanatory variable to create a predictive model. That is, it is possible to obtain a prediction model in which fluctuations of similar explanatory variables are represented by representative explanatory variables.

In this way, by clustering explanatory variables with high relevance and selecting a representative explanatory variable that is a representative value from the cluster and narrowing down the number of explanatory variables, it is easy to obtain explanatory variables by providing feedback from the objective variable. I will be able to do it. In addition, since explanatory variables with high relevance are clustered and representative explanatory variables are selected from the clusters, explanatory variables can be appropriately selected.
(1-3-1) Conditions for Selecting Representative Explanatory Variables In the selection process, the criteria for selecting representative explanatory variables are not particularly limited. In the selection step, a predetermined number of representative explanatory variables can be selected according to predetermined conditions.

As the conditions, for example, one or more selected from conditions 1 to 4 below is preferable.
(Condition 1) Select an explanatory variable with a high correlation coefficient with the objective variable.
(Condition 2) Select explanatory variables related to upstream processes in the operating process.
(Condition 3) Select explanatory variables that are easy to control in the operating process.
(Condition 4) When a prediction model for the objective variable is created using the explanatory variables included in the cluster, an explanatory variable having a high influence on the objective variable is selected.

Note that conditions 1 to 4 can be used in combination. Also, the conditions can be changed, for example, by cluster.
(Regarding Condition 1)
Selecting an explanatory variable with a high correlation coefficient with the objective variable means obtaining the correlation coefficient between the explanatory variables that make up each cluster and the objective variable, and determining the explanatory variables in descending order of the correlation coefficient. It means to select according to the number of explanatory variables to be selected from. An explanatory variable having a high correlation coefficient with the objective variable is a variable that greatly affects the objective variable, and therefore, by selecting the explanatory variable under the above conditions, a particularly appropriate prediction model can be obtained.
(Regarding Condition 2)
Selecting an explanatory variable related to the upstream process in the operating process means that the explanatory variables that make up the cluster are arranged in order from the upstream side of the related operating process, and the number of explanatory variables selected from the cluster is selected according to the number of explanatory variables. It means to select in order from the upstream side. Normally, when there is a correlation between variables, it is more appropriate to control the upstream side of the operating process in many cases, so by selecting under the above conditions, a more appropriate representative explanatory variable can be selected.
(Regarding condition 3)
Selecting an explanatory variable that is easy to control in the operating process means that, for example, for an explanatory variable, the ease of control is quantified in advance, and depending on the number of explanatory variables selected from the cluster, the explanatory variable is selected based on the numerical value. means to select When feedback is provided from the prediction model and explanatory variables for obtaining the desired target are obtained, if the explanatory variables are parameters that are easy to control, the operating process can be easily adjusted, especially yield can be increased. It is from.
(Regarding condition 4)
When creating a predictive model for the target variable using the explanatory variables included in the cluster, when selecting explanatory variables that have a high impact on the target variable, first use the explanatory variables included in the cluster for which the representative explanatory variable is selected. to create a predictive model for the target variable. Then, an explanatory variable having a high degree of influence on the objective variable can be selected in the prediction model. Specifically, for example, when the explanatory variables have the same fluctuation range, the explanatory variables with the highest degree of influence can be set in descending order of the fluctuation range of the objective variable.
(1-3-2) Number of representative explanatory variables selected from each cluster In the selection process, the number of representative explanatory variables selected from each cluster is not particularly limited. It may be one or plural. Also, the number of representative explanatory variables selected for each cluster may be different. For example, no representative explanatory variable may be selected from clusters considered to be of low importance. Although it is possible to select three or more representative explanatory variables from each cluster, it is preferable that the number of representative explanatory variables selected from each cluster is two or less. more preferably one.

It is possible to select the representative explanatory variable in all clusters, but as described above, depending on the cluster, it is possible not to select the representative explanatory variable. You can select explanatory variables. However, since it is preferable to have a plurality of representative explanatory variables when creating a prediction model, it is more preferable to select representative explanatory variables in two or more clusters.
(1-4) Prediction model creation step (S14)
In the prediction model creation step, a prediction model for predicting the objective variable can be created using the representative explanatory variables selected in the selection step.

Prediction models can be created using linear regression, logistic regression, random forest, boosting, neural networks, and the like.
(2) Operation process The operation process to which the prediction model creation method of the present embodiment is applied is not particularly limited, and can be applied to any operation process.

In the prediction model obtained by the prediction model creation method of the present embodiment described above, the representative explanatory variables used in the prediction model are important feature quantities representing each cluster, and the number of them is also narrowed down. there is Therefore, it is easy to feed back the feature amount based on the prediction model.

That is, according to the method of creating a prediction model of the present embodiment, it is possible to create a prediction model that can be fed back to the explanatory variables of the operational process.
[Method of setting operating process conditions]
The method of setting conditions for the operating process of the present embodiment can set a plurality of explanatory variables using, for example, the prediction model obtained by the above-described method of creating a prediction model.

Specifically, the prediction model described above has an objective variable and a representative explanatory variable as a parameter for predicting the objective variable. For this reason, it is possible to determine a target objective variable, back-calculate the prediction model, and obtain each representative explanatory variable.

Then, the conditions of the operating process can be set by the obtained representative explanatory variables.
[Simulation device]
The simulation device of this embodiment is a simulation device for creating a prediction model. For this reason, it can be called a prediction model creation device, and can have the following visualization unit, division unit, selection unit, and prediction model creation unit.

In addition, according to the simulation apparatus of this embodiment, the method of creating the prediction model described above can be carried out. For this reason, the explanation of the already explained matters will be partly omitted.

The visualization unit can visualize the relationship between a plurality of explanatory variables for the objective variable using correlation coefficients between the plurality of explanatory variables.

The dividing unit can divide the multiple explanatory variables into multiple clusters using the relationships between the explanatory variables visualized by the visualization unit.

The selection unit can select, for each cluster, a representative explanatory variable, which is an explanatory variable representing the cluster, from explanatory variables included in each cluster in one or more clusters selected from a plurality of clusters.

The prediction model creation unit can create a prediction model for predicting the objective variable using the representative explanatory variables selected by the selection unit.

As shown in the hardware configuration diagram shown in FIG. 2, the simulation device 20 of the present embodiment is configured by, for example, an information processing device (computer), and physically includes a CPU (Central Processing Unit) which is an arithmetic processing unit. : processor) 21, RAM (Random Access Memory) 22 and ROM (Read Only Memory) 23 as main storage devices, auxiliary storage device 24, input/output interface 25, display device 26 as output device, etc. It can be configured as a computer system. These are interconnected by a bus 27 . Note that the auxiliary storage device 24 and the display device 26 may be provided outside.

The CPU 21 controls the overall operation of the simulation device 20 and performs various types of information processing. The CPU 21 can execute a program (simulation program), which will be described later, for example, stored in the ROM 23 or the auxiliary storage device 24 to calculate the reflectance and the like.

The RAM 22 is used as a work area for the CPU 21 and may include a non-volatile RAM that stores main parameters and information.

The ROM 23 can store programs (simulation programs) and the like.

The auxiliary storage device 24 is a storage device such as an SSD (Solid State Drive) or an HDD (Hard Disk Drive), and can store various data, files, etc. necessary for the operation of the simulation device.

The input/output interface 25 includes both a user interface such as a touch panel, a keyboard, a display screen, and operation buttons, and a communication interface that takes in information from an external data recording server or the like and outputs analysis information to other electronic devices. .

The display device 26 is a monitor display or the like. An analysis screen is displayed on the display device 26 , and the screen is updated according to input/output operations via the input/output interface 25 .

Each function of the simulation device 20 shown in FIG. It can be realized by reading and writing and operating the input/output interface 25 and the display device 26 .

FIG. 3 shows a functional block diagram of the simulation device 20 of this embodiment.

As shown in FIG. 3 , the simulation device 20 can have a reception section 31 , a processing device 32 and an output section 33 . In an information processing device such as a personal computer provided with a CPU, a storage device, various interfaces, etc., of the simulation device 20, each of these units is implemented by executing a program stored in advance in the CPU, for example, software and hardware. are realized in collaboration.

The configuration of each part will be described below.
(1) Reception Unit The reception unit 31 receives commands and data input from the user related to the processing executed by the processing device 32 . The reception unit 31 includes a keyboard and a mouse for inputting commands and the like operated by the user, a communication device for inputting via a network, a reading device for inputting from various storage media such as a CD-ROM and a DVD-ROM, and the like. mentioned.

The receiving unit 31 can receive, for example, an objective variable for an operation process, a list of explanatory variables, data of each explanatory variable for obtaining a correlation coefficient, and the like.
(2) Processing Device The processing device 32 can have a visualization section 321 , a division section 322 , a selection section 323 and a prediction model creation section 324 .
(2-1) Visualization Unit The visualization unit 321 can visualize the relationship between multiple explanatory variables for the objective variable using correlation coefficients between multiple explanatory variables. Specifically, a correlation coefficient between the input explanatory variables is obtained, and a diagram, such as a dendrogram, relating to the explanatory variables can be created and visualized using the correlation coefficient and the above-described distance.
(2-2) Division Unit The division unit 322 can divide a plurality of explanatory variables into a plurality of clusters using the relationships between the explanatory variables visualized by the visualization unit. As described above, the explanatory variables can be divided based on the above correlation coefficients and predetermined thresholds for distances. Also, it may be divided into a plurality of clusters while confirming the physical meaning.
(2-3) Selection unit The selection unit 323 selects one or more clusters selected from a plurality of clusters, and for each cluster, out of the explanatory variables included in the cluster, the representative explanation, which is the explanatory variable that represents the cluster. Variables can be selected. Since the conditions for selecting representative explanatory variables and the like have already been explained, the explanation is omitted here.
(2-4) Prediction Model Creation Unit The prediction model creation unit 324 can create a prediction model for predicting the objective variable using the representative explanatory variables selected by the selection unit.

A method for creating a prediction model is not particularly limited, and can be created using, for example, linear regression, logistic regression, random forest, boosting, neural network, and the like.
(3) Output Unit The output unit 33 can have a display or the like. A prediction model obtained by the prediction model creation unit 324 can be output to the output unit 33 .

According to the simulation apparatus of this embodiment described above, it is possible to create a predictive model that can be fed back to the explanatory variables of the operating process.
[program]
Next, the program of this embodiment will be described.

The program of the present embodiment relates to a program for creating a prediction model, and can cause a computer to function as the following visualization unit, division unit, selection unit, and prediction model creation unit.

The visualization unit can visualize the relationship between a plurality of explanatory variables for the objective variable using correlation coefficients between the plurality of explanatory variables.

The dividing unit can divide the multiple explanatory variables into multiple clusters using the relationships between the explanatory variables visualized by the visualization unit.

The selection unit can select, for each cluster, a representative explanatory variable, which is an explanatory variable representing the cluster, from explanatory variables included in each cluster in one or more clusters selected from a plurality of clusters.

The prediction model creation unit can create a prediction model for predicting the objective variable using the representative explanatory variables selected by the selection unit.

The program of the present embodiment can be stored, for example, in various storage media such as the RAM, ROM, or other main storage device or auxiliary storage device of the aforementioned simulation device. By reading the program and executing it by the CPU, data can be read from and written to the RAM or the like, and the input/output interface and the display device can be operated. For this reason, the description of the items that have already been described in the simulation apparatus will be omitted.

According to the program of the present embodiment described above, it is possible to create a predictive model that can be fed back to the explanatory variables of the operating process.

Hereinafter, the present invention will be described more specifically with reference to examples. However, the present invention is not limited to the following examples.
[Example 1]
A prediction model for predicting objective variables related to the operational process was created by the following procedure.
(Visualization process)
In the visualization step, the correlation coefficients between the explanatory variables and between the explanatory variables and the objective variables are obtained with respect to the explanatory variables and the objective variables shown as feature amounts on the vertical axis in FIG. 4, and the dendrogram shown in FIG. It was created.

When creating the dendrogram, we used the distance defined based on the correlation coefficient. The distance is defined as 1-|r|. In addition, r means the correlation coefficient between each explanatory variable.
(Dividing process)
Based on the dendrogram obtained in the visualization process, explanatory variables were divided into multiple clusters.

Specifically, the distance was divided into 10 clusters, cluster 41 to cluster 50, using 0.65 as a standard threshold.
(Selection process)
In the selection step, for each cluster, representative explanatory variables B1 to B8, which are explanatory variables representing the cluster, were selected from the explanatory variables included in the cluster. When making a selection, we chose explanatory variables that are easy to control in the operating process. No representative explanatory variables were selected from clusters 46, 48, and 50.
(Prediction model creation process)
In the prediction model creation step, the representative explanatory variables B1 to B8 selected in the selection step were used to create a prediction model for the objective variable.

FIG. 5 shows the measured values of the fitting data, which is the data used when creating the prediction model, and the prediction values calculated using the prediction model.

FIG. 5 also shows measured values of test data, which are data not used in creating the prediction model, and predicted values calculated using the prediction model.

According to the results shown in FIG. 5, the test data and the fitting data show almost the same distribution, and it can be confirmed that an appropriate prediction model has been created.

Also, in the prediction model of this embodiment, the 28 feature quantities shown in FIG. 4 are narrowed down to 8 representative explanatory variables for use. That is, since the number of parameters required for the prediction model is small, it can be easily fed back by calculating backward from the prediction model.

S11 visualization step S12 division step S13 selection step S14 prediction model creation step

Claims (3)

A method for creating a prediction model that predicts a predetermined objective variable related to an operating process,
a visualization step of visualizing the relationship between a plurality of explanatory variables for the objective variable using a correlation coefficient between the plurality of explanatory variables;
A division step of dividing the plurality of explanatory variables into a plurality of clusters using the relationships between the explanatory variables visualized in the visualization step;
In one or more of the clusters selected from the plurality of clusters, selecting a representative explanatory variable, which is the explanatory variable representing the cluster, from among the explanatory variables included in the cluster for each cluster. process and
a predictive model creating step of creating a predictive model for predicting the objective variable using the representative explanatory variables selected in the selecting step. In the selection step, a predetermined number of the representative explanatory variables are selected according to predetermined conditions,
The condition selects the explanatory variable having a high correlation coefficient with the objective variable, selects the explanatory variable related to an upstream process in the operating process, and selects the explanatory variable that is easy to control in the operating process. and selecting the explanatory variable having a high influence on the objective variable when a predictive model for the objective variable is created using the explanatory variables included in the cluster. A method for creating a prediction model according to claim 1. 3. A method of setting conditions for an operating process, wherein a plurality of explanatory variables are set using a prediction model obtained by the method of creating a prediction model according to claim 1 or 2.

Images