JP7360016B2 - Data processing method, data processing device, and program - Google Patents

Description

The present invention relates to a data processing method, a data processing device, and a program for a computer to form a prediction module that predicts and outputs a value related to a predetermined feature quantity by inputting the values of a plurality of explanatory variables.

In recent years, technologies have been actively proposed that allow computers to perform machine learning to make various predictions from input data. On the other hand, conventionally, a plurality of rubber materials, fillers, oils, etc. have been blended through trial and error to produce a prototype vulcanized rubber composition, and physical property data have been experimentally measured. For this reason, a large amount of data has been accumulated that links the formulation information of vulcanized rubber compositions with the values of physical property data. Utilizing this accumulated data as a learning data set, it is possible to cause a computer to perform machine learning and predict values of physical property data from input data.

For example, using a neural network method, we learn the mapping relationship between a group of factors that are explanatory variables such as design and formulation and a group of characteristics that are features, and estimate the value of the feature from the value of each explanatory variable. , a technique is known that provides a method for efficiently and easily determining the optimal value of an explanatory variable that creates an arbitrary feature value (Patent Document 1).

Japanese Patent Application Publication No. 2003-58582

In neural network learning using this technology, all prepared original data is uniformly read and used as multiple pieces of training data. Original data often includes data obtained from past experiments and simulations. In other words, the original data includes experimental data whose feature values are experimental values of the measurement target, and simulation calculated values whose feature values are calculated by performing a simulation using a simulation model of the measurement target. In many cases, a plurality of simulation data are held.
In simulation, calculations are performed on a computer using a simulation model to reproduce the experiment, but the results obtained by simulation are the values of the features (values calculated by simulation) obtained when various factors (explanatory variables) are changed. ) corresponds to a change in the value of the feature obtained in the experiment (experimental data), but the value of the feature obtained in the simulation may not match the value of the feature obtained in the experiment. There are often deviations.
For this reason, it is difficult to create a prediction module that performs machine learning on the relationship between explanatory variables and features by combining original data including experimental data and simulation data.

Therefore, the present invention provides an original data set including experimental data and simulation data when a computer defines a prediction module that predicts and outputs a value related to a predetermined feature quantity by inputting the values of a plurality of explanatory variables. Provides a data processing method, a data processing device, and a program for causing a computer to execute the data processing method, which can efficiently create a prediction module with high prediction accuracy by machine learning the relationship between explanatory variables and feature quantities using The purpose is to

One aspect of the present invention is a data processing method for a computer to form a prediction module that predicts and outputs a value related to a predetermined feature amount by inputting values of a plurality of explanatory variables. The data processing method is
Data that is held as a set of the values of each of a plurality of explanatory variables and the value of a feature amount for associating the value of the explanatory variable with the value of the explanatory variable, the value of the feature amount being an experimental value of the measurement target. Using an original data set that holds a plurality of experimental data and simulation data in which the value of the feature amount is a simulation calculation value calculated by performing a simulation using a simulation model of the measurement target, Corrected simulation data in which a computer corrects the value of the feature amount in the simulation data based on the correspondence between the value of the feature amount in the simulation data and the value of the feature amount in the experimental data. creating a modified simulation dataset consisting of;
The computer separates each of the experimental data sets composed of the corrected simulation data set and the experimental data into a learning data set and a verification data set, thereby creating a corrected learning simulation data set. , generating a learning experiment data set, a verification modified simulation data set, and a verification experiment data set;
The computer uses each of the corrected learning simulation data set, the learning experiment data set, and the learning integrated data set in which the learning corrected simulation data set and the learning experiment data set are integrated, a step in which a computer creates a plurality of prediction module candidates by machine learning the relationship between the explanatory variable and the feature amount ;
The computer performs machine learning using the modified verification simulation data set, the verification experiment data set, and the verification integrated data set that integrates the verification modified simulation data set and the verification experiment data set. Evaluating prediction accuracy for each of the plurality of prediction module candidates;
The computer includes the step of determining the prediction module from the plurality of prediction module candidates based on the evaluation result of the prediction accuracy.

The simulation data is a simulation calculated by performing the simulation using the simulation model using the values of the explanatory variables that realize the maximum and minimum values of the feature amounts in the plurality of experimental data. Contains calculated values,
In the correction of the value of the feature amount, a correspondence relationship between the maximum value and the minimum value and simulation calculation values corresponding to the maximum value and the minimum value, respectively, and the value of the feature amount of the experimental data are determined. exists between the maximum value and the minimum value, and the values of the explanatory variables match within a permissible range.Use a correspondence relationship between the simulation calculation value and the value of the feature quantity in the experimental data. It is preferable that the value of the feature amount of the corrected learning simulation data set is corrected.

When evaluating the prediction accuracy, for the prediction module candidates that have been machine learned using the integrated training dataset, the corrected verification simulation dataset, the verification experiment dataset, and the verification integrated dataset are evaluated. It is preferable to evaluate the prediction accuracy when using each.

When evaluating the prediction accuracy, for prediction module candidates that have been machine learned using the integrated learning dataset,
(1) Prediction accuracy when using the experimental data set for verification, and prediction accuracy when using the experimental data set for verification in prediction module candidates that are machine learned using the experimental data set for learning. Compare,
(2) Prediction accuracy when using the corrected simulation data set for verification and prediction when using the corrected simulation data set for verification in prediction module candidates that are machine learned using the corrected simulation data set for learning Compare the accuracy and
Preferably, prediction module candidates subjected to machine learning using the integrated learning data set are evaluated based on the comparison results.

The simulation data includes first simulation data and second simulation data that differ in at least one of the configuration of the simulation model and the method of the simulation,
creating the modified simulation data set using each of the first simulation data and the second simulation data; generating the modified simulation data set for learning and the modified simulation data set for verification; and the prediction. Preferably, the steps of creating module candidates and evaluating the prediction accuracy are performed.

The feature quantity is a physical quantity that acts on the tire,
Preferably, the value of the explanatory variable is a value that defines the tire.

Furthermore, in response to the input of the target value regarding the feature amount, the computer calculates an optimal value regarding the explanatory variable that reproduces the target value using the prediction module,
In the step of calculating the optimal value, the optimal value for the explanatory variable is calculated based on the value of the feature quantity predicted by the prediction module according to the value of the explanatory variable input to the prediction module. It is preferable.

Preferably, the method further includes the step of visualizing the relationship between the value of the explanatory variable and the value of the feature amount.

Another aspect of the present invention is a data processing device configured with a computer that forms a prediction module that predicts and outputs the value of a predetermined feature quantity by inputting the values of a plurality of explanatory variables. The data processing device is
Data that is held as a set of the values of each of a plurality of explanatory variables and the value of a feature amount for associating the value of the explanatory variable with the value of the feature amount, the value of the feature amount being obtained using a testing machine. holding a plurality of experimental data, which are experimental values of the measured object, and simulation data, in which the value of the feature amount is a simulation calculated value calculated by performing a simulation using a simulation model of the measured object. Correcting the value of the feature amount in the simulation data based on the correspondence between the value of the feature amount in the simulation data and the value of the feature amount in the experimental data using the original data set. and a data correction unit that creates a corrected simulation data set composed of the corrected simulation data;
By separating each of the experimental data sets composed of the modified simulation data set and the experimental data into a training data set and a verification data set, the modified simulation data set for learning and the experiment for learning can be separated. a dataset generation unit that generates a dataset, a modified simulation dataset for verification, and an experimental dataset for validation;
The computer uses each of the modified learning simulation data set, the learning experiment data set, and the learning integrated data set that integrates the learning modified simulation data set and the learning experiment data set. a prediction module candidate creation unit that creates a plurality of prediction module candidates by machine learning the relationship between the explanatory variable and the feature amount ;
The computer performs machine learning using the modified verification simulation data set, the verification experiment data set, and the verification integrated data set that integrates the verification modified simulation data set and the verification experiment data set. a prediction module candidate evaluation unit that evaluates prediction accuracy for each of the plurality of prediction module candidates;
The computer includes a prediction module determining unit that determines the prediction module from the plurality of prediction module candidates based on the evaluation result of the prediction accuracy.

Yet another aspect of the present invention causes a computer to execute a data processing method for forming a prediction module that predicts and outputs the value of a predetermined feature amount by inputting the values of a plurality of explanatory variables. It is a program. The program is
Data that is held as a set of the values of each of a plurality of explanatory variables and the value of a feature amount for associating the value of the explanatory variable with the value of the feature amount, the value of the feature amount being obtained using a testing machine. holding a plurality of experimental data, which are experimental values of the measured object, and simulation data, in which the value of the feature amount is a simulation calculated value calculated by performing a simulation using a simulation model of the measured object. Using the original data set, a computer calculates the value of the feature in the simulation data based on the correspondence between the value of the feature in the simulation data and the value of the feature in the experimental data. a step of correcting the values to generate a corrected simulation data set composed of the corrected simulation data;
By causing the computer to separate each of the experimental data sets composed of the corrected simulation data set and the experimental data into a learning data set and a verification data set, a corrected simulation data set for learning is created. , a procedure for generating a learning experiment data set, a verification modified simulation data set, and a verification experiment data set;
Using each of the corrected learning simulation data set, the learning experiment data set, and the learning integrated data set that integrates the learning corrected simulation data set and the learning experiment data set in the computer, A step in which a computer creates a plurality of prediction module candidates by machine learning the relationship between the explanatory variable and the feature amount ;
The computer performs machine learning using the modified verification simulation data set, the verification experiment data set, and the verification integrated data set that integrates the verification modified simulation data set and the verification experiment data set. a step of evaluating prediction accuracy for each of the plurality of prediction module candidates;
and a step of causing the computer to determine the prediction module from the plurality of prediction module candidates based on the evaluation result of the prediction accuracy.

According to the data processing method, data processing device, and program described above, when a computer determines a prediction module that predicts and outputs a value related to a predetermined feature amount by inputting the values of a plurality of explanatory variables, an experiment is performed. Using an original data set containing data and simulation data, it is possible to efficiently create a prediction module with high prediction accuracy by machine learning the relationship between explanatory variables and feature quantities.

FIG. 2 is a diagram schematically explaining an example of the flow of a data processing method according to an embodiment. FIG. 1 is a diagram illustrating an example of the configuration of a data processing device according to an embodiment. (a) to (c) are diagrams illustrating an example of correction of the value of a feature amount in simulation data performed by the data processing method of one embodiment. FIG. 2 is a diagram illustrating an example of an original learning data set used in a data processing method according to an embodiment in a simplified and easy-to-understand manner. (a) to (c) are diagrams showing examples of datasets created from original datasets in the data processing method of one embodiment. FIG. 3 is a diagram illustrating an example of a data set created from an original data set in the data processing method of one embodiment. FIG. 2 is a diagram illustrating an example of how to create a prediction module candidate and use a verification data set in the data processing method of one embodiment. FIG. 3 is a diagram illustrating the correspondence between the value of a feature amount in simulation data and the value of a feature amount in experimental data used in the data processing method of one embodiment.

Hereinafter, a data processing method, a data processing device, and a program according to one embodiment will be described based on the attached figures.
FIG. 1 is a diagram schematically explaining an example of the flow of a data processing method according to an embodiment. FIG. 2 is a diagram illustrating an example of the configuration of a data processing device according to an embodiment.
The data processing method of one embodiment is a method executed by a computer, and is a method of creating a prediction module that predicts and outputs a value related to a predetermined feature quantity by inputting the values of a plurality of explanatory variables. .
The prediction module is evaluated using multiple validation datasets created separately from the original dataset, from among multiple prediction module candidates created using multiple training datasets created from the original dataset. It is determined based on the evaluation results.

The data processing device 10 shown in FIG. 2 is constituted by a computer including a CPU 12 and a memory 14. Connected to the data processing device 10 are a display 30 and an input operation device 32 including a mouse and a keyboard for inputting information.
The input operation device 32 is used by an operator to input desired instructions to the data processing apparatus 10. For example, an operator instructs and inputs information from the input operation device 32 in order to set conditions for creating a prediction module candidate.
The display 30 is used to display set information, for example, an original data set used in the data processing method, an original data set for learning, an original data set for verification, various sub-data sets for learning, and sub-data sets for verification. Displays numerical values of data in the data set, explanatory variables, missing explanatory variables, a condition setting screen for creating prediction module candidates, evaluation results of prediction accuracy in prediction module candidates, etc.

A program is stored in the memory 14, and when the CPU 12 reads and executes the program, the simulation data correction section 15, the sub-data set creation section 16, the prediction module candidate creation section 18, the prediction module candidate evaluation section 20, The prediction module determining section 22 and the prediction section 24 are made to function as software modules. Hereinafter, the functions of the simulation data correction unit 15, sub-data set creation unit 16, prediction module candidate creation unit 18, prediction module candidate evaluation unit 20, prediction module determination unit 22, and prediction unit 24 will be described in an embodiment shown in FIG. The flow of the data processing method will be explained at the same time.

By performing machine learning, the computer previously holds a prediction model that can serve as a prediction module. This prediction model becomes a prediction module candidate by performing machine learning using a plurality of learning sub-data sets created from the original data set. At least one of these prediction module candidates becomes a prediction module. Prediction models include models using neural networks such as the well-known deep learning, models using the well-known random forest method that uses multiple decision trees for "classification" or "regression," and LASSO regression. Including models using . Further, as a prediction model, a nonlinear function using a polynomial, kriging, or RBF (Radial Base Function) can also be used.

The original data set is a group of data that holds multiple sets (for example, tens of thousands of sets) of the values of multiple explanatory variables and the values of features used to correlate these explanatory variable values. . The explanatory variables include, for example, the design dimensions of the product, the structure and physical properties of the constituent materials used in the product, or the manufacturing conditions when manufacturing the product, and the feature quantities include, for example, the characteristic values of the product, the sales volume in the market. Including etc. For example, if the original data set includes the design dimensions of the structure and the structure of the constituent materials as explanatory variables, and the characteristic values of the structure as feature quantities, the data may include various changes in the design dimensions and structure. This refers to data consisting of information on the above-mentioned design dimensions and the above-mentioned structure, and characteristic values. Therefore, in this case, the original data set includes a large amount of data in which the design dimensions, the design dimensions when the structure is variously changed, information on the structure, and characteristic values are set.

The original data set is often a huge amount of data accumulated in the past. The original data set holds a large amount of data that is a set of values of each of a plurality of explanatory variables and values of feature amounts for associating the values of the explanatory variables. The large amount of data includes a large amount of experimental data and a large amount of simulation data. The experimental data is data in which the value of the feature amount is an experimental value of the measurement object. The simulation data is a simulation calculation value in which the value of the feature amount is calculated by performing a simulation using a simulation model of the measurement target object. The simulation data shown in FIG. 1 includes simulation data 1 and simulation data 2 that have simulation calculation values that are feature values calculated using different simulation models or by different simulation methods.

The simulation data modification unit 15 modifies the simulation calculation values (feature values) in the simulation data of the original data set using the experimental data (ST10 in FIG. 1). As a result, a modified simulation data set is created, which is composed of a plurality of modified simulation data, each of which is a set of modified explanatory variables and feature values. A dataset consisting of experimental data is called an experimental dataset.

FIGS. 3(a) to 3(c) are diagrams illustrating an example of correction of the value of a feature amount in simulation data. FIG. 3A shows an example of the values of the feature amounts of the experimental data and the simulation data when the explanatory variables of the experimental data and the explanatory variables of the simulation data match within a permissible range. FIG. 3(b) shows the results of plotting the values of the feature amounts shown in FIG. 3(a) on a graph in which the horizontal axis is the feature amount of simulation data and the vertical axis is the feature amount of experimental data.
As shown in FIG. 3(b), the values of the feature values of simulation data and experimental data have a one-to-one corresponding relationship.Using this relationship, as shown in FIG. 3(c), The value of the feature amount in the simulation data is converted into the value of the experimental data, and the value is determined as the value of the modified simulation data.
Note that when the simulation data includes simulation data 1 and simulation data 2, a relationship as shown in FIG. The value of the quantity is converted into the value of the experimental data to obtain the value of the corrected simulation data.
Note that when the values of the feature values of the experimental data and the values of the feature values of the simulation data are arranged in the order of the size of the values, if the order differs between the experimental data and the simulation data, the experimental data and the simulation data Instead of the values of two feature quantities having different ranks of simulation data, the average value of these two values may be used as the value of the new feature quantity.

Next, the sub-dataset creation unit 16 converts each of the corrected simulation data sets 1 and 2 made up of corrected simulation data and the experimental data set made up of experimental data into a learning data set and a verification data set. By separating the corrected simulation data sets 1 and 2 for learning and the experimental data set for learning as learning sub-data sets, the corrected simulation data sets 1 and 2 for verification and the experimental data set for verification are created as sub-datasets for learning. Create a sub-dataset for verification as a sub-dataset, and further create an integrated training dataset that integrates the modified learning simulation data sets 1 and 2 and the experimental learning dataset as a learning sub-dataset, and create the modified verification simulation data. An integrated verification data set that integrates sets 1 and 2 and the verification experimental data set is created as a verification sub-data set (ST12, ST14 in FIG. 1).

FIG. 4 is a diagram for explaining in a simplified manner an example of an original dataset for learning when the original dataset is separated into an original dataset for learning and an original dataset for verification. ) and FIG. 6 are diagrams illustrating examples of learning sub-data sets created from the original learning data set in the data processing method of one embodiment.
The original data set for learning shown in FIG. 4 includes explanatory variables X ₁ to X _n (n is a natural number) as explanatory variables, and data 1 to data 9 as data for each explanatory variable. In the original data set for learning shown in FIG. 4, the number of data is nine, but the number of data is actually several thousand to tens of thousands.
In the original data set for learning shown in FIG. 4, there is one feature amount, but there may be more than one feature amount.
Here, "..." in FIG. 4 indicates that there is actually a numerical value. Regarding the "simulation index" in the figure, "0" indicates non-simulation data, "1" indicates data obtained by simulation 1, and "2" indicates a simulation model or simulation method. This indicates that the data was obtained by Simulation 2, which differs from Simulation 1 in this respect. The "test machine index" in the figure indicates the type of test machine in the case of non-simulation data. "1" indicates that the data is experimental data obtained by the tester 1, and "0" indicates that the data is not experimental data obtained using the tester. In addition, FIG. 4 shows that the values of the feature quantities of data 1 to 9 are V1 to V9.

FIG. 5(a) shows a learning experiment data set composed of data whose "simulation index" of the original data for learning is "0", and FIG. 5(b) shows a "simulation index" of the original data for learning. Figure 5(c) shows a modified learning simulation set 1 composed of data whose "simulation index" is "1", and FIG. The revised simulation set 2 is shown in FIG. Since the values of the feature amounts in the learning modified simulation sets 1 and 2 are corrected values, the values of the feature amounts shown in FIGS. 5(b) and 5(c) are V4' to V9'.
FIG. 6 shows an integrated data set for learning, which is composed of modified simulation data and experimental data.

In the original data set _, in addition to X ₁ to The number of explanatory variables in the training data sets 1 and 2 and the learning integrated data set is n, X ₁ to X _n .

The prediction module candidate creation unit 18 performs machine learning on a prediction model using the original training data set and the created plurality of training sub-data sets to create prediction module candidates 1 to 5 (ST16 in FIG. 1). In the machine learning of the prediction model, deep learning is used, and, for example, prediction module candidates with a layered structure based on input set conditions are created, for example, prediction module candidates with a layered structure of 1 to 7 layers.

FIG. 7 is a diagram illustrating an example of how to create a prediction module candidate and use a verification data set, which will be described later.
Prediction module candidates 1 to 5 use the above-mentioned original training data set, modified learning simulation data sets 1 and 2, learning experiment data set, and learning integrated data set, so that the prediction model can be used as an explanatory variable. It was created through machine learning of the relationships between features. Transfer learning methods can also be used in the machine learning of the prediction module.

Therefore, in the prediction module candidate 1 created from the original data set for learning, X ₁ to X _n , "simulation index" and "test machine index" are defined as explanatory variables. Therefore, the number of explanatory variables in this case is n+2. In prediction module candidates 2 to 5 created from each of the modified learning simulation data sets 1 and 2, the learning experimental data set, and the integrated learning data set, X ₁ to X _n are defined as explanatory variables. Therefore, there are n explanatory variables in this case.

The prediction module candidate evaluation unit 20 generates a modified verification simulation data set 1 and 2, a verification experiment data set, and a verification integrated data set that integrates the verification modified simulation data sets 1 and 2 and the verification experiment data set. The prediction accuracy is evaluated for each of the prediction module candidates 1 to 5 that have been subjected to machine learning (ST 18 in FIG. 1).

The modified simulation data sets 1 and 2 for verification, the experimental data set for verification, and the integrated data set for verification prepared as sub-datasets for verification are the modified simulation data sets 1 and 2 for learning, the experimental data set for learning, and the learning Similar to the integrated data set for prediction, the corrected simulation data set for verification 1 and ₂ , the _experimental data set for verification, and the integrated data set for verification have the explanatory variables X 1 to It can be used as a sub-data set for each verification. For example, the predicted value of the feature amount can be calculated using each of the prediction module candidate 2, the modified verification simulation data sets 1 and 2, the verification experiment data set, and the verification integrated data set. Therefore, the predicted value of the feature amount calculated by the prediction module 2 can be compared with the value of the feature amount of the corrected simulation data sets 1 and 2 for verification, the experimental data set for verification, and the unified data set for verification. Similarly, for prediction module candidates 3 to 5, the predicted values of the calculated features are used as the correct answer for the corrected simulation data sets 1 and 2 for verification, the experimental data set for verification, and the integrated data set for verification. Can be compared as values.

In evaluating the prediction accuracy of the feature quantities in the prediction module candidates 1 to 5, it is evaluated to what extent the predicted values of the feature quantities predicted by the prediction module candidates 1 to 5 each approximate the correct value. The method of evaluation is not particularly limited, but for example, a value representing the ratio of the predicted value to the correct value is used as the evaluation value. When a plurality of feature quantities are set, the average value of the ratios for each feature quantity or the value where the ratio is farthest from 1 is used as the evaluation value. Alternatively, since there are many sets of actual feature values and predicted values by prediction module candidates, the evaluation value may be the correlation coefficient R or determination coefficient ^R2 between the actual feature values and predicted values. You can also do it.

The prediction module determination unit 22 determines a prediction module with high prediction accuracy based on the prediction accuracy evaluation result (evaluation value) obtained by the prediction module candidate evaluation unit 20 (ST20 in FIG. 1). The prediction module to be determined may be determined by selecting the one with the highest prediction accuracy from a plurality of prediction module candidates, or by determining a plurality of prediction module candidates whose prediction accuracy exceeds a threshold as the prediction module. It's okay. Among the prediction module candidates, prediction module candidate 1 with the most explanatory variables is not necessarily the prediction module candidate with the highest prediction accuracy. There are also explanatory variables that do not contribute to the feature amounts, and these explanatory variables may become noise components and reduce prediction accuracy.
Preferably, the information on the prediction accuracy evaluation results is displayed on the display 30.

The prediction unit 24 sets the determined prediction module and predicts a value related to the feature amount by inputting the value of the explanatory variable. Values related to the predicted feature amounts are output to the display 30.

In this way, in the data processing method described above, when simulation data and experimental data are included, by modifying the feature values of the simulation data, the modified simulation data and experimental data can be used simultaneously as the same learning data. In addition, since the prediction model can be machine learned using learning modified simulation data sets for different types of modified simulation data, such as modified simulation data sets 1 and 2, it is possible to create multiple prediction module candidates. Can be done. Furthermore, as sub-datasets for verification, each of the corrected simulation data for verification, the experimental data set for verification, and the integrated data set for verification are used, that is, the original data set for verification is efficiently used to generate multiple prediction module candidates. It is possible to evaluate the prediction accuracy of Therefore, even when the original data set includes experimental data and simulation data, it is possible to efficiently create a prediction module with high prediction accuracy by machine learning the relationship between explanatory variables and feature amounts.

According to one embodiment, when dividing the original data into a training dataset and a validation dataset, the validation dataset is taken from a different part of the original dataset and the remaining part is the training dataset. is performed multiple times, and each time it is divided, the prediction accuracy of the prediction module candidates created using the training dataset is evaluated, and predictions are made from the prediction module candidates based on the average value of the prediction accuracy evaluation results performed multiple times. Preferably, the module is determined. As a result, it is possible to create a training dataset for machine learning without being biased over a wide range of the original dataset, and it is also possible to use a validation dataset for verification over a wide range without bias. , it is possible to accurately evaluate the prediction accuracy.

According to one embodiment, the simulation data is calculated by performing a simulation using a simulation model using values of explanatory variables that realize the maximum and minimum values of feature quantities in a plurality of experimental data. Including simulation calculation values, correction of feature value values requires the correspondence between the maximum and minimum values and the simulation calculation values corresponding to the maximum and minimum values, and the feature value of experimental data. A learning simulation data set is created by using the correspondence between the simulated calculated values, which exist between the maximum and minimum values, and the values of the explanatory variables match within the allowable range, and the feature values in the experimental data. Preferably, the value of the feature amount is modified. The simulation is not particularly limited, but includes, for example, a simulation using a well-known finite element model.
FIG. 8 is a diagram illustrating the correspondence between the value of the feature amount in the simulation data, that is, the value calculated by the simulation, and the value of the feature amount in the experimental data. If there is a value of the feature quantity of the simulation data corresponding to the value of the explanatory variable that realizes the maximum value and minimum value of the feature quantity in the experimental data, that is, a simulation calculated value, two simulation calculated values corresponding to the maximum value and the minimum value. Correcting the calculated values between the two can be performed with high precision using interpolation. Therefore, if the simulation data does not have the feature values in the simulation data that correspond to the maximum and minimum values of the feature values in the experimental data, by performing a simulation using a simulation model, it is possible to maximize the feature values in the experimental data. Simulation calculation values corresponding to the values and the minimum values can be easily calculated.

In addition, the value of the feature amount in the experimental data exists between the maximum value and the minimum value, and the values of the explanatory variables match within the allowable range.The correspondence between the simulation calculated value and the value of the feature amount in the experimental data. By using the relationship, the value of the feature amount of the learning simulation data set can be corrected with high precision by interpolation. Even in this case, if there is no simulation data in which the explanatory variables in the experimental data match the values of the explanatory variables within the allowable range, by performing a simulation using a simulation model, a simulation that corresponds to the value of the feature in the experimental data can be performed. Calculated values can be easily calculated.
In this way, as shown in FIG. 8, it is possible to associate the values of the feature amounts between the simulation data and the experimental data when the explanatory variables match within the allowable range. Therefore, by interpolation, values can be corrected with high precision.

According to one embodiment, when evaluating prediction accuracy, as shown in FIG. Preferably, prediction accuracy is evaluated using each of the experimental data set and the integrated verification data set. Although it is generally assumed that a prediction module candidate that has undergone machine learning using an integrated training dataset has a higher prediction accuracy than any other prediction module candidate, the prediction accuracy may not always be high. For this reason, for prediction module candidates that have been machine learned using the integrated training dataset, the prediction accuracy should be evaluated by using various validation sub-datasets created from the original validation dataset as much as possible. is preferred.

According to one embodiment, when evaluating prediction accuracy, for prediction module candidates subjected to machine learning using an integrated training dataset,
(1) Compare the prediction accuracy when using the experimental data set for verification and the prediction accuracy when using the experimental data set for verification in the prediction module candidates that were machine learned using the experimental data set for learning,
(2) Compare the prediction accuracy when using the modified simulation dataset for verification and the prediction accuracy when using the modified simulation set for verification for prediction module candidates that were machine learned using the modified simulation dataset for learning. death,
Preferably, prediction module candidates subjected to machine learning using the integrated learning data set are evaluated based on the comparison results. The prediction accuracy of the feature value of the prediction module candidate machine-learned using the integrated training dataset is the same as that of the prediction module candidate machine-learned using the learning experiment dataset when using the verification experiment dataset. The prediction accuracy is improved compared to the prediction accuracy, and the prediction accuracy of the prediction module candidates learned by machine learning using the modified training simulation dataset is improved compared to the prediction accuracy when using the modified verification simulation set. is generally assumed, but the prediction accuracy may not always be high. For this reason, for prediction module candidates that have been machine-learned using the integrated training dataset, the prediction accuracy of the prediction module candidates created from the experimental dataset is compared with the prediction accuracy when the experimental data of the experimental dataset is used as verification data. It is especially preferable to compare the prediction accuracy of the prediction module candidates created from the modified simulation data set with the prediction accuracy when the modified simulation data is used as verification data.

The simulation data may be one type of simulation data in which the configuration of the simulation model and the simulation method are the same, but as shown in FIG. 1 (first simulation data) and simulation data 2 (second simulation data). In this case, it is preferable to perform the processing of ST10 to ST20 shown in FIG. 1 using each of simulation data 1 and simulation data 2. This makes it possible to evaluate the prediction accuracy of a plurality of prediction module candidates based on differences in simulations, so it is possible to determine a prediction module with high prediction accuracy.

According to one embodiment, it is preferable that the feature quantity is a physical quantity that acts on the tire, such as a characteristic value of the tire, and the value of the explanatory variable is a value that defines the tire. This makes it possible to predict with high accuracy the physical quantities that act on the tire using values that define the tire. The values that define the tire include, for example, the rim size on which the tire is mounted, the aspect ratio of the tire, the tire width, the cross-sectional area of the bead filler, the angle of the first steel cord, the rigidity of the first steel cord, the angle of the second steel cord, and the second steel cord. 2 steel cord, the angle of the first carcass cord, the rigidity of the first carcass cord, the angle of the second carcass cord, the rigidity of the second carcass cord, etc.

According to one embodiment, the prediction module can also be used in an optimization process that calculates an optimal value for an explanatory variable that reproduces the target value in response to input of a target value for the feature quantity. That is, in the data processing method of one embodiment, in response to input of a target value regarding a feature amount, the data processing device 10 performs an optimization process of calculating an optimal value regarding an explanatory variable that reproduces the target value using a prediction module. It is preferable to include. In this case, it is preferable to calculate the optimal value for the explanatory variable based on the value of the feature quantity predicted by the prediction module in accordance with the value of the explanatory variable input to the prediction module. For example, it is preferable that an evolutionary algorithm be used as the method for calculating the optimal value. Evolutionary algorithms include Genetic Algorithm, Differential Evolution, Particle Swarm Optimization, Ant Colony Optimization, etc. It is also preferable to use the experimental design method or the Latin hypercube method.

According to one embodiment, it is preferable to visualize the relationship between the value of the explanatory variable and the value of the feature amount.
The relationship between the value of the explanatory variable and the value of the feature amount is displayed on the display 30. The relationship between the value of the explanatory variable and the value of the feature amount is represented by, for example, a self-organizing map. Alternatively, instead of the self-organizing map, a scatter diagram may be used to visualize the relationship between explanatory variables and feature values.

Such a data processing method can be achieved by reading a program to be executed by a computer from the memory 14 and executing it. Therefore, this program:
(1) Using an original data set containing multiple experimental data and simulation data, a computer performs a simulation based on the correspondence between the feature values in the simulation data and the feature values in the experimental data. a step of correcting the value of the feature quantity in the data to generate a corrected simulation data set composed of the corrected simulation data;
(2) By having the computer separate each experimental data set consisting of a corrected simulation data set and experimental data into a learning data set and a verification data set, a corrected simulation data set for learning and a learning data set are created. a procedure for generating an experimental data set for use, a modified simulation data set for verification, and an experimental data set for verification;
(3) A computer uses each of the corrected simulation data set for learning, experimental data set for learning, and integrated data set for learning to perform machine learning on the relationship between explanatory variables and characteristic quantities. A procedure for creating a prediction module candidate,
(4) a step of causing the computer to evaluate the prediction accuracy of each of the plurality of prediction module candidates subjected to machine learning using the corrected simulation data set for verification, the experimental data set for verification, and the integrated data set for verification;
(5) A procedure for causing the computer to determine a prediction module from a plurality of prediction module candidates based on the evaluation result of prediction accuracy.

(Example, comparative example)
In order to confirm the effectiveness of the data processing method described above, 10,881 pieces of experimental data and 3,739 pieces of simulation data were prepared. The explanatory variables include tire dimensions, the dimensions of the constituent materials of the tire, physical property values, and the form of the tire structure as information, and rolling resistance was used as the feature quantity.

In the example, an original data set including experimental data and simulation data was prepared, and this original data set was processed by the above-described data processing method to determine a prediction module. There are three prediction module candidates: a prediction module candidate created from the original data set for learning, a prediction module candidate created from the integrated training data set, and a prediction module candidate created from the experimental data for learning.

On the other hand, in the comparative example, a prediction module was determined using an original data set that included experimental data but did not include simulation data. In this case, only one prediction module candidate is created from the learning experimental data set, and the number of prediction module candidates is one, so this prediction module candidate automatically becomes the prediction module in the comparative example.

The evaluation results of the prediction module candidates in the example were as follows.
The prediction module candidates were created by subjecting the prediction model to machine learning using a deep learning method. The layer structure in deep learning was three layers.
The coefficient of determination R ² between the predicted value using the original data set for verification in the prediction module candidate created from the original data set for learning and the value of the feature amount in the original data set for verification is as low as 0.71.
Predicted values of features using the verification experiment data set, verification modified simulation data set, and verification integrated data set in the prediction module candidates created from the learning experiment data set, and the corresponding features in the above datasets. The coefficient of determination ^R2 between the value of the quantity is 0.77,
Predicted values of feature quantities using the experimental validation dataset, modified simulation dataset for validation, and integrated validation dataset in the prediction module candidates created from the integrated training dataset, and the corresponding features in the above dataset. The coefficient of determination ^R2 between the quantity values was 0.88.

On the other hand, since the prediction module candidate created in the comparative example is only one prediction module candidate created from the above-mentioned learning experiment data set, the determination coefficient R ² of the prediction module candidate is 0.77.

Therefore, the coefficient of determination R ² of the prediction module determined in the example is 0.88, and the coefficient of determination R ² of the prediction module determined in the comparative example is 0.77.
From this, it can be said that the prediction accuracy of the prediction module of the example is high.

Although the data processing method, data processing device, and program of the present invention have been described in detail above, the present invention is not limited to the above embodiments, and various improvements and changes can be made without departing from the gist of the present invention. Of course you can.

10 Data processing device 12 CPU
14 Memory 15 Simulation data modification section 16 Sub-data set creation section 18 Prediction module candidate creation section 20 Prediction module candidate creation section 22 Prediction module determination section 24 Prediction section

Claims (10)

A data processing method for forming a prediction module in which a computer predicts and outputs a value related to a predetermined feature amount by inputting values of a plurality of explanatory variables, the method comprising:
Data that is held as a set of the values of each of a plurality of explanatory variables and the value of a feature amount for associating the value of the explanatory variable with the value of the explanatory variable, the value of the feature amount being an experimental value of the measurement target. Using an original data set that holds a plurality of experimental data and simulation data in which the value of the feature amount is a simulation calculation value calculated by performing a simulation using a simulation model of the measurement target, Corrected simulation data in which a computer corrects the value of the feature amount in the simulation data based on the correspondence between the value of the feature amount in the simulation data and the value of the feature amount in the experimental data. creating a modified simulation dataset consisting of;
The computer separates each of the experimental data sets composed of the corrected simulation data set and the experimental data into a learning data set and a verification data set, thereby creating a corrected learning simulation data set. , generating a learning experiment data set, a verification modified simulation data set, and a verification experiment data set;
The computer uses each of the corrected learning simulation data set, the learning experiment data set, and the learning integrated data set in which the learning corrected simulation data set and the learning experiment data set are integrated, a step in which a computer creates a plurality of prediction module candidates by machine learning the relationship between the explanatory variable and the feature amount;
The computer performs machine learning using the modified verification simulation data set, the verification experiment data set, and the verification integrated data set that integrates the verification modified simulation data set and the verification experiment data set. Evaluating prediction accuracy for each of the plurality of prediction module candidates;
A data processing method comprising: the computer determining the prediction module from the plurality of prediction module candidates based on the evaluation result of the prediction accuracy. The simulation data is a simulation calculated by performing the simulation using the simulation model using the values of the explanatory variables that realize the maximum and minimum values of the feature amounts in the plurality of experimental data. Contains calculated values,
In the correction of the value of the feature amount, a correspondence relationship between the maximum value and the minimum value and simulation calculation values corresponding to the maximum value and the minimum value, respectively, and the value of the feature amount of the experimental data are determined. exists between the maximum value and the minimum value, and the values of the explanatory variables match within a permissible range.Use a correspondence relationship between the simulation calculation value and the value of the feature quantity in the experimental data. The data processing method according to claim 1, wherein the value of the feature amount of the modified learning simulation data set is modified. When evaluating the prediction accuracy, for the prediction module candidates that have been machine learned using the integrated training dataset, the corrected verification simulation dataset, the verification experiment dataset, and the verification integrated dataset are evaluated. The data processing method according to claim 1 or 2, wherein prediction accuracy is evaluated when each is used. When evaluating the prediction accuracy, for prediction module candidates that have been machine learned using the integrated learning dataset,
(1) Prediction accuracy when using the experimental data set for verification, and prediction accuracy when using the experimental data set for verification in prediction module candidates that are machine learned using the experimental data set for learning. Compare,
(2) Prediction accuracy when using the corrected simulation data set for verification and prediction when using the corrected simulation data set for verification in prediction module candidates that are machine learned using the corrected simulation data set for learning Compare the accuracy and
The data processing method according to any one of claims 1 to 3, wherein prediction module candidates subjected to machine learning using the integrated learning data set are evaluated based on the comparison results. The simulation data includes first simulation data and second simulation data that differ in at least one of the configuration of the simulation model and the method of the simulation,
creating the modified simulation data set using each of the first simulation data and the second simulation data; generating the modified simulation data set for learning and the modified simulation data set for verification; and the prediction. 5. The data processing method according to claim 1, further comprising creating module candidates and evaluating the prediction accuracy. The feature quantity is a physical quantity that acts on the tire,
The data processing method according to claim 1, wherein the value of the explanatory variable is a value that defines the tire. Furthermore, in response to the input of the target value regarding the feature amount, the computer calculates an optimal value regarding the explanatory variable that reproduces the target value using the prediction module,
In the step of calculating the optimal value, the optimal value for the explanatory variable is calculated based on the value of the feature quantity predicted by the prediction module according to the value of the explanatory variable input to the prediction module. The data processing method according to any one of claims 1 to 6. The data processing method according to any one of claims 1 to 7, further comprising the step of visualizing the relationship between the value of the explanatory variable and the value of the feature amount. A data processing device configured with a computer that forms a prediction module that predicts and outputs the value of a predetermined feature quantity by inputting the values of a plurality of explanatory variables,
Data that is held as a set of the values of each of a plurality of explanatory variables and the value of a feature amount for associating the value of the explanatory variable with the value of the feature amount, the value of the feature amount being obtained using a testing machine. holding a plurality of experimental data, which are experimental values of the measured object, and simulation data, in which the value of the feature amount is a simulation calculated value calculated by performing a simulation using a simulation model of the measured object. Correcting the value of the feature amount in the simulation data based on the correspondence between the value of the feature amount in the simulation data and the value of the feature amount in the experimental data using the original data set. and a data correction unit that creates a corrected simulation data set composed of the corrected simulation data;
By separating each of the experimental data sets composed of the modified simulation data set and the experimental data into a training data set and a verification data set, the modified simulation data set for learning and the experiment for learning can be separated. a dataset generation unit that generates a dataset, a modified simulation dataset for verification, and an experimental dataset for validation;
The computer uses each of the modified learning simulation data set, the learning experiment data set, and the learning integrated data set that integrates the learning modified simulation data set and the learning experiment data set. a prediction module candidate creation unit that creates a plurality of prediction module candidates by machine learning the relationship between the explanatory variable and the feature amount;
The computer performs machine learning using the modified verification simulation data set, the verification experiment data set, and the verification integrated data set that integrates the verification modified simulation data set and the verification experiment data set. a prediction module candidate evaluation unit that evaluates prediction accuracy for each of the plurality of prediction module candidates;
The data processing device is characterized in that the computer includes a prediction module determining unit that determines the prediction module from the plurality of prediction module candidates based on the evaluation result of the prediction accuracy. A program that causes a computer to execute a data processing method for forming a prediction module that predicts and outputs the value of a predetermined feature quantity by inputting the values of a plurality of explanatory variables, the program comprising:
Data that is held as a set of the values of each of a plurality of explanatory variables and the value of a feature amount for associating the value of the explanatory variable with the value of the feature amount, the value of the feature amount being obtained using a testing machine. holding a plurality of experimental data, which are experimental values of the measured object, and simulation data, in which the value of the feature amount is a simulation calculated value calculated by performing a simulation using a simulation model of the measured object. Using the original data set, a computer calculates the value of the feature in the simulation data based on the correspondence between the value of the feature in the simulation data and the value of the feature in the experimental data. a step of correcting the values to generate a corrected simulation data set composed of the corrected simulation data;
By causing the computer to separate each of the experimental data sets composed of the corrected simulation data set and the experimental data into a learning data set and a verification data set, a corrected simulation data set for learning is created. , a procedure for generating a learning experiment data set, a verification modified simulation data set, and a verification experiment data set;
Using each of the corrected learning simulation data set, the learning experiment data set, and the learning integrated data set that integrates the learning corrected simulation data set and the learning experiment data set in the computer, A step in which a computer creates a plurality of prediction module candidates by machine learning the relationship between the explanatory variable and the feature amount;
The computer performs machine learning using the modified verification simulation data set, the verification experiment data set, and the verification integrated data set that integrates the verification modified simulation data set and the verification experiment data set. a step of evaluating prediction accuracy for each of the plurality of prediction module candidates;
A program comprising: causing the computer to determine the prediction module from the plurality of prediction module candidates based on the evaluation result of the prediction accuracy.

Images