Performance prediction device and program

Abstract

To provide a performance prediction apparatus and program for calculating the accuracy of prediction performance in a short time.SOLUTION: A machine learning model 40 which is trained with multiple pieces of training data 110 representing each of manufacturing conditions of a product and configured to receive, as input, predetermined input data 210, outputs a prediction value 230 indicating property of the product to be manufactured by the manufacturing conditions represented by the input data 210. A dimensional compression unit 50 performs dimensional compression on the multiple training data 110 and the input data 210, respectively. A similarity determination unit 60 determines one piece of the dimensionally compressed training data 110 which is similar to the dimensionally compressed input data 210. An output unit 80 outputs the prediction value 230 and outputs information, as accuracy of the prediction value 230, in accordance with a determination result of the dimensional compression unit 50.SELECTED DRAWING: Figure 2

Description

[Technical Field]

[0001] This disclosure relates to a performance prediction device and program. [Background technology]

[0002] It is known that machine learning models obtained using machine learning techniques such as neural networks are used to make predictions for given input manufacturing condition data. For example, Patent Document 1 describes that when a prediction case that has not yet been predicted is input, a set of similar cases, which is a set of cases similar to the prediction case, is extracted from a set of existing cases, the confidence level of a certain prediction attribute value is calculated from the set of similar cases, a reliability scale for that confidence level is calculated from the set of similar cases and the confidence level, and the confidence level of a certain prediction attribute value and the reliability scale for that confidence level are output. Furthermore, Patent Document 2 describes that an input dataset is divided according to specified division conditions, neighbor data of the input data including feature nodes that represent the characteristics of the distribution structure of each divided dataset is generated, and a score representing the relationship between the explanatory variables and the target variable is calculated based on the explanatory variables of the generated neighbor data and the target variable data obtained by inputting the neighbor data into a machine learning model. [Prior art documents] [Patent Documents]

[0003] [Patent Document 1] Japanese Patent Publication No. 2003-323601 [Patent Document 2] Japanese Patent Publication No. 2019-191895 [Overview of the project] [Problems that the invention aims to solve]

[0004] The accuracy of each prediction varies. Knowing the degree of accuracy for each prediction can help users decide which prediction to rely on for prototyping. In this regard, if the product's manufacturing conditions are similar to those learned by the machine learning model, the accuracy of the prediction output is considered high; conversely, if they are not similar, the accuracy is considered low. Therefore, it is desirable to determine the similarity of the manufacturing conditions.

[0005] However, when manufacturing conditions are represented as multidimensional vector data containing information from numerous factors, determining similarity can be computationally time-consuming.

[0006] This disclosure has been made in view of the above-mentioned problems, and its purpose is to provide a performance prediction device and program that can calculate the accuracy of predicted performance in a short amount of time. [Means for solving the problem]

[0007] The performance prediction device according to this disclosure is characterized by including: a machine learning model that is learned from a plurality of training manufacturing condition vector data representing the manufacturing conditions of a product, and outputs predictive performance data that shows the properties of a product manufactured under the manufacturing conditions represented by a given input manufacturing condition vector data when that input manufacturing condition vector data is input; dimensionality reduction means for reducing the dimensionality of the plurality of training manufacturing condition vector data and the input manufacturing condition vector data, respectively; similarity determination means for determining whether the dimensionality-reduced input manufacturing condition vector data is similar to any of the dimensionality-reduced plurality of training manufacturing condition vector data; and output means for outputting the predictive performance data and outputting information corresponding to the determination result of the similarity determination means as the accuracy of the predictive performance data. According to this, the accuracy of the predictive performance can be calculated in a short time.

[0008] The program relating to this disclosure uses a computer to function as follows: a machine learning model that learns from a plurality of training manufacturing condition vector data representing the manufacturing conditions of a product, and outputs predictive performance data indicating the properties of a product manufactured under the manufacturing conditions shown by a given input manufacturing condition vector data when that input manufacturing condition vector data is input; a dimensionality reduction means for reducing the dimensionality of the plurality of training manufacturing condition vector data and the input manufacturing condition vector data; a similarity determination means for determining whether the dimensionality-reduced input manufacturing condition vector data is similar to any of the dimensionality-reduced plurality of training manufacturing condition vector data; and an output means for outputting the predictive performance data and outputting information corresponding to the determination result of the similarity determination means as the accuracy of the predictive performance data. With this, the accuracy of the predictive performance can be calculated in a short time using a computer. [Brief explanation of the drawing]

[0009] [Figure 1] This figure shows the configuration of a performance prediction device, which is an example of an embodiment of the present disclosure. [Figure 2] This is a functional block diagram showing an example of the functions implemented in the performance prediction device. [Figure 3] This figure shows an example of training data and training-time dimensionality reduction data. [Figure 4] This figure shows an example of input data and input-time dimensionality-compressed data. [Figure 5] This figure shows an example of prediction performance data and the accuracy of that data. [Figure 6A] This figure shows an example of the processes performed when training a machine learning model. [Figure 6B] This figure shows an example of the processes performed when training a machine learning model. [Figure 7A] This figure shows an example of the processing performed when making predictions on input data. [Figure 7B] This figure shows an example of the processing performed when making predictions on input data. [Figure 8]It is a functional block diagram showing another example of the functions implemented in the performance prediction device.

Embodiments for Carrying Out the Invention

[0010] Hereinafter, embodiments of the present disclosure will be described with reference to the drawings. In this embodiment, by inputting input data indicating the manufacturing conditions of a product into a trained machine learning model, prediction performance data indicating the properties of the product manufactured under the manufacturing conditions is output, and information regarding the accuracy of the prediction performance data is also output. Hereinafter, the case where the product is a tire or tire rubber will be described, but the product is not limited to a mixture such as a tire or tire rubber produced by a plurality of raw materials and production processes, and may be any substance produced based on predetermined conditions.

[0011] [1. Hardware Configuration] FIG. 1 is a diagram showing the configuration of a performance prediction device 10 which is an example of an embodiment of the present disclosure. The performance prediction device 10 according to this embodiment is a computer such as a personal computer, a general-purpose computer, or a portable information terminal, and as shown in FIG. 1, includes a processor 11, a storage unit 12, a communication unit 13, a display unit 14, and an operation unit 15. Note that the performance prediction device 10 may include an optical disk drive for reading an optical disk, a USB (Universal Serial Bus) port, and the like.

[0012] The processor 11 is a program control device such as a CPU (Central Processing Unit) that operates according to a program installed in the performance prediction device 10, which is, for example, a computer. The storage unit 12 is a storage element such as a ROM (Read Only Memory) or a RAM (Random Access Memory), or a hard disk drive. Data such as programs executed by the processor 11 is stored in the storage unit 12. The communication unit 13 is a communication interface such as a network board. The display unit 14 is a display device such as a liquid crystal display, and displays various images according to the instructions of the processor 11. The operation unit 15 is a user interface such as a keyboard or a mouse, receives user operation inputs, and outputs a signal indicating the content to the processor 11.

[0013] [[ID=##4]][2. Functional Blocks] The performance prediction device 10 outputs prediction performance data indicating, for example, the performance predicted for a tire or the rubber of a tire (e.g., the rolling resistance of the tire or the hardness of the rubber), and also outputs information regarding the accuracy of the prediction performance data. Hereinafter, the output of the prediction performance data and the information regarding the accuracy by the performance prediction device 10 will be described.

[0014] FIG. 2 is a functional block diagram showing an example of the functions implemented in the performance prediction device 10. As shown in FIG. 2, the performance prediction device 10 functionally includes a learning data acquisition unit 20, an input data acquisition unit 30, a machine learning model 40, a dimensionality reduction unit 50, a dimensionality reduction data storage unit 60, a similarity determination unit 70, and an output unit 80. Note that in the performance prediction device 10, not all of the functions shown in FIG. 2 need to be implemented, and functions other than those shown in FIG. 2 may be implemented.

[0015] [2-1. Learning Data Acquisition Unit] Note: In the translation of , "機能ブロック" is translated as "Functional Blocks" to be more in line with the common expression in English technical texts. If a more literal translation like "Function Blocks" is preferred, it can be adjusted accordingly.The training data acquisition unit 20 acquires training manufacturing condition vector data. "Training manufacturing condition vector data" refers to training manufacturing condition data that indicates the manufacturing conditions of a product, vectorized using a given calculation formula, and is the training data used to train the machine learning model 40 described later. The training data acquisition unit 20 is not limited to this; it may also acquire data converted from the training manufacturing condition data using a given conversion method other than vectorization as training data, or it may acquire the training manufacturing condition data itself. Hereinafter, the data acquired by the training data acquisition unit 20 (in this embodiment, training manufacturing condition vector data) will also be simply referred to as "training data". The training data acquisition unit 20 may be mainly implemented by the processor 11, but is not limited to this; it may also be implemented by the processor of another information processing device connected to the performance prediction device 10 via a network.

[0016] Figure 3 shows an example of training data and training-time dimensionality reduction data. In the example shown in Figure 3, each piece of data identified by an individual ID (1, 2, N, etc.) (one row of data in the table) corresponds to one piece of data.

[0017] As shown in Figure 3, the training data 110 contains multiple factors. More specifically, factors X1 to X M It contains M factors. Furthermore, the training data 110 may include, for example, one or more factors indicating the tire material, and one or more factors indicating the tire structure. In addition, the training data 110 may include one or more factors indicating the tire rubber material, and one or more factors indicating the tire rubber manufacturing conditions.

[0018] Furthermore, the training data 110 includes predictive performance data that shows the performance (properties) of products manufactured under the manufacturing conditions shown therein. Predictive performance data is, for example, data that shows the performance of a tire (such as the rolling resistance of a tire). Alternatively, for example, the predictive performance data may also be data that shows the physical properties of the tire's rubber (such as its hardness). In the example shown in Figure 3, the predictive performance data is shown as a predicted value and is expressed numerically. For example, the training data 110 identified by ID "1" has a predicted value of "15.0", and the training data 110 identified by ID "2" has a predicted value of "14.8".

[0019] Although not shown in the diagram, each of the multiple training data 110 contains multiple factors (factors X1~X M The values ​​differ among the training data. Therefore, the predicted values ​​for each training data are basically different, but due to rounding processes such as truncating the fractional part of the predicted value, the predicted values ​​may sometimes match across multiple training data.

[0020] [2-2. Input Data Acquisition Section] The input data acquisition unit 30 acquires given input manufacturing condition vector data. "Input manufacturing condition vector data" is data that vectorizes the input manufacturing condition data, which indicates the manufacturing conditions of a product, using a given calculation formula, similar to the training manufacturing condition vector data. However, the input data acquisition unit 30 may acquire data converted from the input manufacturing condition data using a given conversion method other than vectorization, or it may acquire the input manufacturing condition data itself. For example, if the training manufacturing condition data itself is used as training data for the machine learning model 40 described later, the input data acquisition unit 30 may acquire the input manufacturing condition data in its unconverted state. Hereinafter, the data acquired by the input data acquisition unit 30 (in this embodiment, input manufacturing condition vector data) will also be simply referred to as "input data". The input data acquisition unit 30 may be mainly implemented by the processor 11, but is not limited to this, and may be implemented by the processor of another information processing device connected to the performance prediction device 10 via a network.

[0021] Figure 4 shows an example of input data and input time-dimensionally compressed data. In the example shown in Figure 4, as in Figure 3, each piece of data identified by an individual ID (1, 2, n, etc.) (one row of data in the table) corresponds to one piece of data. In the example shown in Figure 4, there are multiple (n) pieces of input data, but the number of input data acquired by the input data acquisition unit 30 and input into the machine learning model 40, which will be described later, may be just one.

[0022] The input data 210, like the training data 110, contains multiple factors. More specifically, factors X1 to X M It contains M factors. The input data 210, like the training data 110, may contain one or more factors indicating the tire material and one or more factors indicating the tire structure. In addition, the input data 210 may contain one or more factors indicating the tire rubber material and one or more factors indicating the tire rubber manufacturing conditions.

[0023] In this embodiment, the number of factors included in each of the multiple training data 110 and the input data 210 is M, and they are set to be the same number. However, it is sufficient that the multiple training data 110 and the input data 210 include multiple predetermined factors common to each other, and the total number of factors included in each does not necessarily have to be the same.

[0024] The input data 210 is intended to predict the properties of products manufactured under the various manufacturing conditions shown therein. Therefore, unlike the training data 110, the input data 210 does not include prediction performance data such as predicted values. By creating input data representing various manufacturing conditions and inputting it into the machine learning model 40 (more specifically, the prediction unit 43), the user can obtain information about the properties predicted by the machine learning model 40 (more specifically, the prediction unit 43).

[0025] [2-3. Machine Learning Models] The machine learning model 40 is a model that is trained using multiple training data 110 (training manufacturing condition vector data) that each represent the manufacturing conditions of a product. When given input data 210 (input manufacturing condition vector data) is input, the model outputs predictive performance data that represents the properties of a product manufactured under the manufacturing conditions indicated by this input manufacturing condition vector data. The performance prediction device 10 may have multiple machine learning models 40 as a function, each corresponding to a type of predictive performance data to be predicted (for example, a type of property such as tire rolling resistance or rubber hardness). In this case, machine learning may be performed by inputting training data 110 containing predictive performance data of the type corresponding to this machine learning model 40. Alternatively, the user may input the input data 210 to the machine learning model 40 that corresponds to the type of predictive performance data to be predicted from among the multiple machine learning models 40.

[0026] As shown in Figure 2, the machine learning model 40 includes a parameter storage unit 41, a learning unit 42, and a prediction unit 43. The machine learning model 40 is a model that is trained using a multi-layer neural network, such as a DNN (Deep Neural Network) or a CNN (Convolutional Neural Network). Alternatively, the machine learning model 40 may be a model that is trained using a given statistical method or linear regression.

[0027] [2-3-1. Parameter Storage Unit] The parameter storage unit 41 stores the parameters of the machine learning model 40. More specifically, the parameter storage unit 41 stores the parameters of the machine learning model 40 that output predictive performance data when input data 210 is input to the machine learning model 40. The parameter storage unit 41 may be primarily implemented as the storage unit 12 of the performance prediction device 10, or it may be implemented as an external storage device connected to the performance prediction device 10 by wire or wireless, or as another storage device such as a NAS (Network Attached Storage).

[0028] If the machine learning model 40 is implemented using a multi-layer neural network, the parameter storage unit 41 may store as parameters the machine learning model 40 the multiple nodes that make up the neural network, the weights of each node, the number of layers, and the number of nodes used in each layer. Alternatively, the parameter storage unit 41 may also store as parameters the machine learning model 40 calculation formulas and coefficients of the formulas used to obtain predictive performance data from multiple manufacturing conditions.

[0029] [2-3-2. Learning Department] The learning unit 42 updates the parameters of the machine learning model 40 stored in the parameter storage unit 41 by performing machine learning using multiple training data 110 acquired by the training data acquisition unit 20. In other words, the parameters of the machine learning model 40 stored in the parameter storage unit 41 are learned (updated) using multiple training data. The learning unit 42 may be primarily implemented by the processor 11 of the performance prediction device 10, or it may be implemented by the processor of another information processing device connected to the performance prediction device 10 via a network.

[0030] [2-3-3. Prediction Section] The prediction unit 43 calculates predictive performance data for the input data 210 when the machine learning model 40 receives the input data 210. The prediction unit 43 may be primarily implemented by the processor 11 of the performance prediction device 10, or it may be implemented by the processor of another information processing device connected to the performance prediction device 10 via a network.

[0031] Figure 5 shows an example of prediction performance data and the accuracy of the prediction performance data. In the example shown in Figure 5, the predicted value 230 for input data 210 identified by ID "1" is "15.0", and the predicted value 230 for input data 210 identified by ID "2" is "13.2". Thus, the predicted value 230 for each of the multiple input data 210 identified by individual IDs corresponds to an example of prediction performance data. That is, the prediction unit 43 calculates the predicted value 230 for each input data 210. Note that the prediction performance data is not limited to numerical values ​​such as the predicted value 230, but can be anything that can identify the properties of the product. The prediction performance data may be represented, for example, as a sequence of numbers, or as symbols (circle, cross, etc.) or letters (hard, soft, etc.).

[0032] [2-4. Dimensional Compression Section] The dimensionality reduction unit 50 reduces the dimensionality of each of the multiple training data 110 (training manufacturing condition vector data) and input data 210 (input manufacturing condition vector data). "Dimensionality reduction" means setting and calculating a new method for representing certain data from the factors of that data, thereby reducing the number of factors. The dimensionality reduction unit 50 may reduce the dimensionality of the multiple training data 110 and input data 210 using a given method such as principal component analysis (PCA), independent component analysis (ICA), or t-distributed stochastic neighbor embedding (t-SNE). The dimensionality reduction unit 50 may be mainly implemented by the processor 11 of the performance prediction device 10, or it may be implemented by the processor of another information processing device connected to the performance prediction device 10 via a network.

[0033] The dimensionality reduction unit 50 processes the individual factors X1~X contained in each of the multiple training data 110, as shown in Figure 3, for example. M From new factors A1~A mBy setting and calculating these parameters, dimensionality-reduced training manufacturing condition vector data 120 (hereinafter also referred to as training-time dimensionality-reduced data 120) is generated. As a result, the number of factors m in the training-time dimensionality-reduced data 120 becomes smaller than the number of factors M in the training data 110 before dimensionality reduction.

[0034] Furthermore, the dimensionality reduction unit 50, as shown in Figure 4, for example, measures the individual factors X1~X included in the input data 210. M From new factors A1~A m By setting and calculating these parameters, dimensionality-reduced input manufacturing condition vector data 220 (hereinafter also referred to as input dimensionality-reduced data 220) is generated. As a result, the number of factors m in the input dimensionality-reduced data 220 is smaller than the number of factors M in the input data 210 before dimensionality reduction.

[0035] In the examples shown in Figures 3 and 4, the number of factors in each of the multiple training-time dimensionality-reduced data sets 120 and the input-time dimensionality-reduced data set 220 is m, and this number is assumed to be the same. However, the total number of factors in each set does not necessarily have to be the same.

[0036] Furthermore, if the machine learning model 40 is implemented using a multi-layer neural network such as a Deep Neural Network (DNN), the dimensionality reduction unit 50 may input each of the multiple training data 110 into the machine learning model 40, thereby obtaining the output of the intermediate layer of the machine learning model 40 as multiple training-time dimensionality-reduced data 120 (dimensionality-reduced training manufacturing condition vector data). Similarly, the dimensionality reduction unit 50 may input the input data 210 into the machine learning model 40, thereby obtaining the output of the intermediate layer of the machine learning model 40 as input-time dimensionality-reduced data 220 (dimensionality-reduced input manufacturing condition vector data). By utilizing the intermediate layer of the machine learning model 40 in this way, it is possible to obtain multiple training-time dimensionality-reduced data 120 and input-time dimensionality-reduced data 220 without performing calculations for dimensionality reduction such as principal component analysis (PCA).

[0037] [2-5. Dimensional Compression Data Storage Unit] The dimensionality reduction data storage unit 60 stores multiple training-time dimensionality reduction data 120. The dimensionality reduction data storage unit 60 may store the training-time dimensionality reduction data 120 in association with the type of prediction performance data included in the training-time dimensionality reduction data 120 (for example, the type of properties such as tire rolling resistance or rubber hardness). The dimensionality reduction data storage unit 60 may be mainly implemented as the storage unit 12 of the performance prediction device 10, or it may be implemented as an external storage device connected to the performance prediction device 10 by wire or wireless, or as another storage device such as a NAS (Network Attached Storage).

[0038] [2-6. Similarity determination section, output section] The similarity determination unit 70 determines whether the input dimensionality-compressed data 220 (dimensionality-compressed input manufacturing condition vector data) is similar to any of the multiple training dimensionality-compressed data 120 (dimensionality-compressed training manufacturing condition vector data). The output unit 80 then outputs the prediction performance data calculated by the prediction unit 43 of the machine learning model 40, and also outputs information corresponding to the determination result of the similarity determination unit 70 as the accuracy of this prediction performance data. The similarity determination unit 70 and the output unit 80 may be mainly implemented by the processor 11 of the performance prediction device 10, or they may be implemented by the processor of another information processing device connected to the performance prediction device 10 via a network. The output unit 80 may also output the prediction performance data and accuracy information by displaying the prediction performance data and the accuracy information of the prediction performance data on the display unit 14 of the performance prediction device 10 or other display devices.

[0039] The similarity determination unit 70, for example, estimates the density of the distribution of multiple training-time dimensionally compressed data 120, and determines the similarity between the multiple training-time dimensionally compressed data 120 and the input-time dimensionally compressed data 220 based on the results of this density estimation. The similarity determination unit 70 may perform density estimation at points (locations) specified by the input-time dimensionally compressed data 220 within the distribution range of the multiple training-time dimensionally compressed data 120, and determine that the similarity is high (similar) if the estimated density is high, and low if the density is low (not similar). For such density estimation, for example, kernel density estimation (KDE) can be used. Alternatively, for example, the similarity determination unit 70 may determine the similarity between the multiple training-time dimensionally compressed data 120 and the input-time dimensionally compressed data 220 by utilizing the variability of the predicted estimation data based on Gaussian process regression (GP).

[0040] In addition, the similarity determination unit 70 may determine the similarity between multiple training-time dimensionally compressed data 120 and input-time dimensionally compressed data 220 by calculating the distance between each of the multiple training-time dimensionally compressed data 120 stored in the dimensionally compressed data storage unit 60 and the input-time dimensionally compressed data 220. For example, the similarity determination unit 70 determines the similarity from the distance between each of the multiple training-time dimensionally compressed data 120 identified by IDs "1", "2", ..., "N" in Figure 3 and the input-time dimensionally compressed data 220 identified by ID "1" in Figure 4. The similarity determination unit 70 may determine that the similarity is high (similar) if the calculated distance is close, and that the similarity is low (not similar) if the distance is long.

[0041] In this case, multiple factors A1 to A are included in common in the training-time dimensionality-reduced data 120 and the input-time dimensionality-reduced data 220. mSimilarity may be determined based on the respective distances. The similarity determination unit 70 may determine that a factor (for example, factor A1) is similar if the distance between the training-time dimensionality-compressed data 120 and the input-time dimensionality-compressed data 220 is less than or equal to a threshold, and may determine that the factor is not similar if the distance exceeds the threshold. The similarity determination unit 70 may also determine the similarity of each factor using three or more criteria such as identical, similar, and mismatched, or it may calculate a similarity score according to the distance.

[0042] The output unit 80 outputs a predicted value 230 as prediction performance data, as shown in Figure 5, for example, and also outputs an accuracy value 240 as a value indicating the accuracy of the prediction performance data. In the example shown in Figure 5, the accuracy value 240 is "0.82" for input data 210 identified by ID "1", and the accuracy value 240 is "0.77" for input data 210 identified by ID "2". Note that the information indicating the accuracy of the prediction performance data is not limited to numerical values ​​such as the accuracy value 240, but may also be indicated by symbols (circle, cross, etc.) or letters (high, low).

[0043] Furthermore, if the similarity determination unit 70 determines similarity by calculating the distance between each of the multiple training-time dimensionality-reduced data 120 and the input-time dimensionality-reduced data 220, the output unit 80 may output information based on the similarity with the training-time dimensionality-reduced data 120 that is closest in distance to the input-time dimensionality-reduced data 220 (i.e., the training-time dimensionality-reduced data 120 that is determined to be the most similar) as the accuracy of the prediction performance data.

[0044] [3. Processing Flow] Figures 6A, 6B, 7A, and 7B show examples of the flow of prediction processing performed by the performance prediction device 10. Figures 6A and 6B show examples of processing performed when training the machine learning model 40, and Figures 7A and 7B show examples of processing performed when making predictions on the input data 210.

[0045] [3-1. Processing during training] The processes performed during the training of the machine learning model 40 will be explained with reference to Figures 6A and 6B. First, the training data acquisition unit 20 acquires training data 110, which is manufacturing condition vector data used for training the machine learning model 40 (step S101). Next, the training unit 42 of the machine learning model 40 trains the machine learning model 40 by inputting the training data 110 acquired in step S101 into the machine learning model 40 (step S102). The training unit 42 trains the machine learning model 40 by updating the parameters of the machine learning model 40 stored in the parameter storage unit 41 so that the prediction unit 43 of the machine learning model 40 outputs prediction performance data (for example, the predicted value 230 shown in Figure 3) contained in the training data 110 in response to the input of the training data 110. In step S102, the machine learning model 40 receives training data 110 containing prediction performance data of the corresponding type.

[0046] Next, the dimensionality reduction unit 50 acquires the dimensionality-reduced training data 120, which is obtained from the training data 110 acquired in step S101, and stores it in the dimensionality-reduced data storage unit 60 (step S103), and terminates the process. In step S103, the dimensionality reduction unit 50 may acquire the dimensionality-reduced training data 120 by reducing the dimensionality of the training data 110 using a given method such as principal component analysis (PCA), or it may acquire the data output from the intermediate layer of the machine learning model 40 as the dimensionality-reduced training data 120 through the processing in step S102.

[0047] Furthermore, in step S103, the dimensionality reduction unit 50 may store the training-time dimensionality reduction data 120 in the dimensionality reduction data storage unit 60, associating it with the type of predictive performance data it contains (for example, the type of properties such as tire rolling resistance or rubber hardness). Through the processing of steps S101 to S103 described above, the dimensionality reduction data storage unit 60 stores multiple training-time dimensionality reduction data 120, each associated with the type of predictive performance data, each time the machine learning model 40 is trained.

[0048] [3-2. Processing during prediction] The process of making predictions on input data 210 using the machine learning model 40 will be explained with reference to Figures 7A and 7B. First, the input data acquisition unit 30 acquires input data 210, which is manufacturing condition vector data that the user wishes to predict (step S201). The input data acquisition unit 30 acquires input data 210 specified by the user, for example. The number of input data 210 acquired in step S101 may be multiple, as shown in Figure 4, or it may be just one.

[0049] Next, the prediction unit 43 of the machine learning model 40 calculates, for example, the predicted value 230 shown in Figure 5 as prediction performance data for the input data 210 acquired in step S201, based on the machine learning model 40 (step S202). The prediction unit 43 calculates prediction performance data for the input data 210 by inputting the input data 210 into the machine learning model 40. If the performance prediction device 10 has multiple machine learning models 40 as a function, each corresponding to a type of prediction performance data, the prediction unit 43 may, for example, select a machine learning model 40 according to the user's instructions (a machine learning model 40 that outputs prediction performance data of the type that the user wants to predict) and input the input data 210 to it.

[0050] Next, the dimensionality reduction unit 50 obtains input dimensionality-reduced data 220 from the input data 210 acquired in step S201 (step S203). In step S203, the dimensionality reduction unit 50 may obtain input dimensionality-reduced data 220 by reducing the dimensionality of the input data 210 using a given method such as principal component analysis (PCA), or it may obtain the data output from the intermediate layer of the machine learning model 40 as input dimensionality-reduced data 220 through the processing in step S202.

[0051] Next, the similarity determination unit 70 determines the similarity between the input dimensionality-reduced data 220 acquired in step S203 and the multiple training dimensionality-reduced data 120 stored in the dimensionality-reduced data storage unit 60 in step S103 (step S204). In step S204, the similarity determination unit 70 may determine the similarity between the input dimensionality-reduced data 220 and the multiple training dimensionality-reduced data 120 stored in the dimensionality-reduced data storage unit 60, which are associated with the type of prediction performance data calculated in step S202 (the type corresponding to the machine learning model 40 used for the calculation).

[0052] In step S204, the similarity determination unit 70 may determine the similarity between the multiple training-time dimensionally compressed data 120 and the input-time dimensionally compressed data 220 by performing density estimation of the distribution of the multiple training-time dimensionally compressed data 120 using methods such as kernel density estimation (KDE). Alternatively, the similarity determination unit 70 may determine the similarity between the multiple training-time dimensionally compressed data 120 and the input-time dimensionally compressed data 220 by utilizing the variability of the predicted estimation data based on methods such as Gaussian process regression (GP). In addition, the similarity between the input-time dimensionally compressed data 220 and the multiple training-time dimensionally compressed data 120 may be determined by calculating the distance between each of the multiple training-time dimensionally compressed data 120 and the input-time dimensionally compressed data 220.

[0053] Finally, the output unit 80 outputs, for example, the accuracy value 240 shown in Figure 5, as the accuracy of the predicted value 230 calculated in step S202 and the similarity determination result determined in step S204, and terminates the process. Note that in step S204, if the similarity between the input dimensionally compressed data 220 and the multiple training dimensionally compressed data 120 is determined by calculating the distance between each of the multiple training dimensionally compressed data 120 and the input dimensionally compressed data 220, the output unit 80 may output information based on the similarity with the training dimensionally compressed data 120 that is closest in distance to the input dimensionally compressed data 220 (i.e., the training dimensionally compressed data 120 that is determined to be the most similar) as the accuracy value 240.

[0054] [4. Summary] The probability of property prediction based on the plurality of factors X1 to X included in the input data 210 M varies for each input data 210 (that is, for each prediction). In this regard, according to the similarity determination unit 70 and the output unit 80, the input-time dimensionally compressed data 220, which is the dimensionally compressed input data 210, and the plurality of learning-time dimensionally compressed data 120, which are also the dimensionally compressed plurality of learning data 110, are output as the accuracy of the prediction performance data. Therefore, the user can know the probability of the prediction performance data for each prediction.

[0055] Furthermore, in the present embodiment, since the similarity determination unit 70 determines the similarity or dissimilarity between the input-time dimensionally compressed data 220 dimensionally compressed from the input data 210 and the plurality of learning-time dimensionally compressed data 120 dimensionally compressed from the plurality of learning data 110 respectively, when determining the similarity or dissimilarity between the input data 210, which is vector data including information on a large number of factors (factors X1 to X M ), and the plurality of learning data 110, the similarity or dissimilarity can be determined in a shorter time. That is, the accuracy of the prediction performance can be calculated in a shorter time compared to the case of directly comparing the input data 210 and the plurality of learning data 110.

[0056] [5. Variation] The present invention is not limited to the above embodiments.

[0057] In this embodiment, the dimensionality reduction unit 50 obtains input dimensionality-reduced data 220 obtained by dimensionality reduction from input data 210, and a plurality of training dimensionality-reduced data 120 obtained by dimensionality reduction from a plurality of training data 110, and the similarity determination unit 70 determines the similarity between the plurality of training dimensionality-reduced data 120 and the input dimensionality-reduced data 220. However, the similarity determination unit 70 may also determine the similarity between data from which predetermined factors have been removed from a plurality of training data 110 and data from which predetermined factors have been removed from input data 210. The output unit 80 may output information corresponding to the similarity determined in this way as the accuracy of the prediction performance data predicted by the machine learning model 40 (more specifically, the prediction unit 43).

[0058] Figure 8 is a functional block diagram showing another example of the functions implemented in the performance prediction device 10. As shown in Figure 8, the performance prediction device 10 may not include a dimensionality reduction unit 50 and a dimensionality reduction data storage unit 60 as functions, but instead may include a factor deletion unit 90 and a factor deletion data storage unit 100.

[0059] The factor removal unit 90 removes predetermined factors from each of the multiple training data 110 (training manufacturing condition vector data) and input data 210 (input manufacturing condition vector data). The factor removal unit 90 may be primarily implemented by the processor 11 of the performance prediction device 10, or it may be implemented by the processor of another information processing device connected to the performance prediction device 10 via a network.

[0060] The factor removal unit 90 removes multiple factors X1~X included in the training data 110 acquired by the training data acquisition unit 20. M (See Figure 3) The number of factors included in the training data 110 is reduced by deleting predetermined factors. Similarly, the factor deletion unit 90 deletes multiple factors X1~X included in the input data 210 acquired by the input data acquisition unit 30. MThe number of factors included in the training data 110 is reduced by deleting predetermined factors. It is desirable that the factor deletion unit 90 deletes factors that have a small impact on the calculation of prediction performance data by the prediction unit 43 of the machine learning model 40. The factor deletion unit 90 may calculate the degree of impact on the calculation of prediction performance data by performing analysis of variance or the like, and determine which factors to delete based on this impact.

[0061] The factor-removed data storage unit 100 stores multiple training data sets 110 from which predetermined factors have been removed by the factor-removed unit 90. The dimensionality-reducing data storage unit 60 may store the training data sets 110 from which predetermined factors have been removed, in association with the type of prediction performance data included in the training data sets 110 (for example, the type of properties such as tire rolling resistance or rubber hardness). The dimensionality-reducing data storage unit 60 may be primarily implemented as the storage unit 12 of the performance prediction device 10, or it may be implemented as an external storage device connected to the performance prediction device 10 by wire or wireless, or as another storage device such as a NAS (Network Attached Storage).

[0062] The similarity determination unit 70 determines the similarity between the input data 210 from which predetermined factors have been removed by the factor removal unit 90 and the training data 110 stored in the factor removal data storage unit 100, from which predetermined factors have been removed by the factor removal unit 90. The similarity determination unit 70 may determine the similarity between the input data 210 and the training data 110 stored in the factor removal data storage unit 100 that are associated with the type corresponding to the machine learning model 40 used for prediction (calculation) of prediction performance data.

[0063] The output unit 80 outputs information corresponding to the similarity determined as described above as the accuracy of the prediction performance data. In this way, a large number of factors (factors X1~X) can be used. M Compared to directly determining the similarity between input data 210 containing information about the subject and multiple training data 110, this method allows for similarity determination in a shorter time. In other words, the accuracy of prediction performance can be calculated in a shorter time. [Explanation of symbols]

[0064] 10 Performance prediction device, 11 Processor, 12 Memory unit, 13 Communication unit, 14 Display unit, 15 Operation unit, 20 Training data acquisition unit, 30 Input data acquisition unit, 40 Machine learning model, 41 Parameter storage unit, 42 Learning unit, 43 Prediction unit, 50 Dimensional compression unit, 60 Dimensional compression data storage unit, 70 Similarity determination unit, 80 Output unit, 90 Factor removal unit, 100 Factor removal data storage unit, 110 Training data, 120 Dimensional compression data during training, 210 Input data, 220 Dimensional compression data during input, 230 Predicted value, 240 Accuracy value.

Claims

1. A machine learning model that is trained using multiple training manufacturing condition vector data, each of which vectorizes multiple training manufacturing condition data representing the manufacturing conditions of a product, each of which includes multiple factors representing the manufacturing conditions of the product, and multiple performance data representing the properties of the product manufactured by the manufacturing conditions indicated by the multiple factors included in each of the multiple training manufacturing condition vector data, and which outputs predictive performance data representing the properties of the product manufactured by the manufacturing conditions when input manufacturing condition vector data including multiple factors representing the manufacturing conditions is input, A dimensionality reduction means that inputs each of the plurality of training manufacturing condition vector data into the machine learning model and obtains the output from the intermediate layer of the machine learning model to reduce the dimensionality of each of the plurality of training manufacturing condition vector data, and inputs the input manufacturing condition vector data into the machine learning model and obtains the output from the intermediate layer of the machine learning model to reduce the dimensionality of the input manufacturing condition vector data, Similarity determination means for determining the similarity between the dimensionality-reduced input manufacturing condition vector data and each of the dimensionality-reduced plurality of training manufacturing condition vector data, Output means for outputting the aforementioned prediction performance data and outputting information indicating the degree of similarity between the dimensionality-reduced input manufacturing condition vector data and any of the dimensionality-reduced plurality of training manufacturing condition vector data as the accuracy of the prediction performance data, A performance prediction device characterized by including [a certain feature].

2. In the performance prediction device according to claim 1, The aforementioned product is a tire. The plurality of learning manufacturing condition vector data and the input manufacturing condition vector data include one or more factors representing the material of the tire and one or more factors representing the structure of the tire. The aforementioned predicted performance data indicates the performance of the tire. A performance prediction device characterized by the following features.

3. In the performance prediction device according to claim 1 or 2, The aforementioned product is tire rubber, The plurality of learning manufacturing condition vector data and the input manufacturing condition vector data each include one or more factors representing the rubber material and one or more factors representing the rubber manufacturing conditions. The aforementioned predictive performance data indicates the physical properties of the rubber. A performance prediction device characterized by the following features.

4. A machine learning model that is trained using multiple training manufacturing condition vector data, each of which is a vector of multiple training manufacturing condition data representing the manufacturing conditions of a product, each of which includes multiple factors representing the manufacturing conditions of the product, and multiple performance data representing the properties of the product manufactured by the manufacturing conditions indicated by the multiple factors included in each of the multiple training manufacturing condition vector data, and input manufacturing condition vector data, which is a vector of input manufacturing condition data representing the manufacturing conditions of the product, and outputs predictive performance data representing the properties of the product manufactured by the manufacturing conditions when input manufacturing condition vector data including multiple factors representing the manufacturing conditions is input. Dimensionality reduction means that inputs each of the plurality of training manufacturing condition vector data into the machine learning model and obtains the output from the intermediate layer of the machine learning model to reduce the dimensionality of each of the plurality of training manufacturing condition vector data, inputs the input manufacturing condition vector data into the machine learning model and obtains the output from the intermediate layer of the machine learning model to reduce the dimensionality of the input manufacturing condition vector data, Similarity determination means for determining the similarity between the dimensionality-reduced input manufacturing condition vector data and each of the dimensionality-reduced plurality of training manufacturing condition vector data, and Output means that outputs the aforementioned prediction performance data and outputs information indicating the degree of similarity between the dimensionality-reduced input manufacturing condition vector data and any of the dimensionality-reduced plurality of training manufacturing condition vector data, A program that makes a computer function.

Images