Computer-based machine learning processing method and prediction method for product performance, and computer-based method for proposing raw material blends

Abstract

To enable machine-learning the performances of a product composed of a mixture of raw materials while tracking change in raw materials without waiting for accumulation of new learning data for changed raw materials, so as to obtain learning achievements, even in a case where the raw materials has been changed.SOLUTION: For a product made by mixing and processing raw materials, when machine learning is used to determine performances of a product determined by contents of the mixing or determine the contents of the mixing that satisfies the performances of the product, when the raw materials used in the product are changed, the contents of the mixing of past learning data are converted and re-learned based on a correspondence table 132 between the changed raw materials and the raw materials used for learning before the change, thereby making it possible to evaluate the performances of the product using the changed raw materials and perform machine learning without the need to accumulate such data.SELECTED DRAWING: Figure 1

Description

[Technical Field]

[0001] This invention improves product performance Computer-based Machine learning processing method and prediction method and raw materials Computer-based Mixture presentation method The present invention relates to a method for manufacturing processed foods in which ingredients are frequently changed, and is particularly suitable for the manufacture of processed foods in which ingredients are frequently changed. ... Computer-based Machine learning processing method and prediction method and raw materials Computer-based Mixture presentation method Regarding. [Background technology]

[0002] The performance of products made up of multiple ingredients By computer Machine learning techniques are described in Patent Documents 1 and 2.

[0003] The machine learning uses data in which the composition of raw materials is associated with their performance. To obtain meaningful results through machine learning, a sufficient number of data items are required for each raw material, each of which corresponds to its composition and performance. [Prior art documents] [Patent documents]

[0004] [Patent Document 1] Japanese Patent Publication No. 2022-80491 [Patent Document 2] Japanese Patent Application Publication No. 2018-191073 Summary of the Invention [Problem to be solved by the invention]

[0005] Conventionally, when it comes to raw materials and their procurement, the constituent raw materials of the same product may be changed while maintaining performance due to availability constraints, costs, and seasonal fluctuations. For this reason, when the composition of the relevant raw materials is the input for machine learning, it is necessary to accumulate new learning data using the changed raw materials, and the learning results using the changed raw materials cannot be used until sufficient new data for learning is collected.

[0006] The present invention has been made to solve the above-mentioned conventional problems, and aims to enable machine learning to follow changes in raw materials and obtain learning results, even when the raw materials are changed, without waiting for new data to be accumulated for the changed raw materials. [Means for solving the problem]

[0007] The present invention relates to a product made by blending and processing raw materials, and the performance of the product is determined by the blend or the blend satisfies the performance of the product. Computer-based When using machine learning to find the ingredients, if a change is made to the ingredients used in the product, a correspondence table between the ingredients after the change and the ingredients used for learning before the change is created. data Based on , stored in a storage device Combining past learning data Using a computer By converting and relearning, it is possible to evaluate the performance of products using changed raw materials without the need to accumulate that data. Computer-based By enabling machine learning, the above-mentioned conventional problems are resolved.

[0008] Here, the The correspondence table data is Ingredients before and after the change of , all of which are a combination or blend of multiple ingredients data It can be said that:

[0009] Also, the manufacturing method used to manufacture the product Data Formulated as product composition information data It can be studied in conjunction with the above.

[0010] Also, the combination of past learning data data Conversion of data This can be limited to certain formulations where this is effective.

[0011] Also, the applicable correspondence table data If the product performance evaluation changes depending on the data Parameter information that changes performance according to the training data can be stored, and the performance evaluation of past training data can be corrected and used for re-training.

[0012] The present invention also provides the above-mentioned product performance Computer-based Machine learning processing method Using , the formulation used data from The computer Product performance prediction Computer-based prediction method This provides:

[0013] The present invention also provides the above-mentioned product performance Computer-based Using machine learning processing methods, Specified Product performance satisfy Ingredients data of The computer of raw materials characterized by presenting Computer-based Mixture presentation method This provides: [Effects of the Invention]

[0014] According to the present invention, when the raw materials are changed in the same or equivalent product, a correspondence table of the raw materials before and after the change is created. data By using this, all the compounding information in the data set used for learning is converted and re-learned, and learning results can be obtained while tracking changes in ingredients without accumulating new data for the changed ingredients. Also, even when changes are made to the combination of multiple ingredients, a correspondence table between single or multiple ingredient combinations can be created. data The same effect can be obtained by preparing and converting [Brief explanation of the drawings]

[0015] [Figure 1]FIG. 1 is a block diagram showing the overall configuration of an embodiment of the present invention. [Figure 2] FIG. 1 is a block diagram showing an example of the configuration of a computer used in the embodiment. [Figure 3] 1 is a flowchart showing a processing procedure according to the embodiment; [Figure 4] FIG. 10 is a diagram showing an example of a correspondence table in Example 1 in which the raw materials after the change are a combination of multiple raw materials. [Figure 5] FIG. 10 is a diagram showing an example of correspondence table data in Example 2, in which data on a manufacturing method used in manufacturing a product is used as configuration information of product data together with recipe data as a learning target. [Figure 6] FIG. 10 is a diagram showing an example of correspondence table data in Example 3, in which conversion of combination data in past learning data is limited to some combinations for which the correspondence table data is valid. [Figure 7] FIG. 10 shows an example of a fourth embodiment in which parameter information for changing performance in accordance with correspondence table data is stored, and performance evaluation of past learning data is corrected and used for re-learning. DETAILED DESCRIPTION OF THE INVENTION

[0016] Hereinafter, embodiments of the present invention will be described in detail with reference to the drawings. Note that the present invention is not limited to the contents described in the following embodiments and examples. Furthermore, the constituent elements in the embodiments and examples described below include those that can be easily imagined by a person skilled in the art, those that are substantially the same, and those that are within the so-called equivalent range. Furthermore, the constituent elements disclosed in the embodiments and examples described below may be appropriately combined or appropriately selected for use.

[0017] In the embodiment of the present invention, as shown in FIG. 1, the old recipe table, which is the learning data before the raw material change, is used. Data (also simply called the old recipe table) 112 is stored in an old formula storage unit 110 (for example, a database), an old performance table storage unit 120 is stored in an old performance table 122 before the raw material change, and a correspondence table of raw materials before and after the change. Data (also simply called a correspondence table) A correspondence table storage unit 130 stores the old recipe 112 and the new recipe 132 using the correspondence table 132. Data (also simply called the new recipe table)a data conversion device 100 having a correspondence table conversion unit 140 that replaces the old performance table 122 with the new recipe table 152 and stores it in a new performance table storage unit 150, and corrects the old performance table 122 to obtain a new performance table 162 and stores it in a new performance table storage unit 160; a machine learning device 200 that performs machine learning using the new recipe table 152 and the new performance table 162 stored in the new recipe table storage unit 150 and the new performance table storage unit 160 of the data conversion device 100; a learning result storage unit 210 that saves the learning results of the machine learning device 200; a performance prediction device 300 that predicts product performance (for example, taste and texture in the case of processed foods) using the re-learning results of the learning result storage unit 210 and a new blend to obtain a performance prediction table 310; and a formulation optimization device 400 that uses the learning results stored in the learning result storage unit 210 to create a formulation optimization table 410 that satisfies required performance.

[0018] The correspondence table conversion unit 140, the machine learning device 200, the performance prediction device 300, and the blend optimization device 400 can be configured using, for example, a computer 500 as shown in Fig. 2. The computer 500 is configured to include, for example, a central processing unit (CPU) 510 that performs various arithmetic processing, an input interface (input I / O) 520 for inputting data, a read only memory (ROM) 530 that stores computer programs for processing data, a random access memory (RAM) 540 for temporarily storing data during calculation, an output interface (output I / O) 550 for outputting data, and a system bus 560 that connects the components 510 to 550.

[0019] The above blending includes the manufacturing method as necessary. Also, the ratio of the old and new ingredients is not necessarily 1:1.

[0020] The processing procedure of the embodiment will be described below with reference to FIG.

[0021] First, in step 1000, the new and old mixture data and a table showing the old and new performance (data)The correspondence table 132 is created and stored in the correspondence table storage unit 130. The correspondence table 132 can be created based on, for example, a prototype or actual results.

[0022] Next, in step 1100, the learned old recipe 112 and old performance table 122 stored in the old recipe storage unit 110 and old performance table storage unit 120 are called up.

[0023] Next, proceed to step 1200, replace the old recipe 112 with the new recipe using the correspondence table 132 to create a new recipe 152, and store it in the new recipe storage unit 150.

[0024] Also, at step 1300, a new formulation data The old performance table 122 is corrected to create a new performance table 162, which is then stored in the new performance table storage unit 160.

[0025] Next, the process proceeds to step 1400 , where machine learning is performed again using the new recipe table 152 and the new performance table 162 , and in step 1500 , the re-learning results are stored in the learning result storage unit 210 .

[0026] Next, the process proceeds to step 1600, where the re-learning results and the new blend are used in the performance prediction device 300 to obtain a performance prediction table 310. The performance prediction table 310 thus obtained can be used to predict performance.

[0027] In step 1700, the blend optimization device 400 optimizes the blend using the re-learning results and the required performance, and obtains the blend optimization table 410. The blend optimization table 410 thus obtained can be used to obtain the optimal blend.

[0028] [Example 1] Example 1, in which the raw material after the change is a combination or blend of multiple raw materials, is shown in FIG.

[0029] The old recipe 113 is converted into the new recipe 153 by the correspondence table 133. Here, since the white dashi of the old recipe 113 has become difficult to obtain, the combination has been changed to bonito flakes, kelp, and water, and the new recipe 153 has been obtained.

[0030] [Example 2] FIG. 5 shows Example 2, which is an example of learning the manufacturing method used to manufacture a product together with the composition as product configuration information.

[0031] As shown in the old recipe 114, before the change, raw chikuwabu from Company A was used, but after the change, as shown in the correspondence table 134, quenched chikuwabu from Company B was used, resulting in the new recipe 154. Note that Kansai-style oden did not use chikuwabu from Company A, so the recipe remains the same.

[0032] [Example 3] FIG. 6 shows Example 3, which is an example of a case where changes in combinations in past learning data are limited to a portion of combinations for which the correspondence table is effective.

[0033] In the old recipe table 115, daikon radish of no specified origin was used for Kanto-style oden, Kansai-style oden, and Kyushu-style oden, but as shown in the correspondence table 135, the daikon radish was changed to Nerima-produced daikon radish, so as shown in the new recipe table 155, the daikon radish for Kanto-style oden and Kansai-style oden has been changed to Nerima-produced daikon radish, and Kyushu oden continues to use daikon radish of no specified origin.

[0034] [Example 4] FIG. 7 shows a fourth embodiment, which is an example of a system that stores parameter information for changing performance in accordance with the correspondence table and corrects performance evaluations of past learning data for use in re-learning when the performance evaluation of a product changes depending on the correspondence table to be applied.

[0035] In consideration of Correspondence Table 136, the radish without a specified origin in the old Recipe Table 116 has been replaced with Nerima-produced radish in the new Recipe Table 156, and the aroma of 2.0 and 4.0 in the old Performance Table 126 has changed to aroma of 2.4 and 4.8 in the new Performance Table 166.

[0036] In this way, even if there are changes in raw materials, machine learning can be performed without waiting for the accumulation of new learning data, and performance predictions and formulation suggestions can be made continuously based on the raw materials available at that time.

[0037] It also overcomes the lack of training data for formulations based on changing raw materials, which has been a limitation of machine learning.

[0038] Furthermore, in order to ensure accuracy in machine learning, by applying the present invention each time the raw materials are changed, it is possible to maintain sufficient learning data for the raw materials available at each time.

[0039] In the above embodiment, the present invention is applied to processed foods, but the application of the present invention is not limited to this, and it is clear that the present invention can be similarly applied to other general products. [Explanation of symbols]

[0040] 100...Data conversion device 110...Old recipe storage section 112, 113, 114, 115, 116...Old recipe 120…Old performance table storage section 122, 126...Old performance table 130...Correspondence table storage unit 132, 133, 134, 135, 136... Correspondence table 140...Correspondence table conversion section 150...New combination table storage section 152, 153, 154, 155, 156...New combination table 160…New performance table storage section 162, 166…New performance table 200...Machine learning device 210...Learning result memory unit 300...Performance prediction device 310...Performance prediction table 400... Blend optimization device 410...Mixture optimization table

Claims

1. In the case of a product made by blending and processing raw materials, when the performance of the product is determined by the blend or when a blend that satisfies the product's performance is determined by the blend using computerized machine learning, When the raw materials used in a product are changed, the system converts the composition of the past learning data stored in the storage device using a computer and re-learns based on the correspondence table data between the changed raw materials and the raw materials used for learning before the change. A computer-based machine learning processing method for product performance that enables computer-based machine learning without the need for performance evaluation of products using changed raw materials and the accumulation of such data.

2. 2. The computer-based machine learning processing method for product performance according to claim 1, wherein the correspondence table data is data on raw materials before and after the change, each of which is a combination or blend of multiple raw materials.

3. 2. A computer-based machine learning processing method for product performance according to claim 1, characterized in that data on the manufacturing method used in manufacturing the product is used as the learning target together with composition data as product configuration information.

4. 2. A computer-based machine learning processing method for product performance according to claim 1, wherein the conversion of formulation data in past learning data is limited to a portion of formulations for which the correspondence table data is valid.

5. 2. A computer-based machine learning processing method for product performance according to claim 1, characterized in that, when product performance evaluation changes depending on the applied correspondence table data, parameter information that changes performance in accordance with the correspondence table data is stored, and performance evaluation of past learning data is corrected and used for re-learning.

6. A method for predicting product performance by a computer, comprising: using the machine learning processing method for product performance by a computer according to any one of claims 1 to 5; and predicting product performance by a computer from formulation data used.

7. A method for presenting raw material blending data by a computer, which presents raw material blending data that satisfies specified product performance using a computer-based machine learning processing method for product performance described in any one of claims 1 to 5.

Images