Training data generation device, program, and training data generation method

Abstract

To make it possible to enable a generation of learning data for a machine learning model which makes a plan.SOLUTION: A learning data generation device 100 comprises: an instruction data generation unit 111 that generates new instruction data based on statistical information of instruction data (see facility database 130 and order history database 140) used when generating a plan; a plan generation unit 112 that generates the plan based on the new instruction data; and a plan evaluation unit 113 that displays an evaluation index of the plan, acquires a selected plan, and outputs the selected plan and the new instruction data corresponding to the selected plan.SELECTED DRAWING: Figure 1

Description

[Technical field]

[0001] The present invention relates to a learning data generation device, a program, and a learning data generation method for generating learning data for a machine learning model related to planning. [Background technology]

[0002] In manufacturing production planning, in addition to clear constraints such as production volume, delivery date, and cost, there are also constraints that are difficult to clarify, such as trends by customer or region. In addition, it is necessary to respond to rush orders, interruptions, and delayed delivery of parts. For this reason, it is not only necessary to use mathematical optimization technology to find the optimal solution that satisfies the constraints, but also to relax conditions and set priorities equivalent to the know-how of an expert.

[0003] Machine learning technology is a method for incorporating the know-how of experts. In order to utilize machine learning technology, learning data is necessary. The learning data expansion device described in Patent Document 1 includes an expansion processing unit that, based on an input sentence pair consisting of a sentence written in a source language and a sentence written in a target language, selects replacement candidates for an expression included in the sentence pair using thesaurus information of at least one of the source language or the target language, and replaces the expression to be replaced with the replacement candidate, thereby generating a new sentence pair consisting of a sentence written in a source language and a sentence written in a target language. [Prior art documents] [Patent documents]

[0004] [Patent Document 1] JP 2020-126360 A Summary of the Invention [Problem to be solved by the invention]

[0005] According to the training data augmentation device described in Patent Document 1, it is possible to generate a large amount of bilingual data by expanding existing bilingual data in order to improve the accuracy of machine translation. On the other hand, training data for production planning is product order information, information related to production equipment, and production plans for production equipment. For this reason, it is difficult to generate training data suitable for production planning using the training data augmentation device described in Patent Document 1, which generates training data that is text data. This is true not only for production planning, but also for training data for planning including delivery planning and personnel planning.

[0006] The present invention has been made in consideration of the above background, and has an objective of providing a training data generation device, a program, and a training data generation method that enable the generation of training data for a machine learning model that creates a plan. [Means for solving the problem]

[0007] In order to solve the above problems, the learning data generation device according to the present invention provides instruction data used when generating a plan. About existing instruction data the instruction data generation unit generating new instruction data based on the statistical information; a plan generation unit generating a plan based on the new instruction data; and a plan evaluation unit displaying evaluation indexes of the plan, acquiring a selected plan, and outputting the selected plan and the new instruction data corresponding to the selected plan. Effect of the Invention

[0008] According to the present invention, it is possible to provide a training data generation device, a program, and a training data generation method that enable generation of training data for a machine learning model that formulates a plan. Problems, configurations, and effects other than those described above will become clear from the description of the following embodiments. [Brief description of the drawings]

[0009] [Figure 1] FIG. 2 is a functional block diagram of the training data generating device according to the present embodiment. [Diagram 2]FIG. 2 is a data configuration diagram of a facility database according to the embodiment. [Diagram 3] FIG. 2 is a data configuration diagram of an order history database according to the embodiment. [Figure 4] FIG. 2 is a data configuration diagram of plan data according to the embodiment. [Diagram 5] FIG. 11 is a data configuration diagram of column-related information according to the embodiment. [Figure 6] FIG. 2 is a data configuration diagram of an evaluation index database according to the embodiment. [Figure 7] 11 is a diagram for explaining difference information of instruction data according to the embodiment; FIG. [Figure 8] FIG. 4 is a screen configuration diagram of an evaluation screen according to the embodiment. [Figure 9] 11 is a flowchart of a learning data generation process according to the present embodiment. DETAILED DESCRIPTION OF THE PREFERRED EMBODIMENTS

[0010] Overview of the training data generation device A learning data generation device that generates learning data for a machine learning model that formulates a production plan in a form (embodiment) for carrying out the present invention will be described below. The learning data generation device calculates statistical information for existing order information, and generates new (fictitious) order information based on the statistical information. The statistical information includes a distribution of ordered products and a distribution of the quantity for each product. Next, the learning data generation device generates a production plan corresponding to the new order information while changing an evaluation index. Examples of the evaluation index include productivity and delivery deadline compliance rate. Next, the learning data generation device displays the production plan together with the evaluation index, and outputs the production plan selected as learning data by a user of the learning data generation device and the order information corresponding to the production plan as learning data. In this way, the learning data generation device can generate learning data for machine learning to formulate production plans.

[0011] Configuration of the learning data generation device 1 is a functional block diagram of a training data generation device 100 according to this embodiment. The training data generation device 100 is a computer, and includes a control unit 110, a storage unit 120, and an input / output unit 180. User interface devices such as a display, a keyboard, and a mouse are connected to the input / output unit 180. The input / output unit 180 may include a communication device and may be capable of transmitting and receiving data to and from other devices. A media drive may also be connected to the input / output unit 180, enabling data exchange using a recording medium.

[0012] <Configuration of memory unit> The storage unit 120 includes storage devices such as a read only memory (ROM), a random access memory (RAM), and a solid state drive (SSD). The storage unit 120 stores an equipment database 130 (see FIG. 2 described later), an order history database 140 (see FIG. 3 described later), plan data 150 (see FIG. 4 described later), column-related information 160 (see FIG. 5 described later), an evaluation index database 170 (see FIG. 6 described later), and a program 128. The program 128 includes a description of the processing procedure for the learning data generation process (see FIG. 9 described later).

[0013] <Memory section: Equipment database> 2 is a data configuration diagram of the equipment database 130 according to this embodiment. The equipment database 130 is, for example, data in a table format, and stores, as information related to production equipment (equipment data), the identification information of the equipment (see the "equipment" attribute (column)), the part numbers of products that can be produced by the production equipment, and the production volume of the product per unit time in association with each other. FIG. 2 shows that the equipment with the identification information "P02" can produce 20 products with part number "A01" per unit time, and 25 products with part number "A02" per unit time. The equipment database 130 may also include other attributes, such as the time required to switch between products being produced.

[0014] <Memory section: Order history database> Fig. 3 is a data configuration diagram of the order history database 140 according to this embodiment. The order history database 140 is, for example, data in a table format, and stores, as order data, the identification information (product number) of an ordered product, the order date, the delivery date, and the quantity, in association with each other. Fig. 3 shows that an order for 40 units of a product with product number "A01" was placed on July 1st, and the delivery date is August 3rd. The order history database 140 may also include other attributes, such as the orderer.

[0015] <<Memory section: Project data>> Fig. 4 is a data configuration diagram of the planning data 150 according to this embodiment. The planning data 150 is, for example, data in a table format, and stores the identification information of the production equipment (see the attribute of "equipment"), the identification information of the product manufactured by the production equipment, the manufacturing date, and the quantity in association with each other. Fig. 4 shows that 40 units of a product with part number "A01" are to be manufactured by equipment with identification information "P03" on July 31st. The planning data 150 may include other attributes, for example, the manufacturing order on the manufacturing date.

[0016] <Memory section: Column-related information> 5 is a data configuration diagram of column relation information 160 according to this embodiment. The column relation information 160 indicates the relation (correlation) between columns (attributes) in the equipment database 130 and the order history database 140. The column relation information 160 is, for example, data in a tabular format, and stores data indicating which database the column pertains to, a column, and a relation that is a column related to the column in association with each other. The data value is "equipment" indicating the equipment database 130 or "order" indicating the order history database 140. If there is no related column, the relation becomes "independent."

[0017] Record 161 indicates that the part number column in order history database 140 is not related (correlated) to any other columns. Record 162 indicates that the quantity column in order history database 140 is related (correlated) to the part number.

[0018] <Memory section: Evaluation index database> 6 is a data configuration diagram of the evaluation index database 170 according to this embodiment. An evaluation index is an index for evaluating a production plan, and examples of the evaluation index include productivity, delivery deadline compliance rate, excess inventory, and production volume uniformity. The evaluation index database 170 is, for example, data in a table format, and stores evaluation indexes, weight ranges that are the ranges of weighting of the evaluation index in an evaluation value (evaluation function) described later, and trade-off indexes that are evaluation indexes in a trade-off relationship with the evaluation indexes, in association with each other. FIG. 6 shows that the evaluation index for "productivity" is in a trade-off relationship with the evaluation index for "delivery deadline compliance rate," and the weighting of the evaluation value is 5 to 30.

[0019] <Control Unit Configuration> 1, a description will be given of the configuration of the control unit 110. The control unit 110 is configured to include a CPU (Central Processing Unit), and is provided with an instruction data generation unit 111, a plan generation unit 112, and a plan evaluation unit 113.

[0020] <<Control unit: instruction data generation unit>> The instruction data generating unit 111 generates new instruction data from existing instruction data as input data for generating a production plan (used for generating a production plan). The instruction data in this embodiment is equipment data included in the equipment database 130 and order data included in the order history database 140.

[0021] First, the instruction data generation unit 111 groups existing instruction data and extracts difference information between the groups. The instruction data generation unit 111 groups the instruction data from a time perspective (by month or period), for example, and extracts the difference information. If the ratio of difference information of the instruction data is lower than a predetermined value, the degree of change in the instruction data is considered to be small, and the instruction data generation unit 111 does not generate new instruction data. If the ratio of difference information of the instruction data is equal to or greater than the predetermined value, the instruction data generation unit 111 generates new instruction data based on statistical information of the instruction data. The instruction data generation unit 111 may group the instruction data from a regional perspective, in addition to a time perspective, and extract the difference information.

[0022] FIG. 7 is a diagram for explaining difference information of the instruction data 210, 220 according to this embodiment. The instruction data 210 is order data for July among the instruction data in the order history database 140, and the instruction data 220 is order data for August. The difference information 230 is extracted by calculating the difference for each record (row) of the instruction data 210, 220. In detail, the instruction data generating unit 111 calculates "0" or "1" depending on whether a character string such as a part number matches or does not match. The instruction data generating unit 111 also calculates the difference in days for dates (order date, delivery date), and the difference in numbers for numerical values ​​as difference information. In FIG. 7, the first row of the instruction data 210, 220 matches except for the month (the order date and delivery date do not match in the month, but match in the day), and all differences are 0 (see the solid arrow). As for the second row, there are differences in the part number, order date, and quantity (see the dotted arrow). Supplementally, the order dates of "7 / 1" and "8 / 1" in the instruction data 210, 220 match on the same day except for the month, so the order date field in the difference information 230 is "0." Also, the order dates of "7 / 2" and "8 / 3" do not match on the same day except for the month (one day difference), so the order date field in the difference information 230 is "1."

[0023] The instruction data generation unit 111 obtains difference information 230 for all combinations of order data grouped by month within a specified period. If the difference rate in the difference information 230 is less than a specified value, the instruction data generation unit 111 considers the degree of change in the order data to be low and does not generate new instruction data. If the difference rate is equal to or greater than the specified value, the instruction data generation unit 111 generates order data that is to become new instruction data because the degree of change is high. The difference rate may be the rate of rows with differences (not 0) to all rows in all the difference information 230, or the rate of items with differences.

[0024] Next, the instruction data generation unit 111 calculates the distribution of the attribute values ​​(data values, column values) of the instruction data. If there is no associated column (the association in column association information 160 (see FIG. 5) is "independent"), the instruction data generation unit 111 calculates the distribution of the values ​​of the column. If there is an associated column, the instruction data generation unit 111 calculates the distribution of the values ​​of the column for each associated column value. For example, in the order history database 140, the "quantity" column is associated with "product number" (see record 162 in FIG. 5). The instruction data generation unit 111 calculates the distribution of "quantity" for each "product number".

[0025] Next, the instruction data generation unit 111 generates new instruction data according to the calculated distribution information. For example, if the proportion of order dates that are the 1st to the 5th is 10% each, and the remaining days are approximately equal in proportion, the instruction data generation unit 111 generates order dates to match this proportion. In detail, the instruction data generation unit 111 generates instruction data so that the instruction data with order dates of the 1st to the 5th each account for 10% of the total. Furthermore, the instruction data generation unit 111 generates instruction data so that the order dates of the remaining 50% of the instruction data are the 6th or later, and the proportions of each date are the same.

[0026] For example, if the quantity of products with part number "A03" is 30% for 3, 40% for 40, and 30% for 5, then the quantity of products with part number "A03" is generated to match these proportions.In this way, the instruction data generation unit generates new instruction data based on statistical information of the instruction data, such as the frequency of occurrence of data values ​​and the degree of correlation between data.

[0027] In this embodiment, the equipment database 130 is fixed and no difference occurs. Therefore, the instruction data generation unit 111 does not generate new instruction data for the equipment data. In the following explanation, it is assumed that the degree of change in the order history database 140 is high, and new instruction data is generated.

[0028] <Control Unit: Plan Generation Unit> Returning to FIG. 1, the control unit 110 will be described further. The plan generating unit 112 generates a plurality of production plans based on new instruction data for instruction data (order data) of a type with a high degree of change, and based on existing instruction data for instruction data (equipment data) of a type with a low degree of change. The plan generating unit 112 may generate a production plan while changing an evaluation function (objective function) by using a mathematical optimization method such as an integer mixed programming method or a genetic algorithm. The evaluation function includes weighted evaluation indexes, and the plan generating unit 112 generates a production plan while changing the weighting within a weighting range (see FIG. 6). The plan generating unit 112 generates a production plan while changing the weighting of evaluation indexes (see trade-off indexes in FIG. 6) that are in a trade-off relationship. According to FIG. 6, "productivity" and "delivery date adherence rate" are in a trade-off relationship, and a production plan is generated while increasing the weighting of "productivity" from 5 to 30 and decreasing the weighting of "delivery date adherence rate" from 100 to 10.

[0029] <Control Unit: Planning and Evaluation Unit> The plan evaluation unit 113 inquires of the user of the learning data generation device 100, acquires a production plan to be included in the learning data, and outputs it. First, the plan evaluation unit 113 displays an evaluation screen 250 (FIG. 8 described later) including evaluation indexes of the production plan generated by the plan generation unit 112 on a display connected to the input / output unit 180.

[0030] 8 is a screen configuration diagram of an evaluation screen 250 according to this embodiment. Arranged on the left side of the evaluation screen 250 are, from the top, a difference information display area 251, a column related information display area 252, and a distribution information display area 253. Arranged on the right side of the evaluation screen 250 are an evaluation index graph display area 254 and a save button 255. The difference information display area 251 displays the percentage of difference in the type of instruction data. The column related information display area 252 displays the column related information 160 (see FIG. 5). The distribution information display area 253 displays a graph showing the distribution information of the column selected in the drop-down list 256.

[0031] In the evaluation index graph display area 254, a graph is displayed in which the generated production plans are plotted with the evaluation indexes in a trade-off relationship selected in the drop-down list 257 as axes. In other words, the white and black circles on the graph indicate the production plans, and their coordinates indicate the evaluation indexes of the production plans. The black circles are production plans selected as learning data by the user. In detail, at the initial point when the evaluation screen 250 is displayed, all the plotted production plans are white circles and are not selected (non-selected). For example, when the user clicks on a white circle, the production plan that is the white circle is selected and becomes a black circle. Conversely, when the user clicks on a black circle, the selected production plan that is the black circle is unselected and becomes a white circle. When the save button 255 is pressed, the plan evaluation unit 113 outputs the selected production plan (black circle) together with the corresponding instruction data (instruction data that was input when the production plan was generated) as, for example, a file.

[0032] <Learning data generation process> FIG. 9 is a flowchart of the learning data generation process according to this embodiment. In step S11, the instruction data generating unit 111 generates the difference information 230 (see FIG. 7). In step S12, the instruction data generating unit 111 calculates a difference ratio from the difference information 230 generated in step S11, and identifies the type of instruction data to be generated when the ratio is equal to or greater than a predetermined value.

[0033] In step S13, the instruction data generation unit 111 calculates the distribution of attribute values ​​(data values, column values) for the instruction data to be generated identified in step S12. In step S14, the instruction data generating unit 111 generates new instruction data based on the distribution calculated in step S13.

[0034] In step S15, the plan generating unit 112 executes step S16 while changing the weight of the evaluation index included in the evaluation value. In step S16, the plan generation unit 112 generates a production plan that optimizes the evaluation value based on the instruction data generated in step S14 for instruction data whose difference is equal to or greater than a predetermined value, and based on the existing instruction data for instruction data whose difference is less than the predetermined value.

[0035] In step S17, the plan evaluation unit 113 displays an evaluation screen 250 (see FIG. 8) on a display connected to the input / output unit 180. The user selects a production plan to be adopted as learning data and presses the save button 255. In step S18, the plan evaluation unit 113 outputs the production plan selected by the user as learning data and the corresponding instruction data.

[0036] <Features of the learning data generation device> The learning data generation device 100 calculates the difference between the equipment data in the equipment database 130 and the order data in the order history database 140, which are instruction data, and generates new order data based on statistical information of the order data in which the rate of difference is equal to or greater than a predetermined value. Next, the learning data generation device 100 uses the new order data and the equipment data in the equipment database 130 as input, and generates a production plan while changing the weighting of the evaluation index. The learning data generation device 100 then displays the evaluation index of the generated production plan, and outputs the production plan selected by the user as learning data, and the instruction data that was input when the production plan was generated.

[0037] In this way, the training data generation device 100 can generate training data desired by the user. The training data includes instruction data that has the same statistical information as the existing instruction data and is close to actual data, and a production plan that is generated using the instruction data as an input and that the user judges to be desirable. By learning such training data, the machine learning model can create a production plan desired by the user.

[0038] <<Variation: Column-related information>> In the above embodiment, the relationship between columns is stored in the column relationship information 160. The relationship may be acquired by analyzing existing instruction data and calculating the degree of correlation between columns. In this way, the effort required to set the column relationship information 160 can be reduced.

[0039] <<Variation: Target of the Plan>> In the above embodiment, the plan generated is a production plan, but is not limited to this. The learning data generation device 100 may generate learning data for a machine learning model that creates plans such as a delivery plan in which the instruction data is a demand amount and a delivery capacity, or a personnel plan in which the instruction data is personnel skills and allocation conditions.

[0040] Other variations Although several embodiments of the present invention have been described above, these embodiments are merely illustrative and do not limit the technical scope of the present invention. For example, in the above-described embodiment, the training data generation device 100 stores the order history database 140 and other data in the storage unit 120, but this is not limiting. For example, the training data may be generated while accessing data stored in another device, such as a storage device on the cloud.

[0041] The present invention can take various other embodiments, and furthermore, various modifications such as omissions and substitutions can be made without departing from the gist of the present invention. These embodiments and modifications are included in the scope and gist of the invention described in this specification, etc., and are included in the scope of the invention described in the claims and their equivalents. [Explanation of symbols]

[0042] 100 Learning data generation device 111 Instruction data generation unit 112 Planning Generation Unit 113 Planning and Evaluation Department 130 Equipment database (instruction data) 140 Order history database (instruction data) 150 Planning Data (Plan) 160 Column Related Information 170 Evaluation Index Database 250 Evaluation screen

Claims

1. an instruction data generating unit that generates new instruction data based on statistical information of existing instruction data for use in generating a plan; a plan generating unit that generates a plan based on the new instruction data; a plan evaluation unit that displays an evaluation index of the plan, acquires a selected plan, and outputs the selected plan and the new instruction data corresponding to the selected plan. A training data generating device comprising:

2. The instruction data generating unit In the plurality of types of instruction data included in the instruction data, For the type of instruction data in which the rate of difference between the pieces of instruction data is equal to or greater than a predetermined rate, the new instruction data is generated based on statistical information of the type of instruction data; For the instruction data of the type in which the rate of difference is less than the predetermined rate, the instruction data of the type is set as the new instruction data.

2. The training data generating device according to claim 1 .

3. The statistical information is The indication data includes at least one of an occurrence frequency of a data value included in the indication data and a correlation degree between the data.

3. The training data generating device according to claim 2.

4. The plan generation unit generating the plan in which an evaluation value, which is a combination of the weighted evaluation indexes, is optimal; A plurality of plans are generated by changing the weighting of the plurality of evaluation indexes that are in a trade-off relationship.

2. The training data generating device according to claim 1 .

5. The plan is a production plan for a product, The instruction data is order information including at least one of the identification information, order date, delivery date, and quantity of the product.

2. The training data generating device according to claim 1 .

6. Computer, an instruction data generating unit that generates new instruction data based on statistical information of existing instruction data for use in generating a plan; A plan generating unit that generates a plan based on the new instruction data; and A plan evaluation unit that displays an evaluation index of the plan, acquires a selected plan, and outputs the selected plan and the new instruction data corresponding to the selected plan. A program to function as a

7. A training data generating device, generating new instruction data based on statistical information of existing instruction data for use in generating the plan; generating a plan based on the new prescription data; displaying an evaluation index of the plan, obtaining a selected plan, and outputting the selected plan and the new instruction data corresponding to the selected plan. A method for generating training data.

Images