Learning system

Abstract

To provide a learning system that performs model learning using data collected in a plurality of databases managed by any name.SOLUTION: A base database installed at each of a plurality of bases, stores measurement data measured from a target apparatus installed in the base, in association with a name that represents a type of the measurement data. A name database stores names used in the base database for each base, which are given to a same type of measurement data, in association with each other. A name identification unit reads out, from the name database, a second name that is a name associated with a first name that is a known name related to data used as a learning dataset for machine learning. A learning unit performs learning processing for a learning model using a learning dataset generated from first measurement data, which is measurement data associated with a first name, and second measurement data, which is measurement data associated with a second name, in the plurality of base databases.SELECTED DRAWING: Figure 1

Description

[Technical Field]

[0001] The present invention relates to a learning system. [Background technology]

[0002] A trained model may be created using data collected at a business site. However, if the amount of data collected at the business site is not large, the model may not be trained sufficiently, resulting in low accuracy. Therefore, there is a demand to use data collected not only at the business site but also at other business sites as a training dataset. Patent Document 1 discloses a technology for collecting sensor value data from multiple companies and generating a learning model. [Prior art documents] [Patent documents]

[0003] [Patent Document 1] Japanese Patent Publication No. 2022-128069 Summary of the Invention [Problem to be solved by the invention]

[0004] However, in multiple databases managed in different environments, such as offices or companies, the names of labels indicating the same item may differ depending on the environment. Furthermore, even if the names are the same or similar, the items they represent may differ in the database. Furthermore, not only in databases, but also in different environments, the names of terms indicating the same item may differ. Therefore, when accessing data managed in different environments, the method described in Patent Document 1 may have difficulty obtaining data indicating the desired item. An object of the present invention is to provide a learning system that can learn a model using data collected in multiple databases managed under arbitrary names. [Means for solving the problem]

[0005] [1] According to a first aspect, the learning system includes a plurality of base station databases 10 provided at each of a plurality of base stations, each of which stores measurement data measured from a target device provided at the base station in association with a name representing the type of the measurement data; a name database that stores names given to measurement data of the same type among the names used in the database for each base station in association with each other; a name identification unit that reads out from the name database a second name that is a name associated with a first name that is a known name for data used as a learning dataset for machine learning; and a learning unit that performs a learning process for a learning model using a learning dataset generated in the plurality of base station databases 10 from the first measurement data that is measurement data associated with the first name and the second measurement data that is measurement data associated with the second name.

[0006] [2] According to a second aspect, in the learning system described in [1] above, the name database may store metadata related to measurements of measurement data related to the name in association with the name, and the learning dataset may include data that has been pre-processed to align the units of the first measurement data and the second measurement data based on the metadata associated with the first name and the metadata associated with the second name.

[0007] [3] According to a third aspect, the learning system described in [2] above may include a data processing unit that aligns the units of the first measurement data and the second measurement data based on metadata associated with the first name and metadata associated with the second name, and the learning unit may perform learning processing of the learning model using the learning dataset including the first measurement data and the second measurement data processed by the data processing unit.

[0008] [4] According to a fourth aspect, the learning system according to any one of [1] to [3] above may include an evaluation acquisition unit that acquires an evaluation value related to the measurement data, and a presentation unit that presents the name related to the measurement data whose evaluation value is higher than a threshold value. [Effects of the Invention]

[0009] According to the above aspect, the learning system can learn a model using data collected in multiple databases managed under arbitrary names. [Brief explanation of the drawings]

[0010] [Figure 1] 1 is a schematic block diagram showing the system configuration of a learning system according to an embodiment. [Figure 2] 1 is a schematic block diagram illustrating a functional configuration of an information providing device according to an embodiment. [Figure 3] FIG. 2 is a diagram showing a specific example of a term registration database according to the embodiment. [Figure 4] FIG. 2 is a diagram showing a specific example of a term relation database according to the embodiment. [Figure 5] FIG. 2 is a diagram showing a specific example of a term label relation database according to the embodiment. [Figure 6] FIG. 2 is a diagram showing a specific example of a label registration database according to the embodiment. [Figure 7] FIG. 1 is a schematic block diagram illustrating a functional configuration of a learning device according to an embodiment. DETAILED DESCRIPTION OF THE INVENTION

[0011] Hereinafter, the embodiments will be described in detail with reference to the drawings. Figure 1 is a schematic block diagram showing the system configuration of a learning system 100 according to an embodiment. The learning system 100 executes learning processing of a model used at each base using data collected in a database (base database 10) operated at multiple bases A. Examples of bases A include offices, factories, and companies.

[0012] The learning system 100 includes multiple base databases 10, multiple edge terminals 30, and an information providing device 50. The base database 10 and the edge terminal 30 are provided at each of multiple bases A. In the example shown in FIG. 1, the learning system 100 includes four base databases 10 (base database 10-1, base database 10-2, base database 10-3, and base database 10-4) and four edge terminals 30 (edge ​​terminal 30-1, edge terminal 30-2, edge terminal 30-3, and edge terminal 30-4) provided at four different bases A (base A1, base A2, base A3, and base A4). However, there may be a base where no edge terminal 30 is provided, or there may be a base where multiple edge terminals 30 are provided.

[0013] The base database 10 stores measurement data measured from target equipment installed at each base. The type of measurement data is represented by a label (key) that represents the measurement data. A label is an identifier for a record in the base database 10. The name of the label is determined so that users of the base database 10 can easily understand the contents of the corresponding record. Note that different labels may be assigned to the same type of measurement data between different base databases 10. This is because labels are assigned at each base A, and label names are often not discussed between bases A. The label names may use terms used at the base. The same data may be expressed using different terms at different bases. Hereinafter, terms used at different bases may also be referred to as specific terms. In the case of a steelworks, examples of target equipment include blast furnaces, converters, continuous casting machines, heating furnaces, rolling mills, and the like.

[0014] The edge terminal 30 performs calculation processing using the trained model trained by the learning device 70. The edge terminal 30 inputs measurement data measured from the target device into the trained model to perform calculation processing, and executes processing such as presenting the status of the target device and controlling the target device.

[0015] The information providing device 50 identifies the names of labels associated with the same type of measurement data used in multiple base databases 10. For example, when the information providing device 50 receives the specification of a label used in a specific base database 10 (e.g., base database 10-1), it identifies the names of labels in the other base databases 10 that represent the same type of measurement data as the specified label. The information providing device 50 collects and outputs the measurement data associated with the specified label from each base database 10.

[0016] The learning device 70 performs a learning process for the machine learning model using a learning data set generated based on measurement data collected by the information providing device 50 from a plurality of base station databases 10.

[0017] The base database 10, edge terminal 30, information providing device 50, and learning device 70 are communicably connected to one another via a network N. The network N may be a network using wireless communication or a network using wired communication. The network N may be configured using, for example, the Internet or a local area network (LAN). The network N may also be configured by combining multiple networks.

[0018] 2 is a schematic block diagram showing the functional configuration of an information providing device 50 according to an embodiment. The information providing device 50 is configured using an information processing device such as a personal computer or a server device. The information providing device 50 includes a communication interface 51, a storage 53, a main memory 55, and a processor 57.

[0019] The communication interface 51 performs data communication with other devices (for example, the base database 10) via the network N under the control of the processor 57. The communication interface 51 may perform wireless communication or wired communication.

[0020] The storage 53 is configured using a storage device such as a magnetic hard disk drive or a semiconductor storage device. The storage 53 stores data used by the processor 57. The storage 53 stores, for example, a term registration database 531, a term relation database 532, a term-label relation database 533, and a label registration database 534.

[0021] The term registration database 531 stores information about terms of data used in the base database 10. The terms used in the base database 10 include common terms and specific terms. FIG. 3 is a diagram illustrating a specific example of the term registration database 531 according to the embodiment. The term registration database 531 stores multiple term information records. Each term information record includes values ​​for a term name, environment information, a term storage location, and a description. The term name indicates a character string representing the name of the term. The environment information indicates the environment in which the term is used. The term storage location indicates a location (e.g., a path) where information about the term is stored. The storage location may be, for example, a folder. In the following description, an example will be described in which a folder is used as the concept of a storage location. However, the concept of a storage location does not necessarily have to be limited to a folder. For example, the folder name of each folder indicated by a directory or storage location may contain information indicating the environment in which the term (specific term) is used. With this configuration, for a specific term stored in a folder, the folder name can be used as information indicating the environment in which the specific term is used. For example, if a folder name contains a character string indicating a region name, such as "Area B," the folder name may be used as information indicating the environment (region) in which the specific term is used. The description is metadata that represents the content, properties, and measurement-related information of the measurement data related to the term. Examples of measurement-related information include the units of measurement data and sampling intervals.

[0022] In the example of Figure 3, the term information record shown in the top row has the following values ​​as the term name, environmental information, term storage location, and description: "Tundish ΔT," "Common," "Common / Steelmaking / 08_Continuous Casting," and "XX Temperature Record." This term information record indicates that the term represented by the string "Tundish ΔT" is used as a common term, information about its value is registered in the storage location "Common / Steelmaking / 08_Continuous Casting," and its value indicates "XX Temperature Record." The term information record shown in the second row from the top indicates that the term represented by the string "TD-SH Record" is used as a specific term in Region B, information about its value is registered in the storage location "Regional Terms / Region B / Steelmaking / 09_Continuous Casting Bloom," its value indicates a temperature in degrees Celsius, and its sampling interval is X1 seconds.

[0023] The terminology relation database 532 has information regarding the relationship between terms indicating target matters (common terms) in the base database 10 and terms used to indicate the same target matters (specific terms). Target matters are, for example, events, objects, work performed, and information obtained at each base. Target matters may, for example, indicate a part of the steelmaking process, information obtained in a part of the process, or objects (materials and tools) used in steelmaking. A common term is a term set by the information providing device 50 as a representative name indicating the target matter. In the terminology relation database 532, one common term is stored for the same target matter. A specific term is a term used to indicate the target matter at a specific base A. In the terminology relation database 532, multiple specific terms may be stored for the same target matter.

[0024] FIG. 4 is a diagram showing a specific example of the term relation database 532 according to the embodiment. The term relation database 532 has a plurality of term relation records. Each term relation record has values ​​for the term name of a common term, the storage location of the common term, the term name of a specific term, and the storage location of the specific term. The term name of a common term indicates the character string itself indicating the name of the common term. The storage location of a common term indicates the location (e.g., a path) where information related to the common term is registered. The term name of a specific term indicates the character string itself indicating the name of a specific term used as a term indicating the same subject matter as a common term in the same record. The storage location of a specific term indicates the location (e.g., a path) where information related to the specific term is recorded.

[0025] In the example of Figure 4, term relation records for three specific terms (TD-SH performance, TD-SH performance ΔT, TD molten steel temperature) are registered in association with the common term "tundish ΔT." The term relation record shown in the top row indicates that information about the value of the common term represented by the character string "tundish ΔT" is registered in the recording location "Common / Steelmaking / 08_Continuous Casting," that there is a specific term indicating the same subject matter, "TD-SH performance," and that information about the value of this specific term is registered in the recording location "District Terms / District B / Steelmaking B / 09_Continuous Casting Bloom."

[0026] The common term "tundish ΔT" is a common term used at base A1 (e.g., the head office) in region A (e.g., the region where the head office is located). The specific term "TD-SH actual result" is a specific term used at base A2 in region B. The specific term "TD-SH actual result ΔT" is a specific term used at base A3 in region C. The specific term "TD molten steel temperature" is a specific term used at base A4 in region D. The four terms "tundish ΔT", "TD-SH actual result", "TD-SH actual result ΔT", and "TD molten steel temperature" all indicate values ​​related to the same managed phenomenon, "molten steel temperature in the tundish of continuous casting equipment."

[0027] The term-label relation database 533 has information regarding the relationship between terms used in the base database 10 and one or more labels related to each term. FIG. 5 is a diagram showing a specific example of the term-label relation database 533 according to the embodiment. The term-label relation database 533 has a plurality of term-label relation records. Each term-label relation record has values ​​for a term name, a term storage location, a label name, and a label storage location. The term name indicates the character string itself indicating the name of the term. The term storage location indicates the location (e.g., a path) where information related to the term is stored. The label name indicates the character string itself indicating the name of the label related to the term in the same record. The label storage location indicates the data location (e.g., a path) where measurement data related to the label is stored. Note that the term names included in the term-label relation database 533 may be common terms or specific terms.

[0028] In the example of Figure 5, term-label relation records for four labels (TD-SH, TD-SH_minimum value, TD-SH_maximum value, TD-SH_average value) are registered in association with the term "TD-SH performance." The term-label relation record shown in the top row indicates that information about the term represented by the string "TD-SH performance" is registered in a folder called "Regional terms / Region B / Steelmaking B / 09_Continuous casting bloom," and that the value of one of its labels, "TD-SH," is recorded in a location called "B_Processing Control / Library Region B / CC Casting Section Data."

[0029] 5, it can be seen that there are at least four label values ​​(TD-SH, TD-SH_minimum value, TD-SH_maximum value, TD-SH_average value) associated with the term TD-SH performance. It can also be seen that if one wishes to obtain measurement data related to each label, the measurement data related to that label can be obtained by accessing the recording location of that label (site database 10).

[0030] The label registration database 534 stores information about labels used in the base database 10. Labels used in the base database 10 are associated with each term. FIG. 6 is a diagram illustrating a specific example of the label registration database 534 according to an embodiment. The label registration database 534 stores multiple label information records. Each label information record includes values ​​for a label name, a label storage location, a data type, a length, a description, and an evaluation value. The label name indicates the character string itself indicating the name of the label. The label storage location indicates the location (e.g., a path) in the base database 10 where the measurement data associated with the label is registered. The data type indicates the data type (e.g., numeric, character string, integer, floating-point, binary, etc.) of the measurement data associated with the label. The length indicates the length of the measurement data (e.g., the number of digits, the number of bits, the number of bytes, etc.). The description indicates the content and properties of the value associated with the label. For example, the data update period and the unit of the data value may be recorded as the description value. The evaluation value indicates the usefulness of the measurement data associated with the label. The evaluation value is given by a user who uses the measurement data for machine learning or the like. Thus, the label registration database 534 contains attribute information of the measurement data associated with the label, such as data type, length, description, etc. In other words, the data type, length, and description in the label registration database 534 are metadata that represent information about the measurement.

[0031] In the example of Figure 6, the label information record shown in the top row has the following values ​​for the label name, label recording location, data type, length, description, and evaluation value: "TD-SH", "B_Processing Control / Library B Area / CC Casting Section Data", "NUMBER", "5", "Tundish", and "6.1", respectively. This label information record indicates that the measurement data related to the label represented by the character string "TD-SH" is registered in the location "B_Processing Control / Library B Area / CC Casting Section Data", that the value is a "NUMBER" type with a length of "5", that the content is a value related to "Tundish", and that the evaluation value is "6.1".

[0032] The term registration database 531, term relation database 532, term label relation database 533, and label registration database 534 stored in storage 53 collectively constitute a name database that associates and stores names (term names and label names) given to measurement data of the same type.

[0033] By executing a program, the processor 57 functions as a database control unit 571, a name identification unit 572, a data collection unit 573, a data processing unit 574, and an information provision unit 575. Note that all or part of the functions of the processor 57 may be realized using hardware such as an ASIC, a PLD, or an FPGA. The above program may be recorded on a computer-readable recording medium. Examples of computer-readable recording media include portable media such as flexible disks, magneto-optical disks, ROMs, CD-ROMs, and semiconductor storage devices (e.g., SSDs), as well as storage devices such as hard disks and semiconductor storage devices built into a computer system. The above program may be transmitted via a telecommunications line.

[0034] The database control unit 571 controls the contents of various databases stored in the storage 53. For example, when an instruction to register a new record is received, the database control unit 571 registers the new record in the database to which the record is to be registered. For example, when an instruction to register a new term registration record is received in the term registration database 531, the database control unit 571 registers the new term registration record in the term registration database 531. The instruction to register a new record may be received from another device, for example, via the communication interface 51. Furthermore, the information providing device 50 may be equipped with a user interface such as an input device and an output device. In this case, the database control unit 571 may register a new term registration record input by operating the user interface (input device) in the term registration database 531.

[0035] The name identification unit 572 receives a label name search request from another device. The label name search request includes a term name or label name (first name) known to the user. The name identification unit 572 searches a database stored in the storage 53 for a label name (second name) that represents the same type of measurement data as the name included in the search request. The label name searched for may be a name unknown to the user.

[0036] When a term name is included in a search request, the name identification unit 572 searches for label names as follows. First, the name identification unit 572 identifies a common term corresponding to the input term name from the term relation database 532. Note that the input term name itself may be a common term. The name identification unit 572 reads out a specific term associated with the identified common term from the term relation database 532. The name identification unit 572 reads out label names associated with the identified common term and specific term from the term-label relation database 533. This allows the name identification unit 572 to identify label names that represent the same type of measurement data from the term name.

[0037] When a search request includes a label name, the name identification unit 572 searches for the label name as follows. First, the name identification unit 572 reads out a term name associated with the input label name from the term-label relation database 533. The name identification unit 572 identifies a common term corresponding to the read term name from the term relation database 532. Note that the term name associated with the input label name may itself be a common term. The name identification unit 572 reads out a specific term associated with the identified common term from the term relation database 532. The name identification unit 572 reads out label names associated with the identified common term and specific term from the term-label relation database 533. This allows the name identification unit 572 to identify label names that represent the same type of measurement data from the label name.

[0038] The data collection unit 573 collects, from each base database 10, measurement data associated with the label name included in the search request and the label name searched for by the name identification unit 572. The data processing unit 574 performs a process of aligning the units of the measurement data collected by the data collection unit 573 to the units of the measurement data related to the label included in the search request. Here, the "unit of measurement data" refers to the data type, data length (including the number of pixels if the measurement data is an image), sampling interval (including the frame rate if the measurement data is a video), measurement unit, etc. of the measurement data. For example, when the sampling intervals are different, the data processing unit 574 can align the sampling intervals by performing data interpolation and resampling at an interval that is the least common multiple of the two sampling intervals. Furthermore, when the number of pixels of image data is different, the data processing unit 574 can align the number of pixels by enlarging or reducing the image data. Note that the data processing unit 574 according to other embodiments may align the units of the measurement data to any unit. The data processing unit 574 performs a process of aligning the units based on metadata (description, data type, length) recorded in the term registration database 531 and the label registration database 534. The processing by the data processing unit 574 is preprocessing for learning by the learning device 70.

[0039] The information providing unit 575 outputs the measurement data associated with the label name included in the search request and the data processed by the data processing unit 574 to the device that issued the search request.

[0040] 7 is a schematic block diagram showing the functional configuration of a learning device 70 according to an embodiment. The learning device 70 is configured using an information processing device such as a personal computer or a server device. The learning device 70 includes a communication interface 71, a storage device 73, a main memory 75, and a processor 77.

[0041] The communication interface 71 performs data communication with other devices (for example, the information providing device 50) via the network N under the control of the processor 77. The communication interface 71 may perform wireless communication or wired communication.

[0042] The storage 73 is configured using a storage device such as a magnetic hard disk drive or a semiconductor storage device. The storage 73 stores data used by the processor 77. The storage 73 stores a learning model 731. The learning model 731 is unlearned in an initial state, and the parameter values ​​are learned by a learning process by the learning device 70, resulting in a trained model.

[0043] By executing the program, the processor 77 functions as a request unit 772, a model construction unit 771, a learning unit 773, an execution unit 774, a transfer unit 775, and an evaluation unit 776. Note that all or part of the functions of the processor 77 may be realized using hardware such as an ASIC, a PLD, or an FPGA. The above program may be recorded on a computer-readable recording medium. Examples of computer-readable recording media include portable media such as flexible disks, magneto-optical disks, ROMs, CD-ROMs, and semiconductor storage devices (e.g., SSDs), as well as storage devices such as hard disks and semiconductor storage devices built into computer systems. The above program may be transmitted via a telecommunications line.

[0044] The request unit 772 receives input from a user of the name (first name) of a term or label of measurement data to be used as a learning dataset for machine learning, and causes the information providing device 50 to collect measurement data related to the name and measurement data of the same type. The measurement data provided by the information providing device 50 is preprocessed by the information providing device 50 so that the units are consistent.

[0045] The model construction unit 771 determines the structure of the learning model 731 according to input from the user. The model construction unit 771 may determine the number of dimensions of the input and output of the learning model 731 based on the measurement data acquired by the request unit 772 from the information providing device 50. The model construction unit 771 records the constructed learning model 731 in the storage 73 with initial value parameters.

[0046] The learning unit 773 performs learning of the learning model 731 recorded in the storage 73. The learning unit 773 classifies the measurement data acquired by the request unit 772 into input samples and output samples according to the structure of the learning model 731, and generates a learning dataset. The learning unit 773 may present a list of labels of the measurement data acquired by the request unit 772, and allow the user to select those to be used in the learning dataset. At this time, the list may include an evaluation value assigned to each label. The learning unit 773 executes a learning process for the learning model 731 using the generated learning dataset. The learning unit 773 updates the parameters of the learning model 731 stored in the storage 73 during the learning process.

[0047] The execution unit 774 executes inference processing using the trained learning model 731 whose parameters have been updated by the learning unit 773. For example, the execution unit 774 may perform inference processing by acquiring the latest measurement data from the site database 10 of a specific site A and inputting the measurement data into the learning model 731. By using the learning device 70 to check the functionality of the learning model 731 using the measurement data at the site A where the model is to be introduced, the user can check the usefulness and adjust the model before introducing the model 731 to the edge terminal 30 at site A.

[0048] The transfer unit 775 transfers the program including the learning model 731 recorded in the storage 73 to the edge terminal 30. For example, the transfer unit 775 can perform the transfer to the edge terminal 30 by the following procedure. The transfer unit 775 builds a container-type virtual environment in the edge terminal 30. The transfer unit 775 generates image data that reproduces an execution environment including the program including the learning model 731. The evaluation unit 776 copies the generated image data to the edge terminal 30. This allows the edge terminal 30 to execute the program on the container using the image data. By using the container-type virtual environment, the edge terminal 30 can execute the program in an environment similar to that of the learning device 70, thereby reducing the impact of changes in the environment.

[0049] The evaluation unit 776 receives an input of an evaluation value from a user for the measurement data obtained by the request unit 772. For example, the user can assign a high evaluation value to data that has contributed to improving the accuracy of the learning model 731, or to useful data that has not been collected at other sites A.

[0050] Here, an example of the operating procedure of the learning system 100 will be explained using the example of introducing a learning model 731 that predicts the withdrawal speed (i.e., casting speed) of a cast piece based on the time series of the molten steel temperature in the tundish of a continuous casting equipment to an edge terminal 30 used at base A1 using the above-mentioned learning system 100.

[0051] First, the user accesses the learning device 70 using a terminal device such as a PC. The user instructs the learning device 70 to collect measurement data to be used for learning. In this example, the user inputs into the learning device 70, as first names, a name representing the "molten steel temperature in the tundish" used at site A1 (tundish ΔT in the example of Figure 3) and a name representing the "slab withdrawal speed." The request unit 772 of the learning device 70 generates a search request including the first name input by the user and transmits it to the information providing device 50.

[0052] In response to a search request received from the learning device 70, the name identification unit 572 of the information providing device 50 identifies second names used at other bases that are associated with the first name included in the search request from the term registration database 531, terminology relation database 532, term-label relation database 533, and label registration database 534. For example, the name identification unit 572 identifies label names such as "TD-SH," "TD-SH_minimum value," and "TD-SH_maximum value" used at the base A2 as second names from the first name "Tundish ΔT" used at the base A1. The data collection unit 573 collects measurement data associated with the first and second names from each base database 10. The data processing unit 574 reads metadata associated with the first and second names from the term registration database 531 and label registration database 534 and performs processing to align the unit of measurement data associated with the second name with the unit of measurement data associated with the first name. The information providing unit 575 transmits information related to the first and second names and the measurement data to the learning device 70.

[0053] The model construction unit 771 of the learning device 70 determines the structure of the learning model 731 based on the data format and data length of the two types of acquired measurement data. For example, the model construction unit 771 determines the number of dimensions of the input layer of the learning model 731 according to the data length of the "molten steel temperature in the tundish," and determines the number of dimensions of the output layer of the learning model 731 according to the data length of the "slab withdrawal speed." The learning unit 773 classifies the measurement data acquired from the information providing device 50 into input samples and output samples, and generates a learning dataset. The learning unit 773 executes a learning process for the learning model 731 using the learning dataset, and updates the parameters of the learning model 731 stored in the storage 73.

[0054] Once the learning process is completed according to the specified termination conditions, the user can test the learning model 731 using the measurement data from the site A1 to which it is to be applied. The user specifies the site database 10 and label name associated with the measurement data to be input into the learning model 731 in the learning device 70. That is, the user specifies the site database 10-1 and "tundish ΔT" as the site database 10 and label name. The execution unit 774 acquires the measurement data associated with the specified label name from the specified site database 10 and inputs it into the learning model 731 to perform inference processing. As a result, the execution unit 774 can obtain a predicted value for the "slab withdrawal speed" from the learning model 731. The user can check the predicted value obtained from the learning model 731 and confirm or adjust the usefulness of the learning model 731.

[0055] Once the adjustments for the learning model 731 are complete, the user considers introducing the learning model 731 to the edge terminal 30-1 at site A1. To introduce the learning model 731 to the edge terminal 30, the user inputs a transfer instruction to the learning device 70, including the learning model 731 to be transferred and the edge terminal 30 to which it is to be transferred. The transfer unit 775 builds a container-type virtual environment in the specified edge terminal 30. The transfer unit 775 generates image data that reproduces an execution environment including a program containing the learning model 731. The evaluation unit 776 copies the generated image data to the edge terminal 30. This allows the edge terminal 30 to execute the program on the container using the image data. By executing the learning model 731 on the edge terminal 30, the edge terminal 30 can directly acquire measurement data from site A1 and reduce the computational resource occupation rate of the learning device 70.

[0056] Furthermore, the user can evaluate the data used in generating the learning model 731. The user inputs evaluation information that associates the label names of the measurement data to be evaluated with evaluation values ​​to the learning device 70. The evaluation unit 776 transmits the input label names and evaluation values ​​to the information providing device 50. The database control unit 571 of the information providing device 50 updates the evaluation values ​​of the labels recorded in the label registration database 534 based on the received combination of label names and evaluation values. The database control unit 571 may calculate the updated evaluation value, for example, by using the average value of past evaluation values ​​or the moving average of the most recent evaluation values.

[0057] The learning system 100 configured in this way reads, from the name database, a second name associated with a first name, a known name associated with data used as a machine learning learning dataset, and performs a learning process for a learning model using a learning dataset generated from measurement data associated with the first name and measurement data associated with the second name. This allows the learning system 100 to identify, from an arbitrary label (first name), a label (second name) used in another base database, even if similar items are associated with different labels in each base database. Therefore, the learning system 100 can train a model using data collected in multiple databases.

[0058] Although one embodiment has been described in detail above with reference to the drawings, the specific configuration is not limited to the above, and various design modifications are possible. That is, in other embodiments, the order of the above-described processes may be changed as appropriate. Furthermore, some processes may be executed in parallel. Although the information providing device 50 and the learning device 70 according to the above-described embodiments are implemented by separate computers, this is not limiting. For example, in other embodiments, the information providing device 50 and the learning device 70 may be implemented on the same computer. The information providing device 50 may each be configured by a single computer, or the configuration of the information providing device 50 may be distributed across multiple computers, with the multiple computers cooperating with each other to function as the information providing device 50. Similarly, the learning device 70 may each be configured by a single computer, or the configuration of the learning device 70 may be distributed across multiple computers, with the multiple computers cooperating with each other to function as the learning device 70.

[0059] The information providing device 50 according to the embodiment described above identifies labels representing similar data, then collects data from the base database 10 and performs data preprocessing. However, this is not limited to this. For example, in a learning system 100 according to another embodiment, the learning device 70 may collect data from the base database 10 and perform data preprocessing. In another embodiment, the information providing device 50 may present labels representing similar data to a user, and the user may collect data from the base database 10 and perform data preprocessing. In another embodiment, the collection of data from the base database 10 and the preprocessing of the data may be performed by different entities. Furthermore, if a calculation formula is registered as a term, the information providing device 50 may perform calculations as preprocessing based on the calculation formula. [Explanation of symbols]

[0060] 10...Base database 100...Learning system 30...Edge terminal 50...Information providing device 51...Communication interface 53...Storage 531...Term registration database 532...Term relation database 533...Term label relation database 534...Label registration database 55...Main memory 57...Processor 571...Database control unit 572...Name identification unit 573...Data collection unit 574...Data processing unit 575...Information providing unit 70...Learning device 71...Communication interface 73...Storage 731...Learning model 75...Main memory 77...Processor 771...Model construction unit 772...Request unit 773...Learning unit 774...Execution unit 775...Relocation unit 776...Evaluation unit A...Base N...Network

Claims

1. a plurality of base station databases provided at a plurality of base stations, each of which stores measurement data measured from a target device provided at the base station in association with a name representing a type of the measurement data; a name database that stores names given to the same type of measurement data in association with each other among the names used in the base station database for each base station; a name identification unit that reads, from the name database, a second name that is a name associated with a first name that is a known name related to data used as a learning dataset for machine learning; a learning unit that performs a learning process for a learning model using a learning dataset generated from first measurement data that is measurement data associated with the first name and second measurement data that is measurement data associated with the second name in the plurality of base station databases; A learning system that includes:

2. the name database stores metadata relating to measurement of the measurement data relating to the name in association with the name; The learning dataset includes data that has been preprocessed to align the units of the first measurement data and the second measurement data based on metadata associated with the first name and metadata associated with the second name. The learning system of claim 1 .

3. a data processing unit that aligns the units of the first measurement data and the second measurement data based on metadata associated with the first name and metadata associated with the second name; The learning unit performs a learning process for the learning model using the learning data set including the first measurement data and the second measurement data processed by the data processing unit. The learning system according to claim 2 .

4. an evaluation acquisition unit that acquires an evaluation value related to the measurement data; a presentation unit that presents the name associated with the measurement data whose evaluation value is higher than a threshold; The learning system of claim 1 .

Images