JP7252156B2 - LEARNING DATA GENERATION DEVICE AND LEARNING DATA GENERATION METHOD - Google Patents

Description

The present invention relates to a learning data generation device and a learning data generation method.

When implementing a machine learning system, it is necessary to prepare effective learning data in order to ensure the accuracy of the machine learning model.

As a technique for generating learning data, for example, Patent Literature 1 describes a learning data generation device for learning a determination device using a neural network. The learning data generation device changes the data values of the collected time-series data, changes the time interval of each data of the time-series data, adds distortion to the time-series data, and adds noise to the time-series data. .

Further, Patent Literature 2 describes a technique for generating data that contributes to improvement of learning by processing the learning data when the number of learning data is small. Specifically, a neural network learning device extracts features from learning data using a neural network under learning, generates adversarial features from the extracted features using a neural network under learning, and performs adversarial features with the learning data. The recognition result of the neural network is calculated using the characteristic features and the neural network is trained so that the recognition result approaches the desired output.

Further, Patent Literature 3 describes an anomaly detection system configured for the purpose of promptly detecting an anomaly in the state of an object to be monitored. The anomaly detection system collects observation data for the monitored object, stores it as time series observation data, classifies the observed data into either training data or verification data, and converts the linear state space of the monitored object based on the training data. Identify the model parameters of the model, calculate the estimated value of the probability distribution of the monitored state variables using the model parameters and verification data as input, calculate the degree of abnormality of the monitored target based on the estimated value, and collect the observation data Newly collected observation data is added to the time-series observation data, and when the number of time-series observation data is larger than the threshold value, the earliest collected observation data is discarded.

JP 2019-87106 A WO2018/167900 JP 2019-191836 A

R. B. Cleveland, 3 others, "STL: a seasonal-trend decomposition procedure based on loess", [online], 1990, Journal of official statistics, [searched January 31, 2020], Internet <URL:https:/ /www.wessa.net/download/stl.pdf>

In order to realize a machine learning system that performs inference processing such as predictive diagnosis and anomaly detection based on time-series data, it is necessary to ensure the accuracy of the machine learning model that performs the above inference processing. You need to prepare well. In addition, in order to ensure the accuracy of the machine learning model, it is necessary to prepare time-series data for the required period as learning data.

However, neither Patent Literature 1 nor Patent Literature 2 discloses any technology for generating time-series data for the required period. In addition, the technique described in Patent Document 3 needs to collect observation data for the monitored object, and cannot cope with cases where observation data is not obtained, such as when a machine learning system is introduced.

An object of the present invention is to provide a learning data generation device and a learning data generation method capable of efficiently providing learning data with appropriate content for a required period.

One of the present inventions for achieving the above object is a learning data generation device configured using an information processing device for generating learning data used for learning a machine learning model, comprising: Synthetic data, which is time-series data of a period corresponding to a required period, is generated by concatenating a plurality of replicated data, which are data obtained by replicating source data, which is series data, and applying noise to each of the replicated data. and a learning data generating unit that generates learning data using the artificial data.

In addition, the problems disclosed by the present application and their solutions will be clarified by the description of the mode for carrying out the invention and the drawings.

ADVANTAGE OF THE INVENTION According to this invention, the learning data of the content suitable about the required period can be efficiently provided.

1 is a diagram showing a schematic configuration of a machine learning system; FIG. It is an example of a configuration of an information processing device. It is a figure which shows the main functions with which a learning data generation apparatus is provided. 4 is a flowchart for explaining learning data generation processing; It is a flow chart explaining the details of artificial data generation processing. It is the figure which showed typically the data produced|generated in the execution process of an artificial data production|generation process. It is an example of origin data. This is an example of artificial data. FIG. 11 is a flowchart illustrating details of learning data period setting processing; FIG. It is an example of observation data. 4 is a flowchart for explaining the details of learning data generation processing; It is an example of learning data. It is a flow chart explaining artificial data generation processing in a 2nd embodiment. FIG. 10 is a diagram schematically showing data generated in the process of executing artificial data generation processing in the second embodiment; It is an example of origin data. It is an example of intermediate data. It is an example of copy source data. This is an example of artificial data.

An embodiment of the present invention will be described below with reference to the drawings. In the following description, the same reference numerals may be assigned to components having the same or similar functions, and redundant description may be omitted. Also, in the following description, the letter "S" attached before the reference sign means a processing step. Also, although the term "learning data" is used in the following description, it has the same meaning as "training data." Learning data used for so-called supervised machine learning includes so-called label information, but for the sake of simplicity, the present embodiment omits explanation and examples of labels. Also, in the following description, the period may be specified by date and time, or may be specified by only days or only hours.

[First embodiment]
FIG. 1 shows a schematic configuration of an information processing system (hereinafter referred to as "machine learning system 1") to which the learning data generation device 100 shown as the first embodiment is applied. As shown in the figure, the machine learning system 1 includes an inference device 2 and a learning data generation device 100 .

The inference device 2 has functions of a learning processing unit 21 that performs learning of the machine learning model 23 using learning data 114, which is time-series data, and an inference processing unit 22 that performs inference processing using the machine learning model 23. . The inference processing unit 22 performs inference processing by inputting observation data 113, which is time-series data, into the machine learning model 23, and outputs the result as an inference result 7. FIG. The machine learning model 23 performs, for example, inference processing for predictive diagnosis, abnormality detection, etc. based on time-series data.

The learning data generation device 100 generates learning data 114 based on source data 111 and observation data 113, which are time-series data. The generated learning data 114 is input to the inference device 2 via communication or a recording medium.

FIG. 2 shows an example of the information processing device 10 used for configuring the inference device 2 and the learning data generation device 100 . As shown in the figure, the illustrated information processing apparatus 10 includes a processor 11 , a main storage device 12 , an auxiliary storage device 13 , an input device 14 , an output device 15 and a communication device 16 . These are communicably connected via a communication means such as a bus.

The information processing apparatus 10 is realized using virtual information processing resources provided using virtualization technology, process space separation technology, etc., such as a virtual server provided by a cloud system, for example. good too. Further, all or part of the functions of the information processing apparatus 10 may be implemented by a service provided by a cloud system via an API (Application Programming Interface) or the like, for example. Further, for example, the learning data generation device 100 may be configured using a plurality of information processing devices 10 that are communicably connected. Software such as an operating system, file system, DBMS (database management system) (relational database, NoSQL, etc.) may be installed in the information processing apparatus 10, for example.

The processor 11 is, for example, a CPU (Central Processing Unit), MPU (Micro Processing Unit), GPU (Graphics Processing Unit), AI (Artificial Intelligence) chip, FPGA (Field Programmable Gate Array), ASIC (Application Specific Integrated Circuit), etc. is configured using

The main storage device 12 is a device that stores programs and data, and is, for example, a ROM (Read
Only Memory), RAM (Random Access Memory), nonvolatile memory (NVRAM (Non Volatile RAM)), and the like.

The auxiliary storage device 13 is, for example, an SSD (Solid State Drive), a hard disk drive, an optical storage device (CD (Compact Disc), DVD (Digital Versatile Disc), etc.), a storage system, an IC card, an SD card, or an optical recording device. They are a read/write device for a recording medium such as a medium, a storage area for a virtual server, and the like. Programs and data can be read from the auxiliary storage device 13 via a recording medium reading device or the communication device 16 . Programs and data stored (stored) in the auxiliary storage device 13 are read out to the main storage device 12 at any time.

The input device 14 is an interface that receives input from the outside, and includes, for example, a keyboard, mouse, touch panel, card reader, voice input device, and the like. The output device 15 is an interface for outputting various information such as processing progress and processing results.

The output device 15 includes, for example, a display device (liquid crystal monitor, LCD (Liquid Crystal Display), projector, etc.) that visualizes the above various information, a device that converts the above various information into sound (audio output device (speaker, etc.)), It is a device (printing device, etc.) that converts the above various information into characters.

The input device 14 and the output device 15 constitute a user interface. For example, the information processing device 10 may be configured to input/output information to/from other devices (smartphone, tablet, notebook computer, various mobile information terminals, etc.) via the communication device 16 .

Communication device 16 implements communication with other devices. The communication device 16 is a wireless or wired communication interface that realizes communication with other devices via a communication network. ) module, serial communication module, and the like. Next, functions of each device will be described.

FIG. 3 shows main functions of the learning data generating device 100. As shown in FIG. As shown in the figure, the learning data generation device 100 includes a storage unit 110, an observation data acquisition unit 120, a generation source data acquisition unit 130, an artificial data generation unit 140, a learning data period setting unit 150, a learning data generation unit 160, and each function of the learning data output unit 170 . These functions are realized by the processor 11 of the information processing device 10 constituting the learning data generation device 100 reading out and executing a program stored in the main storage device 12 of the information processing device 10, or by executing the program stored in the information processing device 10. 10 is implemented by hardware (FPGA, ASIC, AI chip, etc.).

Among the functions described above, the storage unit 110 stores and manages the generation data 111 , the artificial data 112 , the observation data 113 , and the learning data 114 . The storage unit 110 stores and manages each data as, for example, a database table provided by a DBMS or a file provided by a file system.

The source data 111 is data used to generate the artificial data 112 . Artificial data 112 is data used to generate learning data 114 . The observation data 113 is data input to the machine learning model 23 after the machine learning system 1 starts actual operation, for example. The learning data 114 is data generated based on the artificial data 112 and the observation data 113 and is data used for learning (training) of the machine learning model 23 .

Among the functions shown in FIG. 3, the observation data acquisition unit 120 acquires the observation data 113 from the inference device 2 via communication or a recording medium. The storage unit 110 stores observation data 113 acquired by the observation data acquisition unit 120 .

The source data obtaining unit 130 obtains or generates the source data 111 . For example, the origin data acquisition unit 130 receives the origin data 1 from the user via a user interface.
11 is obtained. The origin data acquisition unit 130 generates the origin data 111 based on the observation data 113, for example. The user may generate source data 111 by editing observation data 113 via a user interface. The storage unit 110 stores the source data 111 acquired or generated by the source data acquisition unit 130 .

Artificial data generator 140 generates artificial data 112 based on generation source data 111 . Storage unit 110 stores artificial data 112 generated by artificial data generation unit 140 .

The learning data period setting unit 150 performs processing related to setting the period of the learning data 114 (from the start point to the end point of the learning data; hereinafter referred to as “learning data period”). The learning data period setting unit 150 receives, for example, information regarding the setting of the learning data period from the user via the user interface.

The learning data generator 160 generates the learning data 114 based on the artificial data 112 and the observation data 113 .

The learning data output unit 170 outputs the learning data 114 generated by the learning data generation unit 160 . The output learning data 114 is input to the inference device 2 via communication or a recording medium.

FIG. 4 is a sequence diagram illustrating a process performed by the learning data generating device 100 when generating the learning data 114 (hereinafter referred to as "learning data generating process S400"). The learning data generation processing S400 will be described below with reference to FIG. Note that at the start of the process shown in the figure, the storage unit 110 already stores the observation data 113 acquired by the observation data acquisition unit 120 and the generation source data 111 acquired or generated by the generation source data acquisition unit 130. It is assumed that there is

As shown in the figure, first, the artificial data generation unit 140 reads the generation source data 111 stored in the storage unit 110 (S411).

Subsequently, the artificial data generation unit 140 designates the length of the period of the learning data 114 that the user intends to generate (hereinafter referred to as the “request period”) via the user interface, The input of the number of cycles (hereinafter referred to as "the number of cycles") and the variance σ̂2 used to generate noise given to the artificial data 112 is accepted (S412).

Subsequently, the artificial data generation unit 140 generates the artificial data 112 based on the read source data 111 and the received request period, number of cycles, and variance σ^2 (hereinafter referred to as “artificial data generation processing S413 ) is performed (S413).

Subsequently, the learning data period setting unit 150 reads the observation data stored in the storage unit 110 (S421).

Subsequently, the learning data period setting unit 150 sets the period of the generation source data 111 read by the artificial data generation unit 140 in S412, the period of the artificial data 112 generated by the artificial data generation unit 140 in S413, and the observation read out in S421. The period of data 113 is acquired (S422).

Subsequently, the learning data period setting unit 150 acquires the requested period accepted by the artificial data generating unit 140 in S412 (S423).

Subsequently, the learning data period setting unit 150 sets the learning data period based on each period acquired in S422 and the requested period acquired in S423 (hereinafter referred to as "learning data period setting process S424"). (S424).

Subsequently, the learning data generation unit 160 reads the generation source data 111 read by the artificial data generation unit 140 in S411, the artificial data 112 generated by the artificial data generation unit 140 in S413, and the learning data period setting unit 150 in S421. Learning data 114 is generated for the learning data period set by the learning data period setting process S424 based on the observed data 113 obtained (S431).

After that, the learning data output unit 170 outputs the generated learning data 114 . The output learning data 114 is transmitted (provided) to the learning processing unit 21 of the inference device 2 via communication or a recording medium.

FIG. 5 is a flowchart for explaining the details of the artificial data generation processing S413 shown in FIG. FIG. 6 is a diagram schematically showing data generated in the process of executing the artificial data generation process S413. The artificial data generation unit 140 duplicates the generation source data 111 by the number of cycles corresponding to the request period received in S412, replaces the date and time of the generation source data 111 with an appropriate date and time, and adds noise to the observed value. Thus, the artificial data 112 is generated. The artificial data generation processing S413 will be described below with reference to FIGS. 5 and 6. FIG.

FIG. 7 shows an example of the originating data 111. As shown in FIG. The artificial data generation processing S413 will be described below using the generation source data 111 shown in the figure as an example. As shown in the figure, the example generation source data 111 includes a plurality of entries (records) having date/time 701 and observation value 702 items.

Among the above items, in the date and time 701, the date and time when the value of the observed value 702 was obtained is set. The value of the date and time 701 is also used as an identifier for uniquely identifying each entry. An observed value is set in the observed value 702 . Although time-series data may contain categorical variable information, observations are assumed to be quantitative variable information unless otherwise specified. The observed value 702 is, for example, a value itself (raw data) obtained from a sensor device or the like, or a value obtained by processing (addition, subtraction, multiplication, division, tabulation processing, statistical processing, etc.) values obtained from a plurality of observation targets. be. The above values are, for example, the amount of communication and the operating rate when the observation target is an information communication system. Further, the above value is, for example, a value "total traffic" obtained by summing two observation target values of "upstream traffic" and "downstream traffic". Further, for example, the above values are obtained by calculation based on an observed value at a certain point in time and an observed value at another point in time. Further, for example, the above value is the difference between the previous traffic volume and the current traffic volume (amount of change in traffic over time).

The exemplified generation source data 111 consists of information recorded at intervals of 10 minutes from 00:00:00 on November 15, 2019 to 00:00:00 on November 22, 2019. FIG. A) is data for one cycle. In the following description, it is assumed that the generation source data 111 received in S412 has a cycle of one week and includes 1008 entries per cycle. It is also assumed that 28 weeks has been accepted as the requested period in S412.

As shown in FIG. 5, the artificial data generation unit 140 first generates the minimum multiple of the period of one cycle of the generation source data 111 (hereinafter referred to as "minimum number of cycles"), which is longer than the requested period received in S412. ) is obtained (S501).

Subsequently, the artificial data generation unit 140 obtains a value obtained by subtracting the number of cycles included in the generation source data 111 from the obtained minimum number of cycles, and divides the obtained value by the number of cycles included in the source data 111. A value obtained by rounding up the value after the decimal point is set as the number of replications (S502). The reason why the number of cycles included in the generation source data 111 is subtracted from the minimum number of cycles obtained in S501 is to exclude the source data 111 of the replication source from the number of replications. Since the number of cycles of the illustrated source data 111 is 1 and the requested period is 28 weeks, 27 is obtained as the number of replications in this example. Note that the number of times of duplication may be acquired by a method other than the above method. For example, the user may specify the number of times of duplication via a user interface.

Subsequently, the artificial data generating unit 140 generates data (hereinafter referred to as “duplicate data”) by duplicating the original data 111 by the number of times of duplication acquired in S502 (S503).

Subsequently, the artificial data generation unit 140 sequentially assigns natural numbers starting from 1 to each replicated data. The storage unit 110 stores the number assigned to each replicated data (hereinafter referred to as "replicated number") in association with each replicated data (S504).

Subsequently, the artificial data generating unit 140 generates data (hereinafter referred to as “primary artificial data”) by concatenating the duplicate data in reverse order of the assigned duplicate numbers (S505).

Subsequently, the artificial data generation unit 140 generates data obtained by copying the value of the date and time 701 of each entry of the generation source data 111 (hereinafter referred to as “reference date and time”) to each entry of the primary artificial data generated in S507. (S506).

Subsequently, the artificial data generation unit 140 updates the reference source date and time of each entry of the provided primary artificial data to a value that predates the reference date and time (for example, date and time t in FIG. The date and time of each entry is generated by going back by the date and time obtained by multiplying the period of 111 and the copy number of each entry (S507). For example, the date and time after the change of 00:00:00 on November 15, 2019 in the replicated data with the replication number 27 is 00:00:00 on May 10, 2019, which is 27 weeks earlier. The primary artificial data generated by executing S507 is as shown in FIG. 6(B).

Subsequently, the artificial data generation unit 140 generates white noise for the artificial data period using the variance σ̂2 received in S412 of FIG. 4 (S508). The white noise generated by executing S508 is as shown in FIG. 6(C).

Subsequently, the artificial data generation unit 140 generates the artificial data 112 by adding the white noise generated in S508 as a variation value to the primary artificial data (S509). The artificial data 112 generated by executing S509 is as shown in FIG. 6(D).

FIG. 8 shows an example of the artificial data 112 generated by the artificial data generation processing S413. As shown in the figure, the artificial data 112 illustrated includes a plurality of entries each having a date/time 811, an observed value 812, a reference source observed value 813, a variation value 814, a reference source date/time 815, and a replication number 816. .

Among the above items, the date and time generated in S507 is set in the date and time 811 . Note that the value of the date and time 811 is also used as an identifier that uniquely identifies each entry. In the observed value 812, the observed value of the artificial data 112 generated in S509 at the date and time is set. The observed value 702 of the generation source data 111 corresponding to the date and time is set in the reference source observed value 813 . The variation value 814 is set to the white noise value corresponding to the date and time generated in S508. The reference source date and time 815 is set with the date and time 701 of the generation source data 111 corresponding to the date and time. The value of the reference date and time 815 indicates that the entry is based on the entry of the generation source data 111 of the date and time of the value of the reference date and time 815 . The replication number 816 is set with the replication number assigned in S504.

FIG. 9 is a flowchart for explaining the learning data period setting process S424 shown in FIG. The learning data period setting process S424 will be described below with reference to FIG. Note that if the observation data 113 for a period overlapping the requested period acquired in S423 has already been acquired, the artificial data generation unit 140 sets the learning data period so that the observation data 113 is preferentially adopted as the learning data 114. set.

FIG. 10 is an example of observation data 113 used in the following description. As shown in the figure, the illustrated observation data 113 includes a plurality of entries having date and time 1011 and observation value 1012 items. In the date and time 1011 of the above items, the date and time when the observation value of the entry was acquired is set. The observed value 1012 is set with an observed value actually obtained from the observation target.

As shown in FIG. 9, the learning data period setting unit 150 first confirms whether or not the storage unit 110 stores the observation data 113 (whether or not the learning data generation device 100 has acquired the observation data 113). (S901). When the storage unit 110 stores the observation data 113 (S901: YES), the learning data period setting unit 150 sets the end point of the period of the observation data 113 as the end point t_end of the learning data period (S902). . After that, the process proceeds to S904. On the other hand, if the storage unit 110 does not store the observation data 113 (S901: NO), the learning data period setting unit 150 sets the end point of the period of the generation source data 111 as the end point t_end of the learning data period. (S903). After that, the process goes to S90
Proceed to 4.

In S904, the learning data period setting unit 150 sets a date and time (hereinafter referred to as , called “provisional start time tmp_Tstart”).

Subsequently, the learning data period setting unit 150 compares the period of the artificial data 112, the period of the source data 111, and the period of the observation data 113 with the provisional start time tmp_Tstart (S905). If the provisional start time tmp_Tstart is during the period of the artificial data 112 (S905: during the artificial data period), the provisional start time tmp_Tstart is set as the start time t_start of the learning data period (S906). On the other hand, if the temporary start time tmp_Tstart is during the period of the source data 111 or the period of the observation data 113 (S905: during the period of the source data or observation data), the source data 111 start point is set (S907)
.

By the above processing, the start time t_start and the end time t_end of the learning data period are set, and the setting of the learning data period is completed. By the processing of S902, S903, and S907, in the learning data generation processing S431, the observation data 113 or the original data 111 are adopted as the learning data 114 with priority over the artificial data 112.

FIG. 11 is a flowchart for explaining the learning data generation processing S431 shown in FIG. The learning data generation processing S431 will be described below with reference to FIG.

First, the learning data generation unit 160 acquires from the storage unit 110 the observation data 113, the source data 111, and the artificial data 112 of the period overlapping with the learning data period set by the learning data period setting process S424 (S1101 to S1103).

Subsequently, the learning data generation unit 160 generates intermediate consolidated data by connecting the acquired generation source data 111 and the acquired artificial data 112 in the time series direction (connecting the artificial data 112 and the generation source data 111 in chronological order). (S1104).

Subsequently, the learning data generating unit 160 generates learning data 114 by connecting the intermediate connected data and the obtained observation data 113 in the time-series direction (S1105). Note that when there is observation data 113 corresponding to the entire learning data period, all of the learning data 114 is based on the observation data 113 . Also, when the observation data 113 overlaps a part of the learning data period, the period from the start of the learning data period to the start of the observation data 113 in the entire period of the learning data 114 is intermediate concatenated data. Observation data 113 is used from the start time of 113 to the end time of the learning data period. In this way, when the observation data 113 exists, the observation data 113 is preferentially adopted as the learning data 114. Therefore, after the actual operation of the machine learning system 1 is started, the observation data which is the data actually acquired 113 as the learning data 114, it is possible to make an early transition to an operational state in which learning is performed.

FIG. 12 shows an example of the learning data 114 generated by the learning data generation processing S431. The illustrated learning data 114 includes a plurality of entries having date/time 1201 and observation value 1202 items. Among the above items, the date and time 1201 is set based on any of the values of the date and time 811 of the artificial data 112 , the date and time 701 of the source data 111 , and the date and time 1011 of the observation data 113 . The observed value 1202 is set to an observed value based on any one of the observed value 812 of the artificial data 112 , the observed value 702 of the generator data 111 , and the observed value 1012 of the observed data 113 .

As described above, according to the learning data generation device 100 of the first embodiment, even if it is difficult to prepare learning data for a period necessary to ensure the accuracy of the machine learning model 23, Effective learning data can be efficiently generated and provided without troubling the user.

In addition, since the learning data generation device 100 generates the learning data 114 using the artificial data 112 in which noise is individually added to each replicated data, the machine learning model 23 has diversity expected to suppress over-learning. The learning data 114 can be generated, and the inference accuracy of the machine learning model 23 can be improved. Also, by adding white noise to the artificial data 112, it is possible to reproduce fluctuations that are close to actual fluctuations. can be done.

Further, as shown in FIG. 9, in the learning data period setting process S424, the learning data period is set so that the observation data 113 is preferentially adopted over the learning data 114, so that the actual operation of the machine learning system 1 can be performed. After the start, it is possible to quickly transition to an operational state in which learning is performed using only the observation data 113 that is actually acquired data as the learning data 114 . Therefore, the inference accuracy of the inference device 2 can be improved early after the start of the actual operation.

In addition, the learning data generation device 100 can generate artificial data 112 of a period earlier than the generation source data 111, and generates the artificial data 112 so as not to overlap with the period of the newly acquired observation data 113. For example, it is possible to avoid complicated processing such as replacing the artificial data 112 with the observed data 113 .

In the above description, the case of generating the artificial data 112 for a period earlier than the generation source data 111 is illustrated, but the artificial data 112 for a period later than the generation source data 111 may be generated. As a result, for example, it is possible to generate learning data 114 for a desired future period using past time-series data for a period considered to best reflect actual behavior as generation source data 111 . In this case, for example, in S504 of FIG. 5, a negative integer starting from −1 is assigned to each replicated data to be the artificial data 112 in the future period, and a positive number assigned to each replicated data in the past period is assigned. The sum of the replication number and the absolute value of the negative replication number is made to match the number of replications acquired in S502 of FIG. By doing so, the date/time (period) information can be easily calculated simply by adding the value obtained by multiplying the period of the copy data by the copy number in S507 to the reference date/time.

[Second embodiment]
Next, a second embodiment will be described. The learning data generation device 100 of the second embodiment generates artificial data 112 based on components (trends, periodic fluctuations, and residuals to be described later) obtained by decomposing the source data 111 . The basic configuration of the machine learning system 1 of the second embodiment and the flow of processing executed in the machine learning system 1 are basically the same as the machine learning system 1 of the first embodiment described with FIGS. , but a part of the function of the artificial data generation unit 140 is different. The following description will focus on portions that differ from the first embodiment.

FIG. 13 is a flowchart for explaining artificial data generation processing S413 shown as the second embodiment. FIG. 14 is a diagram schematically showing data generated in the process of executing the artificial data generation process S413. FIG. 15 is an example of generation source data 111 used in the following description. The artificial data generation processing S413 of the second embodiment will be described in detail below with reference to these figures.

As shown in FIG. 14A, the example generation source data 111 has a small period Tp (=1 day) and a large period T (=7 days), 00:00:00 on November 15, 2019. to 23:50:00 on Nov. 22, 2019 for 8 days (eight small periods Tp (large period 7 days x 1 + small period 1 day)) at 10-minute intervals. In the following description, t is the starting point of the generation source data 111 . Also, in the following description, the request period received in S412 of FIG. 4 is assumed to be 28 weeks. Also, in S412 of FIG. 4, it is assumed that one cycle (one large cycle) is received as the number of cycles of the generation source data 111 .

As shown in FIG. 13, the artificial data generator 140 first receives input of a short period Tp (1 day) and a large period T (7 days) through the user interface (S1301).

Subsequently, the artificial data generation unit 140 obtains the minimum value (minimum number of cycles) of the multiple of the period of one cycle of the generation source data 111, which is longer than the requested period (28 weeks) received in S412 (S1302). .

Subsequently, the artificial data generation unit 140 obtains a value obtained by subtracting the number of cycles included in the generation source data 111 from the minimum number of cycles obtained in S1301, A value obtained by rounding up the value obtained by dividing by the number after the decimal point and adding 1 is set as the number of times of replication (S1303). Note that the reason for adding 1 is that the period of the generated artificial data 112 satisfies the required period due to the time difference (Tp/2, which will be described later) caused by obtaining the trend, which will be described later, from the moving average of the generation source data 111. because it may disappear. In this example, the cycle number of the generation source data 111 is 1, and the request period received in S412 is 28 weeks, so 28 is obtained as the number of times of duplication.

Subsequently, the artificial data generation unit 140 decomposes the generation source data 111 into constituent elements (trend, periodic fluctuation, residual) using the short period Tp as the fluctuation period (S1304). Here, the term "trend" refers to a component representing long-term fluctuations in time-series data (Trend component). Periodic fluctuation refers to a seasonal component that periodically appears in time-series data at regular intervals. Residuals refer to small fluctuation elements (reddual components) that remain after removing trends and periodic fluctuations in time-series data. In this embodiment, the above decomposition is performed by STL (Seasonal-Trend Decomposition
Procedure Based on Loess), but the decomposition method is not necessarily limited.

FIG. 14(B) shows components obtained by decomposing the generator data 111 of FIG. 14(A). In the figure, (B-1) is the trend, (B-2) is the periodic variation, and (B-3) is the residual.

FIG. 16 shows the data obtained in S1304 (hereinafter referred to as "intermediate data 1600"). As shown in the figure, the intermediate data 1600 includes a plurality of entries having date/time 1601, observed value 1602, trend 1603, periodic variation 1604, and residual 1605 items. In the figure, "-" indicates lack of data. The date and time 1601 and the observed value 1602 correspond to the date and time 1201 and the observed value 1202 in the source data 111 . A trend 1603, a periodic fluctuation 1604, and a residual 1605 are set to values indicating the trend, periodic fluctuation, and residual, which are the components of the observed value 1602 obtained in S1304. Note that both the trend 1603 and the residual 1605 lack values for half the period (=Tp/2) of the short period specified when executing the STL at both ends of the period. In this example, the period from 00:00:00 on November 15, 2019 to 11:50:00 on November 15, 2019 and from 12:00:00 on November 22, 2019 to the year 2019 The values of trend 1603 and residual 1605 are missing in the period up to 23:50:00 on November 22nd.

Returning to FIG. 13, subsequently, the artificial data generation unit 140 determines the total value of the trend 1603 and the periodic fluctuation 1604 (hereinafter referred to as , referred to as “replication source observation values”) (S1305). The process of S1305 corresponds to the process of synthesizing the trend shown in (B-1) and the periodic fluctuation shown in (B-2) in FIG. By executing this process, the data shown in FIG. 14C (hereinafter referred to as "copy source data 1700") is obtained.

FIG. 17 shows an example of duplication source data 1700 . As shown in the figure, source data 1700 includes multiple entries having items of date and time 1701 , observed value 1702 , trend 1703 , periodic variation 1704 , residual 1705 , and source observed value 1706 . Among the above items, in the date and time 1701, the value of the date and time 1601 of the entry having the value of the trend 1703 among the entries of the intermediate data 1600 is set. In the observation value 1702, the value of the observation value 1602 corresponding to the value of the date and time 1701 among the entries of the intermediate data 1600 is set. The value of the trend 1603 corresponding to the value of the date and time 1701 among the entries of the intermediate data 1600 is set to the trend 1703 . The value of the periodic variation 1604 corresponding to the value of the date and time 1701 among the entries of the intermediate data 1600 is set in the periodic variation 1704 . The value of the residual 1605 corresponding to the value of the date and time 1701 among the entries of the intermediate data 1600 is set to the residual 1705 . A duplicate source observed value 1706 is set to a value obtained by summing the value of the trend 1603 corresponding to the value of the date and time 1701 and the value of the periodic variation 1604 corresponding to the value of the date and time 1701 among the entries of the intermediate data 1600 .

Returning to FIG. 13, the artificial data generation unit 140 subsequently duplicates the source data 1700 by the number of times of duplication obtained in S1303 (S1306). Hereinafter, each replicated data is referred to as "replicated data".

Subsequently, the artificial data generation unit 140 sequentially assigns a natural number starting from 1 to each generated replicated data, and the storage unit 110 assigns a number assigned to each replicated data (hereinafter referred to as a “replicate number”). It is stored in association with each copy data (S1307). In the same manner as described in the first embodiment, by assigning a negative integer starting from -1 as a replication number in addition to a natural number starting from 1 in the process, artificial Data 112 may be generated. In this case, as in the case of the first embodiment, the total value of the positive replication number and the absolute value of the negative replication number is made to match the number of replications acquired in S1303.

Subsequently, the artificial data generation unit 140 generates primary artificial data by linking the duplicate data in the time-series direction in reverse order of the assigned duplicate numbers (S1308).

Subsequently, the artificial data generation unit 140 deletes the entry whose copy number is 1 and whose date and time is from t+T to t+T+Tp/2 from the primary artificial data (S1309).

Subsequently, the artificial data generation unit 140 gives each entry of the primary artificial data data obtained by duplicating the value of the date and time 701 of each entry of the generation source data 111 (hereinafter referred to as “reference date and time”). (S1310).

Subsequently, the artificial data generation unit 140 generates the date and time of each entry by updating the reference source date and time of each entry of the primary artificial data to a value that goes back from the reference date and time (S1311). As a result of this process, for example, the changed date and time of November 15, 2019, 12:00:00 in the replicated data with the replication number 2 is changed to November 1, 2019, 12:00:00, two weeks earlier. Become.

Subsequently, the artificial data generation unit 140 obtains the difference d in periodic fluctuations of the duplicated data before and after the boundary at the time of the boundary when connecting the duplicated data. Specifically, the artificial data generator 140 obtains the difference d between the value of the entry at the point of time t+Tp/2 and the value of the entry one point before (S1312). For example, in the example of the intermediate data 1600 in FIG. 16, t+Tp/2 is 12:00:00 on November 15, 2019, so 152 is obtained as the periodic variation 1604 on the same date and time. Also, 151 is obtained as the periodic variation 1604 at 11:50:00 on November 15, 2019, which is one point before the same date and time. Therefore, in this example, 1 is obtained as the difference d.

Subsequently, the artificial data generation unit 140 reflects the value obtained as the product of the difference d and the replication number of each entry for each entry of the primary artificial data in the observed value of each entry of the primary artificial data (for example, , addition or subtraction) (S1313). That is, primary artificial data are generated assuming that the trend (difference d) obtained from short-term data continued during the requested period. After executing the process, the primary artificial data becomes as shown in FIG. 14(D).

Subsequently, the artificial data generation unit 140 obtains the variance ŝ2 of the residuals obtained in S1304 (S1314).

Subsequently, the artificial data generation unit 140 generates white noise for the period corresponding to the period of the primary artificial data having the variance ŝ2 (S1315). White noise generated by executing the processing is as shown in FIG. 14(E).

Subsequently, the artificial data generating unit 140 generates a value obtained by duplicating the observed value of each entry (hereinafter referred to as “reference source observed value”) for the primary artificial data (S1316).

Subsequently, the artificial data generator 140 gives the white noise generated in S1315 as a variation value to the primary artificial data (S1317).

Subsequently, the artificial data generation unit 140 generates artificial data by setting the observed value of each entry in the artificial data 112 to a value obtained by adding each variation value to each reference source observed value (S1318). The artificial data 112 generated by executing the process is as shown in FIG. 14(F).

An example of artificial data 112 is shown in FIG. As shown in the figure, the artificial data 112 includes a plurality of entries having items of date/time 1801 , observed value 1802 , reference source observed value 1803 , variation value 1804 , reference source date/time 1805 , and replication number 1806 . The date and time generated in S1311 is set in the date and time 1801 of the above items. The value of date and time 1801 also functions as an identifier for uniquely identifying each entry. The observed value obtained in S1318 is set in the observed value 1802 . The reference source observation value generated in S1316 is set in the reference source observation value 1803 . The variation value 1804 is set to the white noise value added in S1317. The reference source date and time 1805 is set with the date and time given in S1310. The reference source date and time 1805 corresponds to the date and time 701 of the source data 111 and indicates that the entry is based on the entry of the date and time 701 of the source data 111 . The replication number 1806 is set to the replication number assigned in S1307.

As described above, the learning data generation device 100 of the second embodiment decomposes the source data 111, which is time-series data including two cycles, into trend, periodic fluctuation, and residual, and divides the trend and periodic fluctuation into and white noise is generated based on the variance ŝ2 obtained from the residual, and artificial data is generated from the original data and the white noise. Therefore, it is possible to generate learning data that reproduces a variation process that is close to the variation process that actually occurs, and use this to train the machine learning model 23, thereby improving the inference accuracy of the inference device 2. .

In addition, the learning data generation device 100 acquires the difference d (short-term trend) of periodic fluctuations of the replicated data before and after the point of time that becomes the boundary for connecting the replicated data, and obtains the product of the replication number and the difference d at the boundary. Artificial data 112 is generated by concatenating a plurality of replicated data while changing the observed value by the value. Therefore, the learning data 114 can be generated in consideration of long-term trends, and the machine learning model 23 can be learned with high accuracy.

Although the embodiments of the present invention have been described above, the present invention is not limited to the above-described embodiments, and includes various modifications. Further, for example, the above-described embodiments are detailed descriptions of the configurations for easy understanding of the present invention, and are not necessarily limited to those having all the described configurations. Also, part of the configuration of each embodiment can be added, deleted, or replaced with another configuration.

Further, each of the above configurations, functions, processing units, processing means, and the like may be realized by hardware, for example, by designing them in an integrated circuit. It can also be implemented by a software program code that implements each function shown in the embodiment. In this case, an information processing apparatus (computer) is provided with a storage medium storing the program code, and a processor included in the information processing apparatus reads the program code stored in the storage medium. In this case, the program code itself read from the storage medium implements the functions of the above embodiments, and the program code itself and the storage medium storing it constitute the present invention. Examples of storage media for supplying such program codes include hard disks, SSDs (Solid State Drives), optical disks, magneto-optical disks, CD-Rs,
Flexible disks, CD-ROMs, DVD-ROMs, magnetic tapes, non-volatile memory cards, ROMs, etc. are used.

In the above embodiments, the control lines and information lines are those considered necessary for explanation, and not all control lines and information lines are necessarily shown on the product. All configurations may be interconnected. In the above description, various types of information are illustrated in tabular form, but these information may be managed in forms other than the tabular form.

1 machine learning system, 2 inference device, 21 learning processing unit, 22 inference processing unit, 23 machine learning model, 100 learning data generation device, 110 storage unit, 112 artificial data, 111 generation source data, 113 observation data, 114 learning data , 120 observation data acquisition unit, 130 generation source data acquisition unit, 140 artificial data generation unit, 150 learning data period setting unit, 160 learning data generation unit, 170 learning data output unit, S400 learning data generation processing, S413 artificial data generation processing , S424 learning data period setting process, S431 learning data generation process

Claims (18)

A learning data generation device configured using an information processing device and generating learning data used for learning a machine learning model,
By connecting a plurality of replicated data, which are data obtained by replicating the source data, which is time-series data for a predetermined period, and adding noise to each of the replicated data, time-series data of a period corresponding to the required period is obtained. an artificial data generation unit that generates certain artificial data;
a learning data generation unit that generates learning data using the artificial data;
A learning data generation device comprising: The learning data generation device according to claim 1,
The requested period is the period received from the user via the user interface,
Learning data generator. The learning data generation device according to claim 1,
The artificial data generation unit generates the artificial data for a period earlier than the generation source data,
The learning data generation unit generates the learning data by linking the artificial data to the generation source data in time series.
Learning data generator. The learning data generation device according to claim 1,
The artificial data generation unit generates the artificial data for a future period from the generation source data,
The learning data generation unit generates the learning data by linking the artificial data to the generation source data in time series.
Learning data generator. The learning data generation device according to claim 1 or 2,
The learning data generation unit links observed data, which is time-series data actually input to the machine learning model during actual operation of a machine learning system that performs inference processing using the machine learning model, to the artificial data. generating the learning data by
Learning data generator. The learning data generation device according to claim 5,
The learning data generation unit generates the learning data by adopting the observation data with priority over the artificial data.
Learning data generator. The learning data generation device according to claim 6,
A learning data period setting unit that sets a learning data period that is a period from the start time to the end time of the learning data,
The learning data period setting unit sets the latest point in time of the observation data to the end point of the learning data period,
setting the start time to a time that is the required period of time before the end time;
Learning data generator. The learning data generation device according to claim 1,
The artificial data generation unit gives noise to each of the replicated data individually.
Learning data generator. The learning data generation device according to claim 1,
wherein the noise is white noise;
Learning data generator. The learning data generation device according to claim 1,
The artificial data generation unit decomposes the source data into component elements of a trend, a periodic variation, and a residual, and generates the artificial data based on the trend and the periodic variation among the components. ,
Learning data generator. The learning data generation device according to claim 10,
The artificial data generation unit generates the noise based on the variance of the residual, and adds the generated noise to the generated artificial data.
Learning data generator. The learning data generation device according to claim 10,
The artificial data generation unit obtains a difference in the periodic variation of the duplicated data before and after the boundary at a point of time that becomes a boundary when connecting the duplicated data, and generates a plurality of the duplicated data while reflecting the difference. generating said artificial data by concatenating;
Learning data generator. The information processing device
By connecting a plurality of replicated data, which are data obtained by replicating the source data, which is time-series data for a predetermined period, and adding noise to each of the replicated data, time-series data of a period corresponding to the required period is obtained. generating some artificial data;
generating learning data for use in learning a machine learning model using the artificial data;
A training data generation method that executes The learning data generation method according to claim 13,
The information processing device concatenates, to the artificial data, observed data, which is time-series data actually input to the machine learning model during actual operation of a machine learning system that performs inference processing using the machine learning model. generating the learning data by
A learning data generation method further comprising: The learning data generation method according to claim 14,
a step in which the information processing device generates the learning data by preferentially adopting the observation data over the artificial data;
A learning data generation method further comprising: The learning data generation method according to claim 13,
A step in which the information processing device decomposes the source data into components of a trend, a periodic variation, and a residual, and generates the artificial data based on the trend and the periodic variation among the components. ,
A learning data generation method further comprising: The learning data generation method according to claim 16,
a step in which the information processing device generates the noise based on the variance of the residual and adds the generated noise to the generated artificial data;
A learning data generation method further comprising: The learning data generation method according to claim 16,
The information processing device obtains a difference in the periodic variation of the duplicated data before and after the boundary at a point of time when the duplicated data is concatenated, and concatenates the duplicated data while sequentially adding the difference. generating said artificial data by moving
A learning data generation method further comprising:

Images