Data processing system, data processing program, and data processing method

Abstract

To reduce an error caused by a noise factor.SOLUTION: A data acquisition unit 101 that acquires pieces of time-series data respectively collected by the two or more data collection devices 1011; and a determination unit that determines, based on a difference between the pieces of time-series data acquired by the data acquisition unit 101 and respectively collected by the two or more data collection devices 1011, whether: This device is provided with a noise generation part 102 for generating pseudo noise and a data correction part 103 for correcting at least one piece of time series data among pieces of time series data acquired by a data acquisition part 101 and respectively collected by two or more data collection equipment 1011 by using the pseudo noise generated by the noise generation part 102.SELECTED DRAWING: Figure 1

Description

[Technical Field]

[0001] The present disclosure relates to a data processing system, a data processing program, and a data processing method. [Background technology]

[0002] BACKGROUND ART Conventionally, there is known a technique for detecting the occurrence of an abnormal physical condition or a sign thereof by collecting biometric information of a subject and constructing a learning model or logic using the collected biometric information. On the other hand, in order to build a learning model or logic that can accurately detect the occurrence of an abnormal physical condition in a subject or its signs, a large amount of data on abnormal physical conditions is required.

[0003] When collecting data on such large-scale abnormal physical conditions, errors may occur in the collected data due to differences in the devices that collect the data, differences in the parts of the human body that are collected (for example, whether it is the arm or the fingertip), differences in the biological information that is collected (for example, whether it is heart rate or pulse rate), differences in the environment in which the data is collected (for example, whether it is collected in a vehicle, in a hospital room, or in a shielded room), or differences in the individuals that are the subject of the data collection (movement, the person's attributes (wrinkles or beard, etc.), accessories (glasses, masks, etc.), or whether they are wearing makeup). If errors occur in the collected data, the tendency of changes in biometric information when an abnormality occurs or the feature values ​​used for detection may change, which may reduce the accuracy of detecting an abnormality in the physical condition.

[0004] In response to this, a technique is known in which data values ​​and time intervals of collected time-series data are changed, and distortion and noise are added to the time-series data, thereby increasing the amount of data for learning (see, for example, Patent Document 1). This technique makes it possible to generate a large amount of learning data from collected data, thereby improving the accuracy of judgment. [Prior art documents] [Patent documents]

[0005] [Patent Document 1] Japanese Patent Application Publication No. 2019-87106 Summary of the Invention [Problem to be solved by the invention]

[0006] However, the technology disclosed in Patent Document 1 does not take into account errors due to noise factors such as the information collected, the area collected, or the environment in which it is collected, and if an error occurs between the data used for learning and the data used for judgment, the accuracy of the judgment may decrease.

[0007] For example, in a device that uses heart rate as a basis for logic to determine abnormal physical conditions, if the determination is made based on pulse rate, errors are unlikely to occur in healthy people because the heart rate and pulse rate often match. However, when a person has a heart disease, the heart rate and pulse rate may not match, resulting in errors and reduced accuracy of the determination. Furthermore, even when blood pressure values ​​are used as a reference, blood pressure values ​​at the fingertip and the upper arm are not equivalent, so errors similarly occur and the accuracy of the determination decreases. In conventional technology, there are parts that add noise, but this is not added as noise to absorb differences, but rather is added simply to increase the variation in data during learning.

[0008] The present disclosure has been made to solve the above-mentioned problems, and aims to provide a data processing system that can reduce errors caused by noise factors. [Means for solving the problem]

[0009] The data processing system according to the present disclosure is characterized by including a data acquisition unit that acquires time series data collected by two or more data collection devices, a noise generation unit that generates pseudo-noise based on the difference between the time series data collected by the two or more data collection devices and acquired by the data acquisition unit, and a data correction unit that corrects at least one of the time series data collected by the two or more data collection devices and acquired by the data acquisition unit using the pseudo-noise generated by the noise generation unit. [Effects of the Invention]

[0010] According to the present disclosure, the above-described configuration makes it possible to reduce errors caused by noise factors. [Brief explanation of the drawings]

[0011] [Figure 1] 1 is a block diagram showing an example of the configuration of a data processing system according to a first embodiment. [Figure 2] 3 is a diagram illustrating an example of the operation of the state determination device according to the first embodiment. FIG. [Figure 3] 4 is a flowchart showing an example of operation of the data generating device according to the first embodiment. [Figure 4] 4 is a diagram showing a specific example of pseudo-noise generation by the data generating device according to the first embodiment. FIG. [Figure 5] 4 is a diagram showing a specific example of pseudo noise added by the data generating device according to the first embodiment. FIG. [Figure 6] 4 is a flowchart showing an example of the operation of the state determination device according to the first embodiment. [Figure 7] 4 is a flowchart showing an example of the operation of the state determination device according to the first embodiment. [Figure 8] FIG. 10 is a block diagram showing an example of the configuration of a data processing system according to a second embodiment. [Figure 9] FIG. 11 is a block diagram showing an example of the configuration of a data processing system according to a third embodiment. [Figure 10]10A and 10B are diagrams illustrating examples of hardware configurations of data processing systems according to embodiments 1 to 3. In FIG. DETAILED DESCRIPTION OF THE INVENTION

[0012] Hereinafter, embodiments will be described in detail with reference to the drawings. Embodiment 1 FIG. 1 is a block diagram showing an example of the configuration of a data processing system according to the first embodiment. The data processing system includes a data generating device 1 and a state determining device 2, as shown in FIG.

[0013] The data generating device 1 generates time-series data, which is then output to the state determining device 2 as output data. This data generating device 1 includes a data acquiring unit 101, a noise generating unit 102, and a data correcting unit 103, as shown in FIG.

[0014] The data acquisition unit 101 acquires time-series data collected by two or more data collection devices 1011, respectively.

[0015] The time-series data collected by the data collection device 1011 is, for example, time-series data indicating biometric information, brightness change information of a specific part in an image showing the user's face, the degree of eye opening and closing, the degree of eye opening, the angle of the face's direction, or the line of sight, etc. Examples of the biometric information include an electrocardiogram, heart rate, or blood pressure value.

[0016] The time series data collected by the data acquisition unit 101 and collected by two or more data collection devices 1011 is output as first collected data to the noise generation unit 102. The first collected data is time series data collected simultaneously (including substantially simultaneously) by the two or more data collection devices 1011, that is, multiple time series data that are time-synchronized.

[0017] Furthermore, at least one of the time series data collected by the data acquisition unit 101 and collected by two or more data collection devices 1011 is output as second collected data to the data correction unit 103. The second collected data is time series data collected at a timing different from that of the first collected data.

[0018] FIG. 1 shows a case where the data acquiring unit 101 acquires time-series data collected by two data collecting devices 1011a and 1011b. FIG. 1 also shows a case where time-series data collected by one data collection device 1011b collected by the data acquisition unit 101 is output to the data correction unit 103 as second collected data.

[0019] FIG. 1 also shows a case where the data collection device 1011 is provided inside the data acquisition unit 101. However, the present invention is not limited to this, and the data collection device 1011 may be provided outside the data acquisition unit 101 .

[0020] The noise generation unit 102 generates pseudo noise based on the difference between the time series data collected by two or more data collection devices 1011, which are acquired by the data acquisition unit 101. That is, the noise generation unit 102 generates pseudo noise based on the difference between the time series data included in the first collected data.

[0021] Here, errors occur in the time-series data collected by two or more data collection devices 1011 due to noise factors in the two or more data collection devices 1011. Note that differences in noise factors in the two or more data collection devices 1011 are, for example, individual differences between the two or more data collection devices 1011, differences in the information collected by the two or more data collection devices 1011, differences in the parts of the body where the data is collected, and differences in the environment where the data is collected.

[0022] The collected information includes, for example, heart rate and pulse rate. Areas from which data may be collected include, for example, the chest, face, arms, and fingers. The environments in which the data is collected include, for example, vehicles, hospital rooms, and shielded rooms.

[0023] Therefore, the noise generating unit 102 generates pseudo noise for absorbing errors due to noise factors based on the difference between the time-series data collected by the two or more data collecting devices 1011.

[0024] In addition, the noise generating unit 102 may generate pseudo-noise by taking into consideration not only the difference between the time series data collected by two or more data collection devices 1011 acquired by the data acquiring unit 101, but also information indicating the noise factors of each of the two or more data collection devices 1011. The information indicating the noise factor is at least one of information collected by the data collection device 1011, information on the part collected by the data collection device 1011, and information on the environment collected by the data collection device 1011. In particular, of the above, the information collected by the data collection device 1011 and information indicating the part collected by the data collection device 1011 are considered to be useful information for noise generation. In this way, the noise generation unit 102 generates noise after understanding the noise factors in the time series data collected by the data collection devices 1011, thereby making it possible to further improve the accuracy of generating pseudo-noise, especially when the number of data collection devices 1011 is three or more.

[0025] The pseudo-noise generated by the noise generating unit 102 may be, for example, Gaussian noise. In this case, for example, the noise generating unit 102 generates a Gaussian distribution based on the difference between two or more time series data collected by the data acquiring unit 101, and generates Gaussian noise from the generated Gaussian distribution to absorb errors, thereby generating pseudo-noise. Note that when the noise generating unit 102 generates a Gaussian distribution, for example, the difference between two or more pieces of time-series data, or the median, average, or standard deviation thereof is calculated as an error to generate the Gaussian distribution.

[0026] Furthermore, the pseudo-noise generated by the noise generating unit 102 is not limited to the Gaussian noise described above, and may be, for example, Brownian noise, which is an integral of Gaussian noise.

[0027] The noise generator 102 may also use machine learning to generate the pseudo-noise.

[0028] The data representing the pseudo noise generated by the noise generating unit 102 is output to the data correcting unit 103 as noise data.

[0029] The data correction unit 103 corrects at least one of the time series data collected by the two or more data collection devices 1011 and acquired by the data acquisition unit 101, using the pseudo noise generated by the noise generation unit 102. That is, the data correction unit 103 corrects the second collected data using the pseudo noise generated by the noise generation unit 102.

[0030] More specifically, the data correction unit 103 corrects the second collected data by adding the pseudo noise generated by the noise generation unit 102 to the second collected data to which the pseudo noise is to be added. In this case, for example, the data correction unit 103 may add the pseudo noise by adding the pseudo noise to the second collected data, may add the pseudo noise by multiplying the second collected data by the pseudo noise, or may add the pseudo noise by performing statistical processing on the second collected data using the pseudo noise.

[0031] In addition, the data correction unit 103 may generate spike noise based on the setting information and correct at least one of the time series data acquired by the data acquisition unit 101 and collected by two or more data collection devices 1011 using the spike noise. The spike noise is a noise that simulates a sudden change. The setting information is, for example, information on one or more of the maximum noise amount, the minimum noise amount, the frequency of noise occurrence, and the bias of the noise.

[0032] Furthermore, the data correction unit 103 may correct at least one of the time series data collected by the two or more data collection devices 1011 so as to use the time series data as data for constructing a learning model or logic. That is, the data correction unit 103 corrects and shapes the time series data into time series data that is easy to handle in the state determination device 2. At this time, for example, the data correcting unit 103 may change the sampling rate, change the unit, or calculate the feature amount for the second collected data.

[0033] Changing the sampling rate means, for example, changing the second collected data from 100 Hz (100 pieces of data per second) to 1000 Hz. Here, the data correction unit 103 performs, for example, linear interpolation, spline interpolation, or filling in with the previous value for the missing data or the data to be thinned out.

[0034] The change in the data unit is, for example, changing the second collected data from data in units of ms (1 / 1000 of a second) to data in units of s (1 second).

[0035] The calculation of the feature amount is calculation of the feature amount used in the condition determination device 2, such as calculation of the median, mean, or standard deviation of the second collected data, or calculation of various heart rate variability indices if the second collected data is biological information data. The various heart rate variability indices are information such as LF, HF, LF / HF, and TP obtained from the power spectrum distribution of the biological information data.

[0036] The time series data corrected by the data corrector 103 is output to the state determining device 2 as output data.

[0037] The state determination device 2 constructs a learning model or logic for state determination based on the time-series data generated by the data generation device 1, and performs state determination using the constructed learning model and logic. The state determined by the state determination device 2 is, for example, an abnormal physical condition such as heart disease or epilepsy, a high stress (high load) state, or a drowsy state.

[0038] 1, the state determination device 2 includes a feature acquisition unit 201, a learning unit 202 (construction unit), a model storage unit 203, and a state estimation unit 204. Here, an example is shown in which the state determination device 2 constructs a learning model, but the state determination device 2 may also employ a determination method (construction of logic) using a threshold value.

[0039] The feature acquisition unit 201 acquires feature amounts based on the time-series data generated by the data generation device 1. In this case, for example, the feature acquisition unit 201 may acquire the feature by calculating the feature based on the time-series data generated by the data generating device 1. The calculation of the feature by the feature acquisition unit 201 is the same process as the calculation of the feature performed by the data correcting unit 103 in the data generating device 1. Furthermore, for example, if the feature amount has been calculated by the data correction unit 103, the feature amount acquisition unit 201 may acquire this feature amount. In other words, in this case, the feature amount acquisition unit 201 does not need to calculate the feature amount again.

[0040] The data indicating the feature amount acquired by the feature amount acquisition unit 201 is output to the learning unit 202 .

[0041] The learning unit 202 uses the feature amounts acquired by the feature amount acquisition unit 201 to learn a learning model for state determination (state determination model). Data indicating the learning model obtained by the learning unit 202 is output to the model storage unit 203.

[0042] Here, the learning algorithm used by the learning unit 202 may be, for example, a known supervised learning algorithm. In the following, as an example, a case where a neural network is applied to the learning unit 202 will be described.

[0043] In this case, the learning unit 202 learns the state of the user by so-called supervised learning according to, for example, a neural network model. Here, supervised learning refers to a technique in which a learning device is provided with a set of input and result (label) data, which allows the device to learn the features of the learning data and infer the result from the input.

[0044] A neural network is composed of an input layer consisting of a plurality of neurons, an intermediate layer (hidden layer) consisting of a plurality of neurons, and an output layer consisting of a plurality of neurons. The number of intermediate layers may be one or more. For example, in the case of a three-layer neural network as shown in Figure 2, when multiple inputs are input to the input layers (X1-X3), the values ​​are multiplied by first weights (w11-w16) and input to the intermediate layers (Y1-Y2), respectively. The results are then multiplied by second weights (w21-w26) and output from the output layers (Z1-Z3). This output result varies depending on the values ​​of the first and second weights.

[0045] In the state determination device 2 according to the first embodiment, the neural network learns whether or not the user is in an abnormal state by so-called supervised learning in accordance with the learning data, which are the features acquired by the feature acquisition unit 201. That is, the neural network learns by inputting the features to the input layer and adjusting the first weight and the second weight so that the result output from the output layer approaches the calculated feature that indicates that the user is in an abnormal state. The learning unit 202 generates a learning model by performing the above-described learning, and outputs data indicating the learning model.

[0046] The model storage unit 203 stores data indicating the learning model obtained by the learning unit 202.

[0047] Examples of this model storage unit 203 include non-volatile or volatile semiconductor memory such as RAM (Random Access Memory), ROM (Read Only Memory), flash memory, EPROM (Erasable Programmable ROM), EEPROM (Electrically EPROM), magnetic disk, flexible disk, optical disk, compact disk, mini disk, or DVD (Digital Versatile Disc).

[0048] FIG. 1 shows a case where the model storage unit 203 is provided inside the state determination device 2. However, the present invention is not limited to this, and the model storage unit 203 may be provided outside the state determination device 2.

[0049] The state estimation unit 204 performs state estimation (estimation of whether the user is in an abnormal state) using the feature amount and the learning model obtained by the learning unit 202. At this time, the state estimation unit 204 extracts the learning model from the model storage unit 203 and performs state estimation.

[0050] The feature used by the state estimation unit 204 may be a feature acquired by the feature acquisition unit 201, a feature calculated by the data correction unit 103 in the data generation device 1, or a feature calculated by the feature acquisition unit 201 using information acquired from a data collection device provided outside the state determination device 2.

[0051] Data indicating the state estimation result by the state estimation unit 204 is output to the outside.

[0052] Each process in the data processing system according to the first embodiment may be performed by a terminal or by the cloud.

[0053] Next, an example of the operation of the data generating device 1 according to the first embodiment shown in Fig. 1 will be described with reference to Fig. 3. Here, it is assumed that the pseudo-noise is Gaussian noise. In an example of the operation of the data generating device 1 according to the first embodiment shown in FIG. 1, as shown in FIG. 3, the data acquiring unit 101 first acquires time series data collected by two or more data collecting devices 1011 (step ST101).

[0054] Next, the noise generating unit 102 generates Gaussian noise based on the difference between the time-series data collected by the two or more data collecting devices 1011, which are acquired by the data acquiring unit 101 (step ST102). In this case, the noise generating unit 102 may generate Gaussian noise by taking into consideration not only the difference between the time series data collected by two or more data collection devices 1011 acquired by the data acquiring unit 101, but also information indicating the noise factors in the two or more data collection devices 1011.

[0055] Next, the data correction unit 103 applies Gaussian noise generated by the noise generation unit 102 to at least one of the time series data (data to be corrected) among the time series data collected by two or more data collection devices 1011 and acquired by the data acquisition unit 101 (step ST103).

[0056] Next, the data generating device 1 selects whether to add spike noise (step ST104). The selection of whether to add spike noise by the data generating device 1 may be performed according to a predetermined rule or a user operation, for example.

[0057] In this step ST104, if the data generating device 1 selects to add spike noise (step ST104, YES), the data correcting unit 103 generates spike noise based on the setting information and adds the spike noise to the data to be corrected (step ST105).

[0058] On the other hand, if the data generating device 1 selects not to add spike noise in step ST104 (NO in step ST104), the sequence proceeds to step ST106.

[0059] Next, the data generating device 1 selects whether or not to change the sampling rate as the data correction (step ST106). The selection of the sampling rate change by the data generating device 1 may be performed according to a predetermined rule or in response to a user operation, for example.

[0060] In step ST106, if the data generating device 1 selects to change the sampling rate as the data correction (step ST106, YES), the data correcting unit 103 changes the sampling rate of the data to be corrected (step ST107).

[0061] On the other hand, in step ST106, if the data generating device 1 selects not to change the sampling rate as the data correction (step ST106, NO), the sequence proceeds to step ST108.

[0062] Next, the data generating device 1 selects whether or not to change the unit as data correction (step ST108). The data generating device 1 may select whether or not to change the unit according to a predetermined rule, or according to a user operation, for example.

[0063] In step ST108, if the data generating device 1 selects to change the unit as data correction (step ST108, YES), the data correcting unit 103 changes the unit of the data to be corrected (step ST109).

[0064] On the other hand, in step ST108, if the data generating device 1 selects not to change the unit as data correction (step ST108, NO), the sequence proceeds to step ST110.

[0065] Next, the data generating device 1 selects whether or not to calculate feature quantities as data correction (step ST110). The selection of feature quantity calculation by the data generating device 1 may be performed according to a predetermined rule or a user operation, for example.

[0066] If the data generating device 1 selects to calculate the feature amount as the data correction in step ST110 (YES in step ST110), the data correcting unit 103 calculates the feature amount based on the data to be corrected (step ST111), and then the sequence ends.

[0067] On the other hand, in step ST110, if the data generating device 1 selects not to calculate the feature amount as the data correction (step ST110, NO), the sequence ends.

[0068] Although FIG. 3 shows examples of adding Gaussian noise and spike noise, the present invention is not limited to these examples and various other noises may be added. Furthermore, in FIG. 3, the data correction is exemplified by changing the sampling rate, changing the data unit, and calculating the feature amount, but the present invention is not limited to these, and other corrections may also be performed.

[0069] Next, a specific example of the generation and addition of pseudo-noise by the data generating device 1 according to the first embodiment shown in FIG. 1 will be described with reference to FIGS.

[0070] For example, as shown in the left diagram of FIG. 4, suppose that the data acquisition unit 101 acquires, as time-series data for noise generation, heart rate data (first measurement value) obtained non-contact using a camera in a vehicle and heart rate data (second measurement value) obtained from an electrocardiogram in the vehicle. In this case, for example, as shown in Fig. 4, the noise generator 102 calculates the error by subtracting the second measurement value from the first measurement value, and generates a normal distribution (Gaussian distribution) of the error from the calculated error. This normal distribution of the error becomes a model of the pseudo-noise, and the difference in noise factors is included in this error.

[0071] Also, for example, as shown in the right diagram of FIG. 5, it is assumed that the data acquiring unit 101 acquires data of the heart rate (second measurement value) obtained from an electrocardiogram in a dark room as time-series data for adding noise. In this case, for example, as shown in Fig. 5, the data correction unit 103 adds pseudo noise obtained from a normal distribution of errors to the second measurement value. At this time, the value of the added pseudo noise is determined in accordance with the normal distribution of errors. This makes it possible to treat the second measurement value after adding noise as time-series data similar to the first measurement value. In the example of FIG. 5, addition is used as the method of adding noise, but the method is not limited to this, and multiplication or division may be performed using noise as a coefficient.

[0072] Next, an example of the operation of the state determining device 2 according to the first embodiment shown in FIG. 1 will be described with reference to FIGS. FIG. 6 is a flowchart showing an example of an operation (learning operation example) in which the state determination device 2 uses the time-series data generated by the data generation device 1 to obtain and store a learning model. FIG. 7 is a flowchart showing an example of an operation (an example of a state determination operation) in which the state determination device 2 performs state estimation using feature amounts and a learning model and outputs an estimation result.

[0073] In an example of a learning operation by the state determination device 2 according to the first embodiment shown in FIG. 1, first, as shown in FIG. 6, for example, the feature acquisition unit 201 acquires the time-series data generated by the data generation device 1 (step ST201).

[0074] Next, the feature amount acquiring unit 201 determines whether or not it is necessary to calculate a feature amount based on the acquired time-series data (step ST202). That is, the feature amount acquiring unit 201 determines whether or not the acquired time-series data is data indicating a feature amount.

[0075] In step ST202, if it is determined that calculation of the feature amount is necessary (step ST202, YES), the feature amount acquiring unit 201 calculates the feature amount based on the acquired time-series data (step ST203).

[0076] On the other hand, in step ST202, if the feature acquisition unit 201 determines that calculation of the feature is not necessary (step ST202, NO), the sequence proceeds to step ST204.

[0077] Next, the learning unit 202 uses the feature amounts acquired by the feature amount acquiring unit 201 to learn a learning model for state determination (step ST204).

[0078] Next, the model storage unit 203 stores data indicating the learning model obtained by the learning unit 202 (step ST205), after which the sequence ends.

[0079] In an example of a state determination operation by the state determination device 2 according to the first embodiment shown in FIG. 1, as shown in FIG. 7, for example, first, the feature acquisition unit 201 acquires the time-series data generated by the data generation device 1 (step ST301).

[0080] Next, the feature amount acquiring unit 201 determines whether or not it is necessary to calculate a feature amount based on the acquired time-series data (step ST302). That is, the feature amount acquiring unit 201 determines whether or not the acquired time-series data is data indicating a feature amount.

[0081] In step ST302, if it is determined that calculation of the feature amount is necessary (step ST302, YES), the feature amount acquiring unit 201 calculates the feature amount based on the acquired time-series data (step ST303).

[0082] On the other hand, in step ST302, if the feature acquisition unit 201 determines that calculation of the feature is not necessary (step ST302, NO), the sequence proceeds to step ST304.

[0083] Next, the state estimation unit 204 reads the learning model from the model storage unit 203 (step ST304).

[0084] Next, the state estimation unit 204 performs state estimation (estimation of whether the user is in an abnormal state) using the feature amount and the read learning model (step ST305).

[0085] Next, state estimation section 204 outputs data indicating the state estimation result to the outside (step ST306), after which the sequence ends.

[0086] As described above, in the data processing system according to the first embodiment, the noise generating unit 102 generates pseudo-noise based on the difference between the time series data collected by two or more data collecting devices 1011, and the data correcting unit 103 corrects the data using this pseudo-noise. As a result, the data processing system according to the first embodiment can absorb errors associated with differences in noise factors among the two or more data collection devices 1011, i.e., differences between the individual data collection devices 1011, differences in the information collected by the two or more data collection devices 1011, differences in the parts of the body collected, or differences in the environments in which the information is collected, thereby enabling the construction of a learning model or logic with high accuracy using large amounts of data. Furthermore, in the data processing system according to the first embodiment, by generating pseudo noise from actually collected time-series data, it becomes possible to generate realistic noise and absorb errors appropriately.

[0087] Furthermore, in the data processing system according to the first embodiment, the data correcting unit 103 performs data correction using spike noise. As a result, the data processing system according to the first embodiment can also simulate sudden noise, and can build a learning model or logic with higher accuracy using large amounts of data.

[0088] Furthermore, in the data processing system according to the first embodiment, the data correction unit 103 corrects the data so that it becomes data for constructing a learning model or logic. As a result, in the data processing system of embodiment 1, when constructing a learning model or logic, it is possible to treat the data as the same even if the data collection device 1011 is different, making it possible to construct a learning model or logic with high accuracy using large amounts of data.

[0089] As described above, according to this first embodiment, the data processing system includes a data acquisition unit 101 that acquires time series data collected by two or more data collection devices 1011, a noise generation unit 102 that generates pseudo noise based on the difference between the time series data collected by the two or more data collection devices 1011 and acquired by the data acquisition unit 101, and a data correction unit 103 that corrects at least one of the time series data collected by the two or more data collection devices 1011 and acquired by the data acquisition unit 101 using the pseudo noise generated by the noise generation unit 102. As a result, the data processing system according to the first embodiment can reduce errors caused by noise.

[0090] Furthermore, according to this embodiment 1, the noise generating unit 102 generates pseudo-noise based on the difference between the time series data collected by two or more data collecting devices 1011, which is acquired by the data acquiring unit 101, and information indicating the respective noise factors in the two or more data collecting devices 1011. Furthermore, according to this embodiment 1, the information indicating the noise factor is at least one of information collected by the data collection device 1011, information on the location collected by the data collection device 1011, or information on the environment collected by the data collection device 1011. As a result, the data processing system according to the first embodiment can further reduce errors caused by noise.

[0091] Furthermore, according to this embodiment 1, in the data processing system, the data correction unit 103 corrects at least one of the time series data collected by two or more data collection devices 1011 so as to use the time series data as data for constructing a learning model or logic. Furthermore, according to the first embodiment, a construction unit is provided that constructs a learning model or logic for determining a state based on the time-series data corrected by the data correction unit 103. As a result, the data processing system according to the first embodiment can obtain more appropriate time-series data, improving the accuracy of building a learning model or logic.

[0092] Furthermore, according to the first embodiment, the data processing program causes the computer to function as a data processing system. As a result, the data processing program according to the first embodiment can reduce errors caused by noise.

[0093] Furthermore, according to this embodiment 1, the data processing method includes the steps of: a data acquisition unit 101 acquiring time series data collected by two or more data collection devices 1011; a noise generation unit 102 generating pseudo-noise based on the difference between the time series data collected by the two or more data collection devices 1011 acquired by the data acquisition unit 101; and a data correction unit 103 correcting at least one of the time series data collected by the two or more data collection devices 1011 acquired by the data acquisition unit 101 using the pseudo-noise generated by the noise generation unit 102. As a result, the data processing method according to the first embodiment can reduce errors caused by noise.

[0094] Embodiment 2 The second embodiment shows a configuration example in which some of the functions of the data generating device 1 shown in the first embodiment are incorporated into the state determining device 2 side.

[0095] FIG. 8 is a block diagram showing an example of the configuration of a data processing system according to the second embodiment. In the data processing system according to the second embodiment shown in Fig. 8, data correction unit 103 is removed from data generating device 1, and a noise data storage unit 205, a data acquisition unit 206, and a data correction unit 207 are added to state determining device 2, compared to the data processing system according to the first embodiment shown in Fig. 1. The other configuration examples of the data processing system according to the second embodiment shown in Fig. 8 are the same as the configuration example of the data processing system according to the first embodiment shown in Fig. 1, and the same reference numerals are used, and only the different parts will be described.

[0096] The data representing the pseudo-noise generated by the noise generating unit 102 in the second embodiment is output as noise data to the noise data storage unit 205 in the state determining device 2.

[0097] The noise data storage unit 205 stores data indicating the pseudo noise generated by the noise generation unit 102 in the data generation device 1.

[0098] The noise data storage unit 205 may be, for example, a non-volatile or volatile semiconductor memory such as RAM, ROM, flash memory, EPROM, or EEPROM, a magnetic disk, a flexible disk, an optical disk, a compact disk, a mini disk, or a DVD.

[0099] 8 shows a case where the noise data storage unit 205 is provided inside the state determining device 2. In FIG. However, the present invention is not limited to this, and the noise data storage unit 205 may be provided outside the state determining device 2.

[0100] The data acquisition unit 206 acquires at least one piece of time series data collected by two or more data collection devices 1011, respectively.

[0101] At least one of the time series data collected by the data acquisition unit 206 and collected by two or more data collection devices 1011 is output as second collected data (data for learning) or third collected data (data for state determination) to the data correction unit 207. The second and third collected data are time series data collected at different timings from the first collected data.

[0102] Note that FIG. 8 shows a case where the data acquiring unit 206 acquires time-series data collected by one data collecting device 1011b.

[0103] The data correction unit 207 corrects at least one piece of time series data among the time series data collected by the two or more data collection devices 1011 and acquired by the data acquisition unit 206, using the pseudo noise generated by the noise generation unit 102. That is, the data correction unit 207 corrects the second collected data using the pseudo noise generated by the noise generation unit 102. At this time, the data correction unit 207 extracts the pseudo noise from the noise data storage unit 205 and performs the data correction.

[0104] More specifically, the data correction unit 207 corrects the second collected data by adding the pseudo noise generated by the noise generation unit 102 to the second collected data to which the pseudo noise is to be added. In this case, for example, the data correction unit 207 may add the pseudo noise by adding the pseudo noise to the second collected data, may add the pseudo noise by multiplying the second collected data by the pseudo noise, or may add the pseudo noise by performing statistical processing on the second collected data using the pseudo noise.

[0105] In addition, the data correction unit 207 may generate spike noise based on the setting information and use the spike noise to correct at least one of the time series data collected by two or more data collection devices 1011 and acquired by the data acquisition unit 206.

[0106] Furthermore, the data correction unit 207 corrects at least one of the time series data collected by the two or more data collection devices 1011 so as to use the time series data as data for constructing a learning model or logic. That is, the data correction unit 207 corrects and shapes the time series data into time series data that is easy to handle in the state determination device 2. At this time, the data correcting unit 207 calculates the feature amount for the second collected data. Furthermore, for example, the data correcting unit 207 may change the sampling rate or the unit of the second collected data in addition to calculating the feature amount.

[0107] The time-series data (data indicating the feature amounts) corrected by the data correcting unit 207 is output to the learning unit 202 or the state estimating unit 204.

[0108] The learning unit 202 in the second embodiment uses the feature amount obtained by the data correction unit 207 to learn a learning model for state determination (state determination model). Furthermore, the state estimation unit 204 in the second embodiment uses the feature amount obtained by the data correction unit 207 and the learning model obtained by the learning unit 202 to perform state estimation (estimation of whether the user is in an abnormal state).

[0109] Each process in the data processing device according to the second embodiment may be performed by a terminal or by the cloud.

[0110] Embodiment 3 In the second embodiment, a case has been described in which time-series data requiring noise correction is used as data for state determination, and state determination is performed after noise correction has been performed on this time-series data. In contrast, in the third embodiment, a case will be described in which time-series data not requiring noise correction is used as data for state determination, and state determination is performed without noise correction on this time-series data.

[0111] FIG. 9 is a block diagram showing an example of the configuration of a data processing system according to the third embodiment. In the data processing system according to the third embodiment shown in Fig. 9, a data acquisition unit 208 is added to the state determination device 2 in comparison with the data processing system according to the first embodiment shown in Fig. 1. The other configuration examples of the data processing system according to the third embodiment shown in Fig. 9 are the same as the configuration example of the data processing system according to the first embodiment shown in Fig. 1, and the same reference numerals are used and only the different parts will be described.

[0112] The data acquiring unit 208 acquires at least one of the time series data collected by the two or more data collecting devices 1011. The time series data acquired by the data acquiring unit 208 is time series data that does not require noise correction. In other words, the time series data acquired by the data acquiring unit 208 is time series data that is different from the second collected data (time series data to be subjected to noise correction).

[0113] At least one of the time series data collected by the data acquisition unit 208 and collected by the two or more data collection devices 1011 is output as third collected data (data for state determination) to the state estimation unit 204. The third collected data is time series data collected at a timing different from that of the first collected data.

[0114] 8 shows a case where the data acquiring unit 208 acquires time-series data collected by one data collecting device 1011a.

[0115] The state estimation unit 204 in the third embodiment uses the time-series data collected by the data acquisition unit 208 and the learning model obtained by the learning unit 202 to perform state estimation (estimation of whether the user is in an abnormal state).

[0116] Here, the learning unit 202 creates a learning model using time series data to which pseudo noise has been added by the data correction unit 103, that is, using time series data that takes into account the noise factors of the data collection device 1011. This enables the learning unit 202 to build a learning model that takes into account the measurement error of the data collection device 1011 when determining the state. Therefore, by using this learning model, time series data that does not require noise correction can be used as data for state determination when determining the state, i.e., the measurement data can be used directly for state determination.

[0117] Each process in the data processing device according to the third embodiment may be performed by a terminal or by the cloud.

[0118] Finally, an example of the hardware configuration of a data processing system according to embodiments 1 to 3 will be described with reference to Fig. 10. Below, an example of the hardware configuration of the data generating device 1 according to embodiment 1 will be described, but the same applies to the example of the hardware configuration of the data generating device 1 according to embodiments 2 and 3 and the state determining device 2 according to embodiments 1 to 3. The functions of the data acquisition unit 101, the noise generation unit 102, and the data correction unit 103 in the data generating device 1 are realized by a processing circuit 51. The processing circuit 51 may be dedicated hardware as shown in Fig. 10A, or may be a CPU (also referred to as a central processing unit, processing unit, arithmetic unit, microprocessor, microcomputer, processor, or DSP (Digital Signal Processor)) 52 that executes a program stored in a memory 53 as shown in Fig. 10B.

[0119] When the processing circuit 51 is dedicated hardware, the processing circuit 51 may be, for example, a single circuit, a composite circuit, a programmed processor, a parallel programmed processor, an ASIC (Application Specific Integrated Circuit), an FPGA (Field Programmable Gate Array), or a combination thereof. The functions of the data acquisition unit 101, the noise generation unit 102, and the data correction unit 103 may be realized individually by the processing circuit 51, or the functions of these units may be realized collectively by the processing circuit 51.

[0120] When the processing circuit 51 is a CPU 52, the functions of the data acquisition unit 101, the noise generation unit 102, and the data correction unit 103 are realized by software, firmware, or a combination of software and firmware. The software and firmware are written as programs and stored in the memory 53. The processing circuit 51 realizes the functions of each unit by reading and executing the programs stored in the memory 53. That is, the data generation device 1 includes the memory 53 for storing programs that, when executed by the processing circuit 51, result in the execution of, for example, each step shown in FIG. 3 . These programs can also be said to cause a computer to execute the procedures and methods of the data acquisition unit 101, the noise generation unit 102, and the data correction unit 103. Here, the memory 53 may be, for example, a non-volatile or volatile semiconductor memory such as RAM, ROM, flash memory, EPROM, or EEPROM, a magnetic disk, a flexible disk, an optical disk, a compact disk, a minidisk, or a DVD.

[0121] It is also possible to realize some of the functions of the data acquisition unit 101, the noise generation unit 102, and the data correction unit 103 with dedicated hardware and some with software or firmware. For example, the function of the data acquisition unit 101 can be realized by the processing circuit 51 as dedicated hardware, and the functions of the noise generation unit 102 and the data correction unit 103 can be realized by the processing circuit 51 reading and executing a program stored in the memory 53.

[0122] In this way, the processing circuitry 51 can realize each of the above-described functions by hardware, software, firmware, or a combination of these.

[0123] It should be noted that the embodiments may be freely combined, or any of the components in each embodiment may be modified, or any of the components in each embodiment may be omitted.

[0124] Various aspects of the present disclosure are summarized below as appendices.

[0125] (Appendix 1) a data acquisition unit that acquires time series data collected by two or more data collection devices; a noise generation unit that generates pseudo noise based on the difference between the time-series data collected by the two or more data collection devices, which is acquired by the data acquisition unit; a data correction unit that corrects at least one of the time series data acquired by the data acquisition unit and collected by the two or more data collection devices, using the pseudo noise generated by the noise generation unit; and A data processing system comprising: (Appendix 2) The noise generation unit generates pseudo noise based on the difference between the time-series data collected by the two or more data collection devices, which is acquired by the data acquisition unit, and information indicating each noise factor in the two or more data collection devices. 2. The data processing system of claim 1. (Appendix 3) The information indicating the noise factor is at least one of information collected by a data collection device, information on a location collected by the data collection device, and information on an environment collected by the data collection device. 3. The data processing system of claim 2. (Appendix 4) The data correction unit corrects at least one of the time series data collected by the two or more data collection devices so as to use the time series data as data for constructing a learning model or logic. 4. The data processing system according to any one of claims 1 to 3. (Appendix 5) A construction unit that constructs a learning model or logic for state determination based on the time-series data corrected by the data correction unit. 5. The data processing system of claim 1, wherein: (Appendix 6) A data processing program for causing a computer to function as the data processing system according to any one of appendices 1 to 5. (Appendix 7) a step in which a data acquisition unit acquires time series data collected by two or more data collection devices; a noise generating unit generating pseudo noise based on a difference between the time series data collected by the two or more data collection devices, the difference being acquired by the data acquiring unit; a data correction unit correcting at least one of the time series data acquired by the data acquisition unit and collected by the two or more data collection devices, using the pseudo noise generated by the noise generation unit; A data processing method comprising: [Explanation of symbols]

[0126] 1 Data generation device, 2 State determination device, 51 Processing circuit, 52 CPU, 53 Memory, 101 Data acquisition unit, 102 Noise generation unit, 103 Data correction unit, 201 Feature acquisition unit, 202 Learning unit (construction unit), 203 Model storage unit, 204 State estimation unit, 205 Noise data storage unit, 206 Data acquisition unit, 207 Data correction unit, 1011 Data collection device.

Claims

1. a data acquisition unit that acquires time series data collected by two or more data collection devices; a noise generation unit that generates pseudo noise based on the difference between the time-series data collected by the two or more data collection devices, which is acquired by the data acquisition unit; a data correction unit that corrects at least one of the time series data acquired by the data acquisition unit and collected by the two or more data collection devices, using the pseudo noise generated by the noise generation unit; A data processing system comprising:

2. The noise generation unit generates pseudo noise based on the difference between the time-series data collected by the two or more data collection devices, which is acquired by the data acquisition unit, and information indicating each noise factor in the two or more data collection devices.

2. The data processing system according to claim 1.

3. The information indicating the noise factor is at least one of information collected by a data collection device, information on a location collected by the data collection device, and information on an environment collected by the data collection device.

3. The data processing system according to claim 2.

4. The data correction unit corrects at least one of the time series data collected by the two or more data collection devices so as to use the time series data as data for constructing a learning model or logic.

2. The data processing system according to claim 1.

5. A construction unit that constructs a learning model or logic for state determination based on the time-series data corrected by the data correction unit.

2. The data processing system according to claim 1.

6. A data processing program for causing a computer to function as the data processing system of claim 1.

7. a step in which a data acquisition unit acquires time-series data collected by two or more data collection devices; a noise generating unit generating pseudo noise based on a difference between the time series data collected by the two or more data collection devices, the difference being acquired by the data acquiring unit; a data correcting unit correcting at least one of the time series data acquired by the data acquiring unit and collected by the two or more data collecting devices, using the pseudo noise generated by the noise generating unit; A data processing method comprising:

Images