JP7359009B2 - Data analysis device, program and method - Google Patents

Description

The present invention relates to a data analysis device, program, and method, and for example, to a method for managing input data used for data analysis.

BACKGROUND ART Data analysis systems that install devices such as sensors on various objects and collect and analyze sensor data and the status of the devices are becoming popular.

For example, there is an anomaly detection system that uses data such as sensor data to detect the occurrence of an abnormality or a sign of an abnormality in a monitored object such as a device or a structure. Such an anomaly detection system includes a sensor terminal, a data collection device, a data analysis device, a monitoring terminal, an alert display device, and the like. In this abnormality detection system, a data collection device periodically collects sensor data from sensor terminals, and a data analysis device analyzes the sensor data each time it receives the sensor data to determine whether the monitored object is normal or abnormal.

If the sensor collection cycle is long, even if a sign of a change to an abnormal state is detected, the object to be monitored changes rapidly by the time of the next sensor data collection, and by the time the abnormal state is detected, the equipment is The abnormality may progress to a state where recovery control cannot be performed.

Patent Document 1 discloses a data management system that can provide appropriate data when performing analysis using sensor data. More specifically, in Patent Document 1, when the server detects that an abnormality or a sign of abnormality has occurred in the monitored object based on sensor data, the server sets a collection rule to change the amount of sensor data. Select and send the collection rule to the gateway. The gateway obtains and manages the state of the sensor and selects the first condition according to the collection rule received from the server. Then, the gateway changes settings for acquiring sensor data based on the first condition and the state of the sensor. The gateway acquires sensor data from the sensor based on this setting and sends it to the server.

In this way, in Patent Document 1, when it is determined that there is a sign of an abnormality, the frequency of collecting sensor data is changed so that the progress of the abnormality can be grasped without delay.

JP2019-95883A

By the way, in order to discover signs of abnormality, it is necessary to measure subtle changes in sensor values.

However, changes in sensor values are not limited to abnormalities in the monitored object. For example, the sensor value may change due to an abnormality in the sensor or the influence of the surrounding environment other than the sensing target. In this case, even if the state of the object to be monitored has not changed, the sensor value may change slightly, and the object may be determined to be abnormal.

More specifically, errors may occur in sensor values due to, for example, changes in ambient temperature or humidity, sensor malfunction, or obstacles approaching the sensor. As described above, the occurrence of measurement errors may cause a problem of erroneously determining the state of the device.

An object of the present invention is to determine whether sensor data can be used for data analysis, taking into account the state of the sensor and its surroundings, and to provide the usable sensor data to a data analysis unit.

In order to solve this problem, the data analysis device according to the first aspect of the present invention includes a data storage means for storing sensor data used for data analysis, and a data storage means for storing information about related sensors related to the sensor used for data analysis. an information storage means, a learning means for learning the correlation of the time-series changes using the time-series changes in sensor data used for data analysis, and the time-series changes in the sensor data of related sensors; and a data analysis means that performs a predetermined analysis process using the sensor data that is determined to be usable by the usability determination means. It is characterized by

A data analysis program according to a second aspect of the present invention includes a computer, a data storage means for storing sensor data used for data analysis, and a related information storage means for storing information about related sensors related to the sensors used for data analysis. A learning means that learns the correlation of time-series changes using time-series changes in sensor data used for data analysis and time-series changes in sensor data of related sensors, and a learning method that uses the learning results of the learning means to It is characterized by functioning as a usability determination means for determining whether or not it can be used for analysis, and a data analysis means for performing a predetermined analysis process using the sensor data determined to be usable by the usability determination means.

In the data analysis method according to the third aspect of the present invention, the data storage means stores sensor data used for data analysis, the related information storage means stores information about related sensors related to the sensors used for data analysis, and the data analysis method stores sensor data used for data analysis. The means learns the correlation of the time-series changes using the time-series changes in the sensor data used for data analysis and the time-series changes in the sensor data of the related sensors, and the usability determination means uses the learning results by the learning means. It is characterized in that the sensor data is determined whether or not it can be used for data analysis, and the data analysis means performs a predetermined analysis process using the sensor data determined to be usable by the usability determination means.

According to the present invention, it is possible to determine whether sensor data can be used for data analysis in consideration of the state of the sensor and its surroundings, and to provide the usable sensor data to the data analysis section.

1 is an internal configuration diagram showing an internal configuration of a data analysis device according to a first embodiment; FIG. 1 is an overall configuration diagram showing the overall configuration of a monitoring system including a data analysis device according to a first embodiment; FIG. 2 is a flowchart showing learning of sensor data (sensor values) in the data analysis device according to the first embodiment. It is a flowchart which shows abnormality detection processing of a sensor value in a data analysis device concerning a 1st embodiment. FIG. 3 is a diagram showing time-series changes in sensor values in the first embodiment. FIG. 3 is a diagram illustrating data validity determination processing according to the first embodiment. FIG. 2 is an internal configuration diagram showing the internal configuration of a data analysis device according to a second embodiment. It is a flowchart which shows abnormality detection processing of a sensor value in a data analysis device concerning a 2nd embodiment. FIG. 7 is a diagram showing time-series changes in sensor values according to the second embodiment.

(A) First Embodiment Below, a first embodiment of a data analysis device, program, and method according to the present invention will be described in detail with reference to the drawings.

(A-1) Configuration of first embodiment [Data analysis device]
FIG. 1 is an internal configuration diagram showing the internal configuration of a data analysis device according to the first embodiment.

In FIG. 1, the data analysis device 1 according to the first embodiment includes a data acquisition section 11, a data temporary storage section 12, an availability determination section 13, a data analysis section 14, a sensor data storage section 15, and a related information storage section 16. , has a learning section 17.

The data analysis device 1 can be realized by, for example, a device having a CPU, ROM, RAM, EEPROM, input/output interface section, and the like. The CPU can be implemented by executing a processing program (for example, a data analysis program, etc.) stored in the ROM. Further, the system may be constructed by installing a processing program (for example, a data analysis program, etc.). In that case, the data analysis program can be represented by functional blocks illustrated in FIG.

The data acquisition unit 11 acquires sensor data used for analysis. The sensor from which the data acquisition unit 11 collects sensor data may be one sensor or a plurality of sensors. Further, the sensor may be one type of sensor or multiple types of sensors. FIG. 1 illustrates a case where sensor data is collected from each of a plurality of types of sensors.

The data temporary storage unit 12 temporarily stores sensor data acquired by the data acquisition unit 11. That is, the data temporary storage unit 12 temporarily stores data used for analysis. For example, when analyzing changes over the past hour, the data temporary storage unit 12 stores one hour's worth of data used for data analysis.

The usability determining unit 13 determines whether the sensor value can be used for data analysis, according to the filter rules described in the related information storage unit 16. The usability determination section 13 provides the data that has been determined to be usable for analysis to the data analysis section 14 . In the first embodiment, the availability determination unit 13 inputs only valid data to the data analysis unit 14 based on the filter rule, and does not input invalid data to the data analysis unit 14. In order to distinguish between valid data and invalid data in this way, the availability determining unit 13 is also called a data filter unit.

Here, "valid data" is data (sensor values) determined to be usable for data analysis. “Invalid data” is data (sensor values) that has been determined to be unusable for data analysis.

For example, errors may occur in sensor values (data values) due to changes in the temperature or humidity around the sensor, malfunction of the sensor, or obstacles approaching the sensor. Although errors in sensor values are minute, if such sensor values are used to perform data analysis, the analysis results will be affected. In the first embodiment, the availability determining unit 13 determines whether or not data can be used for data analysis based on filter rules, and identifies data that can be used for data analysis as "valid data". shall be. In other words, even if learning is performed using sensor values and an error (a value exceeding the threshold) occurs in the sensor value using a threshold, the error is within the expected range and there will be no impact even if it is used for data analysis. Sensor values that can be considered to have no effect on the environment are defined as "valid data."

The "filter rule" defines a procedure, setting values, etc. for determining whether a sensor value is valid data. Filter rules are also called "usability determination algorithms." For example, the filter rules include a learning procedure for time-series changes in sensor values, a unit time for determining the correlation of time-series changes in sensor values, a threshold for determining whether a sensor value is abnormal, This includes predicted values of sensor values, etc. The filter rule may be determined depending on the type of sensor or the like.

Note that the method of determining the validity of the sensor value by the availability determining unit 13 (also referred to as "availability determination") will be described in detail in the operation section.

The data analysis unit 14 performs data analysis using the data to be analyzed that is input from the availability determination unit 13, and provides the analysis result to an external device (for example, the monitoring device 4 in FIG. 2). The data analysis method is not particularly limited, and various methods can be applied.

The sensor data storage unit 15 stores sensor data that may be used for data analysis. The sensor data storage unit 15 stores not only sensor data but also data analysis results obtained by data analysis by the data analysis unit 14.

Here, the data used in data analysis by the data analysis device 1 is not limited to new data input into the data temporary storage section 12 via the data acquisition section 11. Data acquired in the past and analysis results by the data analysis unit 14 may be stored in the sensor data storage unit 15, and various data stored in the sensor data storage unit 15 may be read out and used for data analysis. . In addition, even if the data is input to the data acquisition unit 11 but not used for data analysis, the data is not deleted, but such data is marked as "not used for analysis", for example. The sensor data may be stored in the sensor data storage unit 15 with an identifier indicating the .

The related information storage unit 16 stores related information such as sensor address information (identification information), installation location, data analysis program, and filter rules when providing data for analysis. Note that the related information is not limited to the above-mentioned types of information.

Further, the related information storage unit 16 stores information regarding sensors used for data analysis and sensors related to the sensors. For example, it is assumed that the sensor is installed near the sensor to be analyzed. The type of sensor does not matter.

More specifically, for example, it is assumed that a plurality of sensors are installed on a monitoring target (monitored object). Information regarding the installation position of each sensor installed in the monitoring target is stored in the related information storage section 16. If a certain sensor is used as a data analysis target among multiple sensors, other sensors associated with the sensor may be used to recognize one or more other sensors installed near the sensor. Stores sensor information.

Then, based on the related information stored in the related information storage section 16, data to be acquired by the data acquisition section 11 (for example, which sensor to acquire sensor data from) is specified, and the data is stored in the data temporary storage section 12. The retention period of the data can be determined. Furthermore, based on the related information stored in the related information storage section 16, filter rules are specified in the availability determination section 13, and furthermore, the analysis method in the data analysis section 14 and the learning method in the learning section 17 are specified. It is specified.

The learning unit 17 uses the sensor values stored in the data temporary storage unit 12, the time-series changes in the sensor values of the sensor to be analyzed, and the time-series changes in the sensor values of the related sensors. Learning the correlation of time series changes. The learning unit 17 is a type of data analysis, and has a function of learning time-series changes in data, correlations between analysis data and related sensors, and the like. There are no particular limitations on the learning method.

[Example of overall configuration]
FIG. 2 is an overall configuration diagram showing the overall configuration of a monitoring system including the data analysis device 1 of the first embodiment. The overall configuration of the monitoring system 5 illustrated in FIG. 2 is an example.

In FIG. 2, the monitoring system 5 includes a plurality of sensor devices 2 (2-1 to 2-n), a relay device 3, a data analysis device 1, and a monitoring device 4.

In the monitoring system 5, the data analysis device 1 collects sensor data and performs data analysis, and the monitoring device 4 determines abnormalities in the monitored object based on the analysis results of the data analysis device 1.

Here, an abnormality in the monitored object is a learning process that uses the measured values (sensor values) of the sensor device 2 that detects the characteristics and status of the monitored object to learn the time-series changes in sensor values. The result obtained by the sensor is compared with a certain sensor value, and the comparison result is evaluated using a threshold value. When a result exceeding the threshold value is obtained, it can be said that an abnormality has occurred. The object to be monitored is not particularly limited.

The sensor devices 2-1 to 2-n are installed on the object to be monitored, and detect the characteristics, state, etc. of the object to be monitored. The sensor devices 2-1 to 2-n each include, for example, a sensor, a communication device, a control unit, etc., and the sensor measures a sensor value, and the communication device transmits a signal including the sensor value via the relay device 3. and transmits it to the data analysis device 1. The types of sensors of the sensor devices 2-1 to 2-n are not particularly limited, and for example, acceleration sensors, temperature sensors, humidity sensors, etc. can be applied. The sensor devices 2-1 to 2-n may each have different types of sensors, or may have the same type of sensors. Further, the communication method of the communication devices of the sensor devices 2-1 to 2-n is not particularly limited, and the communication devices send and receive communication signals via a wireless line or a wired line.

The relay device 3 collects sensor values from the sensor devices 2-1 to 2-n and transfers them to the data analysis device 1. For example, a gateway can be used as the relay device 3.

The data analysis device 1 performs data analysis using the sensor values received via the relay device 3 and provides the analysis results to the monitoring device 4 .

The monitoring device 4 displays the analysis results by the data analysis device 1 and sets the data analysis results. The monitoring device 4 has a user interface such as a computer, a display unit such as a display, and an input unit such as a keyboard and a mouse. For example, a personal computer, a mobile terminal such as a smartphone or a tablet terminal, etc. can be applied.

(A-2) Operation of First Embodiment Next, processing in the data analysis device 1 according to the first embodiment will be described in detail with reference to the drawings.

[Initial stage: Learning sensor data]
FIG. 3 is a flowchart showing learning of sensor data (sensor values) in the data analysis device 1 according to the first embodiment.

In order for the usability determining unit 13 to determine the validity of sensor values that are subject to data analysis, it is necessary to know the range of change in values for determining that the sensor values are valid data.

In the first embodiment, a case will be exemplified in which sensor values (sensor data) to be analyzed are accumulated and a change range of the sensor values is determined from time-series changes in the sensor data.

For example, each of the sensor devices 2-1 to 2-n measures a sensor value indicating the characteristics or state of the object to be monitored, and a signal including the sensor value is collected by the relay device 3. The relay device 3 transfers the collected sensor values to the data analysis device 1.

[S101] Data Input In the data analysis device 1, sensor values (sensor data) are input to the data acquisition unit 11 (S101). The related information storage unit 16 stores information regarding sensor values of sensors targeted for data analysis. The data acquisition unit 11 receives sensor values of sensors to be analyzed based on the related information in the related information storage unit 16 .

[S102] Temporary storage The sensor value (sensor data) input to the data acquisition unit 11 is temporarily stored in the data temporary storage unit 12 (S102). The related information storage unit 16 stores the temporary holding time of the sensor values to be analyzed. The data temporary storage unit 12 temporarily holds the sensor value based on the temporary holding time of the related information storage unit 16.

After the temporary holding time has elapsed, the temporarily held sensor value may be deleted from the temporary data storage unit 12. Note that the sensor values stored in the data temporary storage section 12 may be stored in the sensor data storage section 15.

[S103] Acquisition and storage of related data Data related to the sensor value (sensor data) to be analyzed is input to the data acquisition unit 11, and the related data is temporarily held in the data temporary storage unit 12 (S103 ).

Here, the related information storage unit 16 stores, for example, a sensor targeted for data analysis and information regarding the sensor related thereto. Therefore, the data acquisition unit 11 also acquires sensor values detected by sensors related to the sensor targeted for data analysis, and stores them in the data temporary storage unit 12 .

[S104] Derivation of change range of sensor value The learning unit 17 uses the sensor values held in the data temporary storage unit 12 to determine the time-series change in the sensor value, and calculates a predetermined range of change from the time-series change in the sensor value. A range of changes in sensor values within time is derived (S104). That is, the change range of the sensor value of the sensor to be data analyzed is determined using the time-series change in the sensor value of the sensor to be data analyzed and the time-series change in the sensor value of the related sensor. A detailed explanation of the process for deriving the change range of the sensor value will be described later using FIG. 5.

For example, when one hour's worth of sensor values is held in the data temporary storage section 12, time-series changes in the one hour's worth of sensor values can be seen. In that case, the change range of the sensor value can be derived based on the difference value between the largest sensor value and the smallest sensor value within a predetermined period of time. Furthermore, as time passes thereafter, new sensor values are held in the data temporary storage section 12, and the time-series changes also vary. Therefore, for example, an averaged change range may be derived using the past change range (the range between the largest value and the smallest value).

Alternatively, for example, the learning unit 17 may calculate a correlation between sensor values between sensors related to the target sensor data, and may calculate a correlation between time-series changes in sensor values within a predetermined period of time.

Here, the correlation of time-series changes is, for example, the correlation of changes in the differences between the respective sensors at the start time and end time of a certain period of time. More specifically, for a certain sensor, a change value of the difference between the sensor value at the start time and the sensor value at the end time of a predetermined period of time (herein referred to as a first change value) is determined. Furthermore, for other related sensors, a change value (second change value) of the difference between the sensor value at the start time and the sensor value at the end time is determined. Then, a correlation between the first change value and the second change value may be determined. This is because the individual performance of the sensor may differ from sensor to sensor, and the measured values may differ slightly, but even in that case, the reliability is maintained in the difference in the change in the measured value, so the above-mentioned It is also possible to use the values measured by the sensor as a correlation of time-series changes. Furthermore, for example, correlations including intermediate changes in the values of each sensor within a certain period of time may be learned.

[S105] Storing Related Information Thereafter, the learning unit 17 repeatedly performs the process of S104 using the sensor values stored in the temporary data storage unit 12 one after another. Then, when the correlation information between sensors is updated, it is stored in the sensor data storage section 15.

[Sensor value abnormality detection processing]
Next, the sensor value abnormality detection process in the data analysis device 1 according to the first embodiment will be described with reference to the drawings.

FIG. 4 is a flowchart showing sensor value abnormality detection processing in the data analysis device 1 according to the first embodiment.

[S201] Data Input Sensor values (sensor data) of the sensor to be analyzed are input to the data acquisition unit 11 (S201). Basically, the same process as S101 in FIG. 3 can be performed.

[S202] Temporary storage The sensor value (sensor data) input to the data acquisition unit 11 is temporarily stored in the data temporary storage unit 12 (S202). Basically, the same process as S102 in FIG. 3 can be performed. Note that the time for temporarily retaining the sensor value in the data temporary storage unit 12 may be the same as the time in S102 of FIG. 3, or may be different.

[S203] Data Validity Determination Process The availability determination unit 13 checks whether the input sensor value is within the range of predicted values. If the data is within the predicted value range, it is considered valid data. If the predicted value is outside the range, the data is considered invalid.

[S204-S206]
If the usability determining unit 13 determines that the data is invalid (S204/No), the sensor value is deleted (S205). If the data is determined to be invalid, the corresponding data is deleted. Alternatively, data may be input to the data analysis unit 14 even when the data is invalid, but the weight of the use of the data in the data analysis unit 14 may be lowered to make it less likely to affect the analysis.

On the other hand, if the usability determination unit 13 determines that the data is valid (S204/Yes), the sensor value is given to the data analysis unit 14 for data analysis processing (S206). . The analysis results are given to the monitoring device 4 and displayed.

Next, an example of abnormality detection as data analysis will be described.

FIG. 5 is a diagram showing time-series changes in sensor values according to the first embodiment. In FIG. 5, time-series changes in sensor values of sensor device 2-1 (hereinafter simply referred to as “sensor 2-1”) that is subject to data analysis and sensor device 2-2 (hereinafter simply referred to as “sensor 2-1”) related to sensor 2-1 are shown. , simply referred to as "sensor 2-2"). It is assumed that there is a correlation between the time-series change in the sensor value of the sensor 2-1 and the time-series change in the sensor value of the sensor 2-2.

In FIG. 5, the learning unit 17 calculates the change value of the difference between the sensor values at predetermined time intervals with respect to the time-series changes in the sensor values of the sensor 2-1 and the sensor 2-2, and learns the correlation of the time-series changes. do. For example, the predetermined times for capturing the change range of sensor values are T1 (start time tL1 to end time tL3), T2 (start time tL2 to end time tL4), T3 (start time tL3 to end time tL5), T4 (start time tL3), and T4 (start time tL3 to end time tL5). tL4 to end time tL6), T5 (start time tL5 to end time tL7), and so on. For example, for the sensor 2-1, a change value (first change value) of the difference between the sensor value at the start time tL1 and the sensor value at the end time tL2 of the predetermined time T1 is determined. Similarly, for the sensor 2-2, a change value (second change value) of the difference between the sensor value at the start time tL1 and the sensor value at the end time tL2 of the predetermined time T1 is determined. The same process is performed for other predetermined times to learn the correlation between the change values of sensor 2-1 and sensor 2-2 for each predetermined period.

Through learning, the range of change in sensor value per unit time of sensor 2-1 is determined. Then, the usability determining unit 13 determines that the sensor value is invalid data if it is outside this change range. On the other hand, if the sensor value is within the change range, it is determined to be valid data.

FIG. 6 is a diagram illustrating the data validity determination process of the first embodiment. FIG. 6 illustrates a validity determination method for determining whether input sensor values are data that can be used for data analysis.

The threshold value Vth is a threshold value for determining whether or not an abnormality has occurred in the monitored object. If the sensor value is equal to or greater than the threshold value Vth, it is assumed that an abnormality has occurred in the object to be monitored. The sensor value at time t11 is Vt11, the sensor value at time t12 is Vt12, the sensor value at time t13 is Vt13, and the sensor value at time t14 is Vt14.

For example, from time t11 to t12, the sensor value changes from Vt11 to V12 and exceeds the threshold value Vth. The predicted value of the sensor value at time t12 by learning is set from Vt12_min to Vt12_max. Then, since the sensor value Vt12 is outside this range (from Vt12_min to Vt12_max) when viewed from the sensor value Vt11, the sensor value Vt12 is treated as invalid data and is not determined to be abnormal.

Further, for example, from time t13 to t14, the sensor value changes from Vt13 to Vt14 and exceeds the threshold value Vth. The predicted value of the sensor value at time t14 through learning is set from Vt14_min to Vt14_max. Since the sensor value Vt14 is within this range (Vt14_min to Vt14_max), the sensor value Vt14 is used as valid data for data analysis and determined to be abnormal. Then, the monitoring device 4 is notified and displayed that an abnormality has occurred in the object to be monitored.

In this way, the value Vt12 at time t12 can be determined to be a sensor value that has temporarily changed significantly due to, for example, a sensor failure.

On the other hand, it can be determined that the sensor value Vt14 at time t14 indicates that there is no abnormality in the sensor itself and that a correct value has been input. Therefore, since the threshold value Vth for determining the abnormality of the monitored object has been exceeded, the monitored object is determined to be abnormal.

(A-3) Effects of the first embodiment As described above, in the first embodiment, the sensor data of the target sensor and related sensors are held for a predetermined period of time, and the sensor values between each sensor change over time. The system learns the correlation between the two and uses the correlation to determine whether the next input sensor data is valid or invalid. This allows only valid data to be used for analysis and improves the accuracy of analysis results.

Moreover, the probability that a sensor value shows an abnormal value due to the influence of the sensor or the measurement environment and that the monitored object is erroneously determined to be abnormal even though there is no abnormality is reduced, and false notifications are also reduced.

(B) Second Embodiment Next, a second embodiment of the data analysis device, program, and method according to the present invention will be described in detail with reference to the drawings.

(B-1) Configuration of Second Embodiment FIG. 7 is an internal configuration diagram showing the internal configuration of a data analysis device according to the second embodiment.

In FIG. 7, a data analysis device 1A according to the second embodiment includes a data acquisition section 11, a data temporary storage section 12, an availability determination section 13, a data analysis section 14, a sensor data storage section 15, and a related information storage section 16. , a learning section 17, a sensor value prediction section 21, and a correction information input section 22.

The data analysis device 1A has a configuration in which a sensor value prediction unit 21 and a correction information input unit 22 are added to the components shown in FIG. 1 of the first embodiment. Components other than the sensor value prediction unit 21 and the correction information input unit 22 have been described in the first embodiment, so detailed description thereof will be omitted here.

The sensor value prediction unit 21 uses information regarding time-series changes in sensor values of sensors targeted for data analysis and information regarding time-series changes in sensor values of related sensors. (sensor value of the sensor) is predicted.

The correction information input unit 22 inputs correction information (correction formula) such as the degree of correlation between a sensor subject to data analysis and related sensors from an external source such as the monitoring device 4, and stores the correction information in related information. It is stored in section 16.

The availability determination unit 13 inputs the corrected valid data to the data analysis unit 14. For data that is valid even without correction, the data before correction may be input to the data analysis section 14.

(B-2) Operation of Second Embodiment FIG. 8 is a flowchart showing sensor value abnormality detection processing in the data analysis device 1A according to the second embodiment. In FIG. 8, the same processes as in FIG. 4 of the first embodiment are given the same reference numerals.

[S201] Data input Sensor values (sensor data) are input to the data acquisition unit 11 (S201).

[S202] Temporary storage The sensor value (sensor data) input to the data acquisition unit 11 is temporarily stored in the data temporary storage unit 12 (S202). Note that the time for temporarily retaining the sensor value in the data temporary storage unit 12 may be the same as the time in S102 of FIG. 3, or may be different.

[S203] Data Validity Determination Process The availability determination unit 13 checks whether the input sensor value is within the range of predicted values. If the data is within the predicted value range, it is considered valid data. If the predicted value is outside the range, the data is considered invalid.

[S204]
If the usability determination unit 13 determines that the data is valid data, the process moves to S303 in order to use it for data analysis. On the other hand, if it is determined that the data is invalid, the process moves to S301.

[S301] Estimated value prediction When the data is determined to be invalid by the availability determining unit 13, the sensor value predicting unit 21 predicts a value that can be considered normal as an estimated value (predicted value) for the sensor value at the time determined to be invalid. do.

Various methods can be applied to estimate the estimated value (predicted value) by the sensor value prediction unit 21. For example, in a time-series change in a sensor value to be analyzed, a value within the change range of the sensor value at a time determined to be invalid may be used as the estimated value. More specifically, for example, the maximum value or minimum value that specifies the change range of the sensor value may be used as the estimated value, or, for example, the median value of the change range may be used as the estimated value.

[S302] Data Correction The estimated value predicted by the sensor value prediction unit 21 is input to the data temporary storage unit 12. The estimated value is also input to the sensor data storage section 15 . Note that the sensor data storage section 15 may store only the estimated value, or may store both the sensor value and the estimated value input from the data acquisition section 11.

Then, the learning unit 17 uses the sensor values stored in the data temporary storage unit 12 to correct the time-series changes in the sensor values. At this time, time-series changes in the sensor values to be data analyzed are corrected using the degree of correlation (value indicating correlation) between the sensor values to be analyzed and related sensor values.

Learning may be performed using either the estimated value or the sensor value input from the sensor subject to data analysis, or both data. When using estimated values, learning processing is performed after the estimated values are input from the sensor value prediction unit 21 to the data temporary storage unit 12.

[S303] Data analysis processing Normal data or corrected data is input to the data analysis section 14, and the data analysis section 14 performs data analysis. The data analysis method may be the same as in the first embodiment. Alternatively, data may be weighted so that normal data is weighted higher than corrected data so that it is more likely to be influenced by analysis.

Next, an example of abnormality detection will be explained as data analysis.

FIG. 9 is a diagram showing time-series changes in sensor values according to the second embodiment. 9(A) shows a time-series change in the sensor value input from the sensor targeted for data analysis, and FIG. 9(B) shows a time-series change in the sensor value of the related sensor as correction information. (C) shows time-series changes when sensor values are corrected using correlation with related monitoring target data.

In FIG. 9(A) and FIG. 9(C), Vth is a threshold value, and if it is equal to or greater than the threshold value Vth, it is assumed that the monitored object is abnormal.

Data analysis uses corrected data. As shown in FIG. 9(A), at time t21, the sensor value Vt21 exceeds the threshold value Vth, but as shown in FIG. 9(C), the corrected sensor value Vt21' does not exceed the threshold value Vth. The monitored object is not determined to be abnormal.

On the other hand, in FIG. 9C, at time t22, the corrected sensor value Vt22' exceeds the threshold value Vth. In this case, the object to be monitored is determined to be abnormal, and the monitoring device 4 is notified and displayed that the object to be monitored is abnormal.

In this way, it is assumed that the sensor value Vt21 at time t21 has temporarily changed significantly due to, for example, a sensor failure. On the other hand, at time t22, it is assumed that there is no abnormality in the sensor itself and a correct value has been input, and since the threshold value Vth for abnormality determination of the monitored object has been exceeded, the monitored object is determined to be abnormal.

(B-3) Effects of the second embodiment As described above, in the second embodiment, the estimated value of the sensor value at the relevant time is obtained by referring to information other than the relevant sensor, and this estimated value is used to obtain the estimated value of the sensor value at the relevant time. I decided to analyze it. As a result, compared to the first embodiment, it is possible to increase the amount of data used for data analysis and further improve the reliability of the analysis results.

(C) Other Embodiments (C-1) In the embodiments described above, the data analysis device determines whether or not data can be used, and the data that can be used for data analysis is provided to the data analysis unit. In the embodiment described above, an example of a method in which the learning unit learns time-series changes in sensor values in the data analysis device is shown, but the learning method is not limited. Further, although the data analysis unit analyzes abnormality detection of the monitored object using time-series sensor values, the data analysis method is not limited to abnormality detection.

(C-2) In the embodiments described above, the monitoring system has been explained separately as a relay device (for example, a gateway) and a data analysis device. However, they may be provided on the same hardware. In other words, the relay device may include a data analysis device. Further, each function of the data analysis device may be implemented in separate hardware to perform distributed processing.

(C-3) Although the case in which the data input to the data analysis device is sensor data has been exemplified, data other than sensor data may be input to the data analysis device.

(C-4) In the embodiment described above, the case where there is one sensor related to the sensor to be analyzed is illustrated, but each of the time-series changes in the sensor values of a plurality of related sensors and the sensor to be analyzed are A correlation with time-series changes in sensor values may be derived.

1 and 1A...Data analysis device, 11...Data acquisition section, 12...Data temporary storage section, 13...Usability determination section, 14...Data analysis section, 15...Sensor data storage section, 16...Related information storage section, 17... Learning section, 21...Sensor value prediction section, 22...Correction information input section.

Claims (6)

a data storage means for storing sensor data used for data analysis;
Related information storage means for storing information regarding related sensors related to the sensors used for the data analysis;
learning means for learning correlations between time-series changes using time-series changes in sensor data used for the data analysis and time-series changes in sensor data of the related sensors;
Usability determination means for determining whether or not the sensor data can be used for data analysis using the learning result by the learning means;
A data analysis device comprising: a data analysis device that performs a predetermined analysis process using the sensor data determined to be usable by the availability determination device. The learning means uses the correlation of the time-series changes to derive a change range of the sensor data used for the data analysis,
The data analysis device according to claim 1, wherein the availability determining means refers to the change range and determines whether or not the sensor data can be used for data analysis. The availability determination means determines that the device can be used when the value of the sensor data is within the change range, and determines that the device is not available when the value of the sensor data is outside the change range. The data analysis device according to claim 2. comprising a prediction means for predicting an estimated value for the value of the sensor data determined to be unavailable by the availability determination means,
The data analysis device according to claim 3, wherein the learning means corrects the time-series change in the sensor data of the data analysis using the time-series change in the sensor data of the related sensor. computer,
a data storage means for storing sensor data used for data analysis;
Related information storage means for storing information regarding related sensors related to the sensors used for the data analysis;
learning means for learning correlations between time-series changes using time-series changes in sensor data used for the data analysis and time-series changes in sensor data of the related sensors;
Usability determination means for determining whether or not the sensor data can be used for data analysis using the learning result by the learning means;
A data analysis program that functions as a data analysis unit that performs a predetermined analysis process using the sensor data determined to be usable by the availability determination unit. The data storage means stores sensor data used for data analysis,
the related information storage means stores information regarding related sensors related to the sensor used for the data analysis;
a learning means learns a correlation between time-series changes using a time-series change in sensor data used for the data analysis and a time-series change in sensor data of the related sensor;
Usability determination means determines whether the sensor data can be used for data analysis using the learning result by the learning means,
A data analysis method characterized in that a data analysis means performs a predetermined analysis process using the sensor data determined to be usable by the availability determination means.

Images