JP2021111034A - Abnormality detection program, abnormality detection method, and information processing device - Google Patents

Abstract

To improve detection accuracy for abnormal data.SOLUTION: An information processing device acquires first plurality of values obtained according to measurement by a first sensor, and second plurality of values obtained according to measurement by a second sensor. Next, the information processing device classifies a plurality of sets obtained by combining each of the first plurality of values and each of the second plurality of values into a plurality of groups. Then the information processing device detects an abnormality on the basis of a variation in the number of sets respectively contained in the plurality of groups.SELECTED DRAWING: Figure 11

Description

The present invention relates to an abnormality detection technique.

Various sensors are attached to the objects to be managed in the operating state (managed objects). The measurement data including the measurement value measured by the sensor can be collected by the server using a technology such as IoT (Internet of Things). The server can analyze the measurement data of a large number of collected sensors and grasp the state of the managed object. For example, the server can grasp the operation status and performance of a ship based on the time-series changes of the measured values of a large number of sensors attached to the ship. The server can also perform machine learning using the measured values of a large number of sensors attached to the ship as input data and generate a prediction model for diagnosing the performance of the ship. Then, the server can estimate the presence or absence of an abnormality that hinders the operation of the ship based on the measurement data of the sensor by using the generated prediction model.

As a data processing technique based on the measured data, a time-series data abnormality monitoring device that appropriately detects an abnormality due to unspecified fluctuations in time-series data such as communication traffic of a network device has been proposed.

JP-A-2017-143399

The measurement data of the sensor may include abnormality data caused by measurement abnormality such as spike noise and measurement error data. These anomalous data do not accurately represent the state of the controlled object to which the sensor is attached. If the measurement data of a plurality of sensors attached to the controlled object contains abnormal data, the state of the controlled object cannot be correctly determined using the measured data. Therefore, for example, when a predetermined index value based on the measured value obtained from the sensor exceeds the threshold value, the server determines that the measured data contains abnormal data and excludes the measured data from the analysis target. Can be done.

However, the detection of data abnormality by comparison with the threshold value is based on the premise that the threshold value for determination is set in advance by the expert, and it is not possible to detect the abnormality of the data for which the threshold value is not set by the expert. Moreover, unlike managed objects in a controlled environment such as network devices, managed objects operated in a natural environment such as ships have abnormal data in the sensor measurement data due to the influence of the external environment. It often appears. Therefore, it is difficult to set the threshold value assuming all situations. As described above, in the prior art, it is not possible to detect abnormal data caused by an unexpected event, and it is difficult to improve the detection accuracy of the abnormal data.

On one side, the present case aims to improve the detection accuracy of anomalous data.

One proposal provides an anomaly detection program characterized by causing a computer to perform the following processes.
The computer acquires the first plurality of values obtained based on the measurement of the first sensor and the second plurality of values obtained based on the measurement of the second sensor. Next, the computer classifies a plurality of sets in which each of the first plurality of values and each of the second plurality of values are combined into a plurality of groups. Then, the computer detects the abnormality according to the change in the number of pairs included in each of the plurality of groups.

According to one aspect, the detection accuracy of abnormal data can be improved.

It is a figure which shows an example of the abnormality detection method which concerns on 1st Embodiment. It is a figure which shows an example of the system which concerns on 2nd Embodiment. It is a figure which shows an example of the hardware of a server. It is a block diagram which shows the monitoring function of a ship in a server. It is a figure which shows an example of the measurement data stored in a data storage part. It is a block diagram which shows the detailed function of a preprocessing part. It is a figure which shows an example of the correlation frequency storage part. It is a figure which shows an example of an abnormality history storage part. It is a figure which shows the generation example of the correlation diagram. It is a figure which shows an example of the region-divided coordinate system. It is a figure which shows the meaning of the point plotted in each area. It is a figure which shows an example of the region which shows the correlation characteristic of group A. It is a figure which shows an example of the region which shows the correlation of group B. It is a figure which shows the judgment example of the presence or absence of abnormality of the measurement data. It is a flowchart which shows an example of the procedure of abnormality data detection processing. It is a flowchart which shows an example of the procedure of the correlation pair selection process. It is a flowchart which shows an example of the procedure of the correlation characteristic investigation processing. It is a flowchart which shows an example of the procedure of the correlation characteristic extraction processing. It is a flowchart which shows an example of the procedure of A group classification processing. It is a flowchart which shows an example of the procedure of B group classification processing. It is a flowchart which shows an example of the procedure of C group classification processing. It is a figure which shows the difference of the correlation diagram depending on whether or not the rate of change is calculated. It is a figure which shows an example of the setting method of the time interval ΔT in the change rate calculation. It is a figure which shows the comparison example of the time series data, and the time change of the change rate after the change rate calculation. It is a figure which shows an example of the generated correlation diagram. It is a figure which shows an example of the correlation diagram when the correlation characteristic is broken.

Hereinafter, the present embodiment will be described with reference to the drawings. It should be noted that each embodiment can be implemented by combining a plurality of embodiments within a consistent range.
[First Embodiment]
FIG. 1 is a diagram showing an example of an abnormality detection method according to the first embodiment. FIG. 1 shows an example in which the abnormality detection method is carried out by using the information processing apparatus 10. The information processing device 10 can execute the processing related to the abnormality detection method by executing, for example, an abnormality detection program in which the processing procedure of the abnormality detection method is described.

The information processing device 10 has a storage unit 11 and a processing unit 12. The storage unit 11 is, for example, a memory or a storage device included in the information processing device 10. The processing unit 12 is, for example, a processor or an arithmetic circuit included in the information processing device 10.

The first sensor 1 and the second sensor 2 are connected to the information processing device 10 via a network. The first sensor 1 and the second sensor 2 are, for example, sensors attached to a device (managed device) whose state is to be managed. Controlled equipment also includes equipment used outdoors, such as ships and automobiles.

The storage unit 11 stores the time-series measured values measured by the first sensor 1 and the second sensor 2. For example, in the storage unit 11, for each predetermined unit period P1 and P2, a fixed time interval is set by each of the first sensor 1 and the second sensor 2 within the corresponding unit period (time interval shorter than the unit period P1 and P2). Memorize the measured value measured in.

The processing unit 12 acquires the time-series measured values measured by the first sensor 1 and the second sensor 2 from the first sensor 1 and the second sensor 2, and stores the measured values in the storage unit 11. .. Then, the processing unit 12 detects abnormal data that may be included in the measured values of the first sensor 1 or the second sensor 2 based on the measured values stored in the storage unit 11. Abnormal data is the measured value when spike noise or measurement error occurs. For example, the first sensor 1 or the second sensor 2 outputs abnormal data as a measured value due to the influence of disturbance or the like. If the data measured within a certain unit period contains abnormal data, it is inappropriate to use the data measured within that unit period in analysis processing such as machine learning. Therefore, the processing unit 12 detects the abnormal data from the time-series measured values measured by the first sensor 1 and the second sensor 2, and prevents the data including the abnormal data from being used in the analysis process. do.

For example, the processing unit 12 acquires a first plurality of values obtained based on the measurement of the first sensor 1 and a second plurality of values obtained based on the measurement of the second sensor 1. For example, the processing unit 12 acquires the rate of change of the predetermined time interval ΔT of the measured value obtained by the measurement of the first sensor 1 as the first plurality of values, and is obtained by the measurement of the second sensor 2. The rate of change of the measured value at a predetermined time interval ΔT is acquired as a second plurality of values. The rate of change is, for example, the average rate of change during a predetermined time interval ΔT. For example, when the time change of the measured value can be expressed by the equation f (T) (T is a variable indicating the time), the average rate of change is obtained by the difference quotient "{f (T + ΔT) -f (T)} / ΔT". Can be done. The processing unit 12 can also perform a differential operation of the equation f (T) and use the differential coefficient f'(T) for each time interval ΔT as the rate of change.

Next, the processing unit 12 classifies a plurality of sets in which each of the first plurality of values and each of the second plurality of values are combined into a plurality of groups. Each set is a set of values obtained by measuring the same time in the first sensor 1 and the second sensor 2. For example, one set is a set of the rate of change of the measured value of the first sensor 1 at a certain time and the rate of change of the measured value of the second sensor 2 at the same time.

For example, the processing unit 12 can classify a plurality of sets based on the positions on a predetermined coordinate system in which the points corresponding to each set are plotted. Specifically, the processing unit 12 generates a coordinate system having a first coordinate axis corresponding to the first plurality of values and a second coordinate axis corresponding to the second plurality of values. The processing unit 12 generates a plurality of regions corresponding to any of the plurality of groups on the generated coordinate system. Then, the processing unit 12 classifies each of the plurality of sets into a group corresponding to the area to which the points having the two values included in the set as the coordinate values belong.

In the example of FIG. 1, the processing unit 12 includes, as a plurality of regions, a region including the origin of the coordinate system, a region including the first coordinate axis, a region including the second coordinate axis, and a first coordinate axis and a second coordinate axis. The area that does not contain is generated. The area including the origin of the coordinate system is associated with the A group. The area including the first coordinate axis and the area including the second coordinate axis are associated with the B group. The area that does not include the first coordinate axis and the second coordinate axis is associated with the C group. In this case, the processing unit 12 classifies each of the plurality of sets into a group corresponding to the area to which the point indicating the set belongs among the three groups A, B, and C.

Then, the processing unit 12 detects an abnormality according to a change in the number of sets included in each of the plurality of groups. For example, the processing unit 12 detects an abnormality in data when the ratio of the number of pairs included in each of the plurality of groups changes with the passage of time.

Specifically, when the processing unit 12 acquires the first plurality of values and the second plurality of values, the first plurality of values and the second plurality of values are obtained for each unit period P1 and P2. And get. Next, the processing unit 12 executes the classification of a plurality of sets for each unit period P1 and P2. Further, the processing unit 12 sorts by the number of pairs including a plurality of groups for each unit period P1 and P2. Then, the processing unit 12 determines that there is an abnormality when the sorted order of the plurality of groups in the first unit period and the sorted order of the plurality of groups in the second unit period do not match.

In the example of FIG. 1, in the unit period P1, many pairs are included in the B group, but are not included in the C group. On the other hand, in the unit period P2, the B group contains a small amount of pairs, but the C group contains many pairs. That is, in the unit period P1 and the unit period P2, the number of pairs included in the B group and the number of pairs included in the C group are reversed. In this case, the processing unit 12 detects that there is an abnormality in the data.

In this way, it is possible to detect an abnormality in the data. In the detection of such an abnormality of data, it is not necessary to set a threshold value for determining an abnormality assuming a specific event, and it is possible to detect an abnormality of data caused by an unknown event. As a result, the accuracy of detecting data abnormalities is improved.

By the way, the abnormal data output from the sensor often outputs a value that is significantly different from the value taken by the physical quantity that the sensor should originally measure. Therefore, in the detection of abnormal data, it becomes easy to detect the abnormal data by converting so as to emphasize the sudden change in the measured value.

Therefore, for example, the processing unit 12 acquires the rate of change of the measured value of the first sensor 1 and the rate of change of the measured value of the second sensor 2 as the first plurality of values and the second plurality of values. .. This emphasizes abrupt changes in the measured values. In the example of FIG. 1, the measured value of the first sensor 1 is rapidly increasing at a specific time. Taking the rate of change of the measured value of the first sensor 1, an extremely large value is detected at the time when the measured value suddenly changes. By emphasizing the sudden change in the measured value, the accuracy of detecting the measured value with the abnormal value is improved.

Further, the processing unit 12 can detect the collapse of the correlation appearing in the correlation diagram by classifying each set into a group corresponding to the area to which the points having the values included in the set as the coordinate values belong. For example, it is assumed that the first sensor 1 and the second sensor 2 are attached to a certain managed object. At this time, there is a certain kind of value between the first plurality of values obtained based on the measurement of the first sensor 1 and the second plurality of values obtained based on the measurement of the second sensor 2. It may be known that there is a correlation (positive correlation, negative correlation, no correlation, etc.). The correlation appears in the way the points plotted on the correlation diagram are scattered. Therefore, the processing unit 12 determines a group for classifying the set corresponding to the point according to which region among the plurality of regions set in the coordinate system representing the correlation diagram belongs to the point on the correlation diagram. As a result, when the correlation is broken, the number of pairs included in each group fluctuates greatly. As a result, it is possible to detect the collapse of the correlation indicating the possibility that abnormal data has occurred according to the change in the number of pairs included in the group.

When setting a plurality of regions on the coordinate system, the processing unit 12 can set, for example, a region near the origin of the coordinate system, a region near the seat axis, and a region far from the origin and the coordinate axis. Setting such a region is particularly effective when the points on the coordinate system corresponding to each of the plurality of sets tend to gather around the origin. For example, when the rate of change of the measured value of the first sensor 1 or the second sensor 2 is set to the first plurality of values or the second plurality of values, if the change of the measured value of each sensor tends to be small, Points corresponding to multiple pairs are plotted near the origin of the coordinate system. Then, the state of correlation can be easily determined by how many points are plotted in a region other than the region near the origin.

The processing unit 12 determines whether or not the number of pairs included in each of the plurality of groups has changed, for example, based on the difference in the order of the sorted groups based on the number of pairs included in the plurality of groups for each unit period. to decide. This makes it possible to detect an abnormality without presetting a threshold value for determining whether or not the value obtained based on the measurement of the sensor is abnormal. As a result, it becomes easy to detect a data abnormality due to an unexpected disturbance.

The number of sensors attached to the object to be managed may be three or more. In this case, the processing unit 12 generates, for example, a plurality of sensor pairs obtained by selecting two sensors from a plurality of sensors attached to the object to be managed. Then, the processing unit 12 obtains a first plurality of values and a second plurality of values for each sensor pair by using one of the sensor pairs as the first sensor and the other as the second sensor. Classify groups and detect abnormalities. By detecting anomalies for each sensor pair, the accuracy of anomaly detection can be improved.

When anomaly detection is performed for each sensor pair, for example, the processing unit 12 determines whether or not there is a measurement abnormality in at least a part of the plurality of sensors based on the ratio of the sensor pair in which the abnormality is detected among the plurality of generated sensor pairs. To judge. That is, when the processing unit 12 detects a data abnormality in a sensor pair having a predetermined ratio or more, it is considered that the managed object is under the influence of a strong disturbance, and there is a measurement abnormality in some sensors, and the measured value. Judges that abnormal data is included in. If some sensors have measurement anomalies, the overall reliability of the values obtained based on the measurements of all sensors will be low. Therefore, when the processing unit 12 determines that there is a measurement abnormality of at least a part of the sensors, the processing unit 12 suppresses the use of the entire value obtained based on the measurement of the plurality of sensors in the analysis processing. This improves the reliability of the analysis process using the value obtained based on the measurement of the sensor.

[Second Embodiment]
The second embodiment detects the presence or absence of abnormal data caused by the sensor measurement abnormality, not the abnormality related to the operation state of the ship, from the measurement data including the measurement value measured by the sensor attached to the ship. Is.

FIG. 2 is a diagram showing an example of the system according to the second embodiment. A plurality of sensors 31 to 33 are attached to the ship 30 during voyage. The sensors 31 to 33 of the ship 30 include a sensor for measuring the shaft horsepower of the power, a sensor for measuring the ship speed, a sensor for measuring the rotation speed of the power shaft, and a wing angle response of a variable pitch propeller (CPP). Sensors that measure values, sensors that measure the instantaneous flow rate of fuel in the main engine (M / E: Main Engine), sensors that measure the instantaneous flow rate of fuel in the generator (D / G: Diesel Generator), true wind direction and true There are sensors that measure wind speed. The sensors 31 to 33 output measurement data at predetermined intervals. The measurement data includes, for example, a measured value which is a physical quantity measured by the sensors 31 to 33 and time information indicating a measurement time.

The ship 30 is equipped with a communication device 34 for satellite communication. The communication device 34 is connected to the sensors 31 to 33. Further, the communication device 34 can communicate with the communication satellite 41. The communication satellite 41 can communicate with the server 100 via the ground antenna 42. The communication device 34 transmits the measurement data output by the sensors 31 to 33 to the server 100 via the communication satellite 41.

The server 100 grasps, for example, the operation status of the ship 30 based on the measurement data acquired from the ship 30. For example, the server 100 can detect an abnormality of the ship 30 based on the measurement data.

FIG. 3 is a diagram showing an example of server hardware. The entire device of the server 100 is controlled by the processor 101. A memory 102 and a plurality of peripheral devices are connected to the processor 101 via a bus 109. The processor 101 may be a multiprocessor. The processor 101 is, for example, a CPU (Central Processing Unit), an MPU (Micro Processing Unit), or a DSP (Digital Signal Processor). At least a part of the functions realized by the processor 101 executing a program may be realized by an electronic circuit such as an ASIC (Application Specific Integrated Circuit) or a PLD (Programmable Logic Device).

The memory 102 is used as the main storage device of the server 100. At least a part of an OS (Operating System) program or an application program to be executed by the processor 101 is temporarily stored in the memory 102. Further, various data used for processing by the processor 101 are stored in the memory 102. As the memory 102, for example, a volatile semiconductor storage device such as a RAM (Random Access Memory) is used.

Peripheral devices connected to the bus 109 include a storage device 103, a graphic processing device 104, an input interface 105, an optical drive device 106, a device connection interface 107, and a network interface 108.

The storage device 103 electrically or magnetically writes and reads data to and from the built-in recording medium. The storage device 103 is used as an auxiliary storage device for a computer. The storage device 103 stores an OS program, an application program, and various data. As the storage device 103, for example, an HDD (Hard Disk Drive) or an SSD (Solid State Drive) can be used.

A monitor 21 is connected to the graphic processing device 104. The graphic processing device 104 causes the image to be displayed on the screen of the monitor 21 in accordance with the instruction from the processor 101. The monitor 21 includes a display device using an organic EL (Electro Luminescence), a liquid crystal display device, and the like.

A keyboard 22 and a mouse 23 are connected to the input interface 105. The input interface 105 transmits signals sent from the keyboard 22 and the mouse 23 to the processor 101. The mouse 23 is an example of a pointing device, and other pointing devices can also be used. Other pointing devices include touch panels, tablets, touchpads, trackballs and the like.

The optical drive device 106 reads the data recorded on the optical disk 24 by using a laser beam or the like. The optical disk 24 is a portable recording medium on which data is recorded so that it can be read by reflection of light. The optical disk 24 includes a DVD (Digital Versatile Disc), a DVD-RAM, a CD-ROM (Compact Disc Read Only Memory), a CD-R (Recordable) / RW (ReWritable), and the like.

The device connection interface 107 is a communication interface for connecting peripheral devices to the server 100. For example, a memory device 25 or a memory reader / writer 26 can be connected to the device connection interface 107. The memory device 25 is a recording medium equipped with a communication function with the device connection interface 107. The memory reader / writer 26 is a device that writes data to or reads data from the memory card 27. The memory card 27 is a card-type recording medium.

The network interface 108 is connected to the network 20. The network interface 108 transmits / receives data to / from another computer or communication device via the network 20.

The server 100 can realize the processing function of the second embodiment by the hardware as described above. The device shown in the first embodiment can also be realized by the same hardware as the server 100 shown in FIG.

The server 100 realizes the processing function of the second embodiment, for example, by executing a program recorded on a computer-readable recording medium. The program that describes the processing content to be executed by the server 100 can be recorded on various recording media. For example, a program to be executed by the server 100 can be stored in the storage device 103. The processor 101 loads at least a part of the program in the storage device 103 into the memory 102 and executes the program. Further, the program to be executed by the server 100 can be recorded on a portable recording medium such as an optical disk 24, a memory device 25, and a memory card 27. The program stored in the portable recording medium can be executed after being installed in the storage device 103 under the control of the processor 101, for example. The processor 101 can also read and execute the program directly from the portable recording medium.

The server 100 acquires the measurement data of the sensor from the ship 30 and monitors the operation status of the ship 30. For example, the server 100 monitors the presence or absence of anomalies in the ship 30 using a trained model generated by machine learning. At that time, if the measurement data of the sensor includes abnormality data due to the influence of the external environment, the server 100 may make an erroneous determination as to whether or not there is an abnormality in the ship 30. For example, when the server 100 determines whether or not there is an abnormality in the ship 30 based on the measurement data including the abnormality data, there is a possibility that the server 100 determines that the ship 30 has an abnormality even though the ship 30 has no abnormality. Therefore, the server 100 determines whether or not the measurement data acquired from the ship 30 includes the abnormality data caused by the measurement abnormality of the sensor before determining the presence or absence of the abnormality of the ship 30. When the measurement data includes the abnormality data, the server 100 suppresses the determination of the presence or absence of the abnormality of the ship 30 based on the measurement data including the abnormality data.

FIG. 4 is a block diagram showing a ship monitoring function on the server. The server 100 has a data acquisition unit 110, a data storage unit 120, a preprocessing unit 130, and a machine learning unit 140 as functions for monitoring the ship 30.

The data acquisition unit 110 acquires the measurement data of the sensor installed on the ship 30 from the ship 30 via a communication satellite line or the like. For example, the data acquisition unit 110 acquires the measurement data measured within the immediately preceding fixed time at regular time intervals. The data acquisition unit 110 stores the acquired measurement data in the data storage unit 120.

The data storage unit 120 stores measurement data. For example, the data storage unit 120 stores the measurement data of each of the plurality of sensors 31 to 33 for each unit period, with a period having a fixed time width (for example, several hours) as a unit period. The data storage unit 120 is realized, for example, by a part of the storage area of the memory 102 or the storage device 103.

The preprocessing unit 130 detects the presence or absence of abnormal data in the measurement data for each unit period. The preprocessing unit 130 excludes the measurement data for which the abnormal data is detected from the input data to the machine learning unit 140. Further, the preprocessing unit 130 inputs the measurement data for which the abnormality data has not been detected to the machine learning unit 140.

The machine learning unit 140 uses the DNN (Deep Neural Network) 141 to detect the presence or absence of an abnormality in the ship 30 based on the input measurement data. For example, when the measurement data for a unit period is input, the machine learning unit 140 uses the measurement data as the input data of the trained model to estimate the presence or absence of an abnormality using the trained model. The machine learning unit 140 can also train the model based on the input measurement data and generate or modify the training model.

The server 100 monitors the operating state of the ship 30 by coordinating such functions. The function of each element shown in FIG. 4 can be realized, for example, by causing a computer to execute a program module corresponding to the element.

FIG. 5 is a diagram showing an example of measurement data stored in the data storage unit. The data storage unit 120 stores measurement data 121, 122, ... For each unit period P1, P2, ....

The measurement data 121 for each unit period P1, P2, ... Includes time-series data 121a, 121b, ... For each sensor within the corresponding unit period. In the time series data 121a, 121b, ..., The data type of the data measured by the sensor is associated with the sensor ID of the sensor, and the sensor measures the data periodically (for example, every few seconds) within a unit period. The measured value of the physical quantity is set. The measurement time of the measured value is given to the measured value. Similar to the measurement data 121, the other measurement data 122, ... Also include time series data for each sensor.

Next, the details of the preprocessing unit 130 will be described.
FIG. 6 is a block diagram showing a detailed function of the preprocessing unit. The preprocessing unit 130 includes a measurement data input unit 131 and an abnormality data detection unit 132. The measurement data input unit 131 acquires measurement data for one unit period from the data storage unit 120. Then, the measurement data input unit 131 transmits the acquired measurement data to the abnormality data detection unit 132, and requests the abnormality data detection unit 132 to perform the abnormality data detection process. When the abnormality data detection unit 132 determines that the measurement data does not include the abnormality data, the measurement data input unit 131 inputs the measurement data to the machine learning unit 140 as input data for machine learning. Further, when the abnormality data detection unit 132 determines that the measurement data contains the abnormality data, the measurement data input unit 131 suppresses the input of the measurement data to the machine learning unit 140.

The abnormality data detection unit 132 performs an abnormality data detection process from the measurement data acquired from the measurement data input unit 131. The abnormality data detection unit 132 includes a correlation pair selection unit 132a, a data smoothing processing unit 132b, a change rate calculation unit 132c, a correlation characteristic monitoring unit 132d, a correlation frequency storage unit 132e, an abnormality determination unit 132f, an abnormality history storage unit 132g, and an abnormality. It has a data removing unit 132h.

The correlation pair selection unit 132a selects a correlation pair for monitoring the correlation characteristics between the measured values. The correlation pair is an example of the sensor pair shown in the first embodiment. For example, when the abnormal data has not been detected in the past, the correlation pair selection unit 132a comprehensively extracts a combination of any two sensors, and sets the extracted sensor set as a correlation pair. Exhaustive extraction means that each sensor is extracted so as to be included in at least one correlation pair. By comprehensively extracting, it is possible to suppress the generation of sensors that are excluded from the detection of abnormal data.

The correlation pair selection unit 132a can also select all the combinations that can be selected when arbitrarily selecting two from all the sensors as a correlation pair. However, if such a round-robin selection of correlation pairs is performed when there are a large number of sensors, the number of correlation pairs becomes enormous, and the calculation load for abnormal data detection becomes excessive. Therefore, the correlation pair selection unit 132a suppresses the leakage of the sensor from the monitoring target of the abnormality data detection processing and suppresses the processing load by selecting the correlation pair at least comprehensively.

The correlation pair selection unit 132a refers to, for example, the abnormality history storage unit 132g, and sets a pair of sensors included in the correlation pair that is frequently determined to have abnormality data in the past, and the correlation pair to be monitored in this unit period. Priority is given to selection. As a result, a set of sensors having a high possibility of detecting abnormal data can be selected as a correlation pair.

The data smoothing processing unit 132b smoothes data by removing measurement noise (power supply noise, etc.) by filtering, moving average method, or the like. For example, the data smoothing processing unit 132b removes a frequency component higher than a predetermined cutoff frequency from the frequency component of the time series data measured by the sensor by a low-pass filter. Further, when the moving average method is applied, the data smoothing processing unit 132b replaces two measured values having consecutive measurement times by the sensor with the average value of those measured values. At this time, the data smoothing processing unit 132b sets the measurement time of the measured value after the replacement as an intermediate time between the continuous measurement times before the replacement.

The change rate calculation unit 132c calculates the change rate of the time change of the measured value in the smoothed data. By calculating the rate of change, the time change of the amount of change (rate of change) of the measured value per unit time can be obtained. For example, when the time-series data of the sensor represents the position of the ship 30 at each measurement time, the time-series change of the speed of the ship 30 can be obtained by calculating the rate of change. Specifically, the rate of change calculation unit 132c divides the difference between the two measured values measured at two times T1 and T2 (T2-T1 = ΔT) at a fixed time interval ΔT by the time interval ΔT. The rate of change calculation unit 132c uses the division result as the rate of change in the period (T1 to T2) between those times.

The rate of change calculation unit 132c determines the time interval ΔT based on the sharpness of the rise or fall of the waveform of the curve representing the time change of the measured value in the original time series data, or the sampling frequency. For example, the rate of change calculation unit 132c obtains the sampling period based on the sampling frequency of the physical quantity measured by the sensor, and sets N times the sampling period (N is a positive real number) as the time interval ΔT.

The correlation characteristic monitoring unit 132d obtains a correlation diagram between the rate of change obtained from each of the time-series data of the sensors selected as the correlation pair, and quantifies the correlation characteristic. For example, the correlation characteristic monitoring unit 132d classifies the points plotted on the correlation diagram into a plurality of groups based on the coordinates of the points. Then, the correlation characteristic monitoring unit 132d sets the number of points (correlation frequency) belonging to each of the plurality of groups obtained for the correlation pair as the characteristic of the correlation pair. The correlation characteristic monitoring unit 132d stores the correlation frequency calculated for each of the plurality of correlation pairs in the correlation frequency storage unit 132e.

In the classification of the points plotted on the correlation diagram, the correlation characteristic monitoring unit 132d is, for example, A group near the origin, B group near the x coordinate axis, B'group near the y axis coordinate, and C in the first or third quadrant. Classify each point into a group, a C'group within the second or fourth quadrant. The B group and the B'group may be combined into one group. Further, the C group and the C'group may be combined into one group.

The correlation frequency storage unit 132e stores the calculation history of the correlation frequency for each correlation pair. The correlation frequency storage unit 132e is, for example, a part of the storage area of the memory 102 or the storage device 103.

The abnormality determination unit 132f compares the correlation frequencies obtained for each unit period with respect to the correlation pair to be detected, and determines the presence or absence of an abnormality. For example, the abnormality determination unit 132f creates a histogram of the correlation frequency of each group for each unit period of the correlation pair, and compares the histograms for each unit period. Then, when the histogram in the latest unit period has different characteristics from the histogram in the previous unit period, the abnormality determination unit 132f determines that the measurement data in the latest unit period includes the abnormality data.

When the abnormality determination unit 132f determines that the abnormality data is included for at least a part of the correlation pairs, the abnormality determination unit 132f notifies the abnormality data removal unit 132h that there is an abnormality data. Further, the abnormality determination unit 132f counts up the cumulative number of abnormalities of the correlation pair determined to include the abnormality data in the abnormality history storage unit 132g.

The abnormality history storage unit 132g stores the cumulative number of abnormalities indicating the number of times abnormal data is detected for each correlation pair. The abnormality history storage unit 132g is, for example, a part of the storage area of the memory 102 or the storage device 103.

The abnormality data removing unit 132h removes the measurement data for the unit period determined to have the abnormality data from the input data to the machine learning unit 140. For example, the abnormal data removing unit 132h notifies the measurement data input unit 131 of information indicating that it is determined that there is abnormal data. Then, the measurement data input unit 131 recognizes that the measurement data for the unit period acquired from the data storage unit 120 contains abnormal data, and suppresses the input of the measurement data to the machine learning unit 140.

Next, with reference to FIGS. 7 and 8, the data stored in the correlation frequency storage unit 132e and the abnormality history storage unit 132g will be described in detail.
FIG. 7 is a diagram showing an example of the correlation frequency storage unit. The correlation frequency storage unit 132e stores the correlation frequency information 51, 52, ... For each unit period. The correlation frequency information 51 includes correlation frequency tables 51a, 51b, ... For each correlation pair generated when abnormal data is detected in the measurement data of the corresponding unit period. In the correlation frequency tables 51a, 51b, ..., The number of points (appearance frequency) included in the group is set in association with the group name of the group in which the points plotted on the correlation diagram are classified. The correlation frequency tables 51a, 51b, ... Show the sensor IDs of the corresponding correlation pairs. Similar to the correlation frequency information 51, the other correlation frequency information 52, ... Also includes a correlation frequency table for each correlation pair generated when anomalous data is detected in the measurement data of the corresponding unit period.

FIG. 8 is a diagram showing an example of an abnormality history storage unit. For example, the abnormality history management table 60 is stored in the abnormality history storage unit 132g. In the abnormality history management table 60, the cumulative number of abnormalities is set in association with the set of sensor IDs of the two sensors included in the correlation pair. The cumulative number of abnormalities indicates the number of times it is determined that there is abnormal data from the correlation frequency table of the corresponding correlation pair.

Next, the abnormality data detection process will be specifically described. In the anomaly data detection process, a correlation diagram of the rate of change between the correlation pairs is generated.
FIG. 9 is a diagram showing an example of generating a correlation diagram. In the example of FIG. 9, the pair of the sensor with the sensor ID "001" and the sensor with the sensor ID "002" is selected as a correlation pair. The rate of change calculation unit 132c calculates the rate of change at regular intervals for each sensor based on the time series data 121a and 121b of each sensor included in the correlation pair. As a result, the change rate information 61 of the sensor with the sensor ID "001" and the change rate information 62 of the sensor with the sensor ID "002" are obtained.

The sensor IDs of the corresponding sensors are assigned to the rate of change information 61 and 62 for each sensor. Further, in the change rate information 61 and 62, the change rate of the time series data at the corresponding time is set for each time interval ΔT. When the formula showing the time change of the measured value shown in the time series data 121a is f (x), the rate of change of the time series data 121a at time T11 is “{f (T11 + ΔT) −f (T11)} / ΔT. ". When the formula showing the time change of the measured value shown in the time series data 121b is g (x), the rate of change of the time series data 121b at time T11 is "{g (T11 + ΔT) -g (T11)} /. It becomes ΔT ”.

The correlation characteristic monitoring unit 132d generates a correlation diagram of the rate of change between the correlation pairs based on the rate of change information 61 and 62 for each sensor. For example, the correlation characteristic monitoring unit 132d uses the sensor with the sensor ID "001" as the first sensor and the sensor with the sensor ID "002" as the second sensor. The correlation characteristic monitoring unit 132d defines a coordinate system in which the rate of change of the first sensor is on the horizontal axis (x-axis) and the rate of change of the second sensor is on the vertical axis (y-axis).

The correlation characteristic monitoring unit 132d plots points in the defined coordinate system at positions corresponding to the rate of change of the first sensor and the rate of change of the second sensor at each time corresponding to the rate of change. do. For example, the point 63 of T11 is plotted at a position corresponding to the rate of change of the first sensor and the rate of change of the second sensor at time T11. Correlation diagram 64 is generated by the correlation characteristic monitoring unit 132d plotting the corresponding points for all the rate of change calculated based on the time series data within the unit period.

The correlation characteristic monitoring unit 132d determines that the time between the change rate information 61 and the change rate information 62 is the same time within a predetermined error range. For example, it is assumed that the sampling period of each sensor included in the correlation pair is 1 s. In this case, if the time shown in the change rate information 61 and the time shown in the change rate information 62 are the same in seconds (ignoring the decimal point), the correlation characteristic monitoring unit 132d has the same time. Judge that.

Further, the correlation characteristic monitoring unit 132d divides the coordinate system of the correlation diagram 64 into a plurality of regions.
FIG. 10 is a diagram showing an example of a region-divided coordinate system. The correlation characteristic monitoring unit 132d sets a region 65a within a predetermined range from the origin of the coordinate system 65 as a region representing the correlation characteristic of the A group. Correlation characteristic monitor unit 132d, the region 65b in the range of the width L ^B about the x axis, 65c (the excluded range included in region 65a), the region representing the correlation properties of the B group. Correlation characteristic monitor unit 132d has a width L ^B about the y-axis ^'area within the range of 65d, 65e (the excluded range included in region 65a), B' is a region representing the correlation properties of the group. The correlation characteristic monitoring unit 132d sets the region 65f in the first quadrant and the region 65g in the third quadrant, which are not included in any of the regions 65a to 65e, as regions representing the correlation characteristics of the C group. The correlation characteristic monitoring unit 132d sets the second quadrant region 65h and the fourth quadrant region 65i, which are not included in any of the regions 65a to 65e, as regions representing the correlation characteristics of the C'group.

FIG. 11 is a diagram showing the meaning of the points plotted in each area. The points plotted in the region 65a representing the correlation characteristics of group A indicate that the time series data of each of the two sensors of the correlation pair did not change significantly over time at the time corresponding to that point (no change). ) Is shown.

The points plotted in the regions 65b and 65c representing the correlation characteristics of group B show that the time series data of the first sensor changes significantly with time at the time corresponding to the points, but at the time of the second sensor. The series data show that there is no significant change over time. That is, at the time corresponding to the points plotted in the regions 65b and 65c representing the correlation characteristics of the B group, the change timings of the measured values shown in the time series data of the two sensors do not match.

The points plotted in the regions 65d and 65e representing the correlation characteristics of the B'group show that the time series data of the second sensor changes significantly with time at the time corresponding to the points, but that of the first sensor Time series data show that there is no significant time change. That is, at the time corresponding to the points plotted in the regions 65d and 65e representing the correlation characteristics of the B'group, the change timings of the measured values shown in the time series data of the two sensors do not match.

The points plotted in the regions 65f and 65g representing the correlation characteristics of the C group indicate that the values of the time series data of each of the two sensors of the correlation pair are increasing or decreasing at the time corresponding to the points. In the correlation characteristic of group C, when the value of the time series data of the first sensor is increasing, the value of the time series data of the second data is also increasing, and the value of the time series data of the first sensor is decreasing. When doing so, the value of the time series data of the second data also decreases. As described above, at the time corresponding to the points plotted in the regions 65f and 65g representing the correlation characteristics of the C group, the change timings of the measured values shown in the time series data of the two sensors are the same.

The points plotted in the regions 65h and 65i representing the correlation characteristics of the C'group indicate that the values of the time-series data of each of the two sensors of the correlation pair are increasing or decreasing at the time corresponding to the points. .. In the correlation characteristic of the C'group, when the value of the time series data of the first sensor is increasing, the value of the time series data of the second data is decreasing, and the value of the time series data of the first sensor is increasing. When it is decreasing, the value of the time series data of the second data is increasing. As described above, at the time corresponding to the points plotted in the regions 65h and 65i representing the correlation characteristics of the C'group, the change timings of the measured values shown in the time series data of the two sensors are the same.

The region 65a representing the correlation characteristic of Group A is, for example, an ellipse calculated based on the degree of variation in the rate of change of the first sensor and the degree of variation in the rate of change of the second sensor.

FIG. 12 is a diagram showing an example of a region representing the correlation characteristic of Group A. For example, the correlation characteristic monitoring unit 132d calculates the degree of variation in the rate of change of each correlation pair. The variation in the rate of change can be represented by, for example, the variance value. Assuming that the variance value of one of the correlation pairs (first sensor) is s _x ² and the variance value of the other (second sensor) is _sy ² , each variance value can be calculated by the following equation.

In the equations (1) and (2), n is the total number of elements (a set of the rate of change of the first sensor and the rate of change of the second sensor at a certain time) (n is an integer of 1 or more). x _i is the rate of change of the first sensor in the i-th element (i is an integer of 1 or more and n or less). y _i is the rate of change of the second sensor in the i-th element. The overlined x is the average value of the rate of change of the first sensor. The overlined y is the average value of the rate of change of the second sensor.

Next, the correlation characteristic monitoring unit 132d generates an equation of an ellipse 66 indicating the outer edge of the region 65a near the origin based on the variance value. For example correlation characteristic monitor unit 132d is the rate of change of the first sensor of variance s _x ² a, when the rate of change of the second sensor the variance s _y ² and is b, the length of one axis Generates an ellipse 66 with a length of 2a (= 2s _x ² ) and a length of the other axis of 2b (= 2s _y ^2). Such an ellipse 66 is represented by the following equation.
x ² / a ² + y ² / b ² = 1 (3)
When a ≧ b, the positions of the focal points F and F'of this ellipse 66 are F ((a ² −b ² ) ^1/2 , 0), F'(− (a ² −b ² ) ^1/2 , respectively. It becomes 0). The correlation characteristic monitoring unit 132d sets the inside of such an ellipse 66 as a region 65a representing the correlation characteristic of the A group.

The correlation characteristic monitoring unit 132d can also expand the ellipse 66 by taking a margin of about 10% from the dispersion value of each correlation pair. When a margin of 10% is taken, the correlation characteristic monitoring unit 132d sets a = 1.1 × s _x ² and b = 1.1 × s _y ² for a and b in the equation (3) of the ellipse 66.

By adjusting the amount of such a margin, it is possible to adjust the sensitivity of normal / abnormal determination. For example, when the ship 30 is sailing calmly at a constant speed, it is expected that the time series data of most sensors will not change significantly. Then, most of the points plotted based on the rate of change obtained from the time series data of those sensors gather near the origin. In this case, there is a possibility of abnormal data when a large number of points plotted at positions outside the region 65a representing the correlation characteristic of group A occur. By increasing the amount of the margin, the region 65a represented by the ellipse 66 becomes wider. Then, the ratio of the plotted points included in the A group increases. That is, the sensitivity for determining an abnormality is lowered. On the other hand, by reducing the amount of the margin, the region 65a represented by the ellipse 66 becomes narrower. Then, the ratio of the plotted points included in the A group becomes low. That is, the sensitivity for determining an abnormality is increased.

The correlation characteristic monitoring unit 132d plots when the conditional branch (if statement) is true by executing the following processing on the points (xi, yi) using the generated ellipse 66 equation (3). The characteristics indicated by the points marked with are judged to be group A.
if (1> = xi ^ 2 / a ^ 2 + yi ^ 2 / b ^ 2) {
coordinate data = true} # A group.
elseif (1 <xi ^ 2 / a ^ 2 + yi ^ 2 / b ^ 2) {
coordinate data = false} # Not in group A.

In this way, it can be determined whether or not the plotted points are group A. The correlation characteristic monitoring unit 132d determines whether or not the correlation characteristic is the B group or the B'group with respect to the point determined that the correlation characteristic is not the A group.

FIG. 13 is a diagram showing an example of a region representing the correlation of the B group. For example correlation characteristic monitor unit 132d has a variance s _y ² rate of change of the second sensor, and the y-axis direction of the width L ^B of the region indicating the correlation of B group. In this case, the correlation characteristic monitor unit 132d is the point (xi, yi) y-coordinate values of "yi" is, ^{"- L B / 2 <yi <} L B / 2 " satisfies the correlation characteristics indicated by the point Is determined to be group B.

For example, the correlation characteristic monitoring unit 132d can determine the correlation characteristic of the plotted points as the B group by executing the following processing when the conditional branch (if statement) is true.
if (-LB / 2 <= yi <= LB / 2) {
coordinate data = true} # B group.
elseif (-LB / 2> yi || LB / 2 <yi) {
coordinate data = false} # Not in B group.

The correlation characteristic monitor unit 132d has a dispersion value s _x ² rate of change of the first sensor, and B 'width L ^B of the x-axis direction of a region showing a correlation characteristic of the ^group'. In this case, if the x-coordinate value "xi" of the point (xi, yi) ^{satisfies "-LB'} / 2 <xi <LB ^' / 2", the correlation characteristic monitoring unit 132d is indicated at that point. It is judged that the correlation characteristic is the B'group.

For example, the correlation characteristic monitoring unit 132d can determine the correlation characteristic indicated by the plotted points when the conditional branch (if statement) is true as the B'group by executing the following processing.
if (-LB'/2 <= xi <= LB' / 2) {
coordinate data = true} # B'group.
elseif (-LB'/2> xi || LB' / 2 <xi) {
coordinate data = false} # Not in the B'group.

Note the correlation characteristic monitor unit 132d takes the margin of about 10% from the variance value of each correlation pair, it is also possible to enlarge the width L ^B or the width L ^{B '.} When taking the 10% margin, the correlation characteristic monitor unit 132d has a width _{^{L B = 1.1 × s y 2}} , and the width _{^{L B '= 1.1 × s x}} 2.

In this way, it is possible to determine whether or not the correlation characteristics of the plotted points are in the B group and in the B'group. The correlation characteristic monitoring unit 132d determines whether or not the group is a C group or a C'group with respect to a point determined to be none of the A group, the B group, and the B'group.

For example, when the plotted point (xi, yi) is in the first quadrant (xi> 0, yi> 0), the correlation characteristic monitoring unit 132d determines that the correlation characteristic indicated by that point is group C. Even when the plotted points (xi, yi) are in the third quadrant (xi <0, yi <0), the correlation characteristic monitoring unit 132d determines that the correlation characteristic indicated by that point is group C. ..

When the plotted point (xi, yi) is in the second quadrant (xi <0, yi> 0), the correlation characteristic monitoring unit 132d determines that the correlation characteristic indicated by that point is in the C'group. The correlation characteristic monitoring unit 132d determines that the correlation characteristic indicated by the plotted point (xi, yi) is in the C'group even when it is in the fourth quadrant (xi> 0, yi <0). do.

When the determination of the correlation characteristic indicated by the points for all the plotted points is completed, the correlation characteristic monitoring unit 132d counts the frequency of each group of the correlation characteristics and generates a histogram. Then, the abnormality determination unit 132f determines whether or not there is an abnormality in the measurement data of the sensor of the corresponding correlation pair based on the histogram for each correlation pair.

FIG. 14 is a diagram showing an example of determining whether or not there is an abnormality in the measurement data. Was monitored correlation characteristics for a correlated pairs, the histogram 71 of the unit period P _1, point A group is most often, then many respects point or B 'groups of Group B, a point C group the next Or the points of the C'group increased. It is assumed that the same tendency continues until the unit period P _x-1.

Then, in _{the histogram 72 of the unit period P x} , the points of the A group were the most, the points of the C group or the C'group were the most, and then the points of the B group or the B'group were the most. .. That is, from the unit period P ₁ to the unit period P _x-1 , the points of the group B or the group B'are more than the points of the group C or the group C', but in the unit period P _x , the points of the group C or the group C' There are more points than points in group B or group B'. When the frequency of appearance of points in each group changes significantly (when the correlation is broken), the abnormality determination unit 132f has an abnormality in the measurement data of at least one sensor of the monitored correlation pair. Is determined.

For example, the abnormality determination unit 132f arranges the groups of correlation characteristics for each unit period in descending order of appearance frequency of the plotted points. If there is no change in the arrangement of groups in a predetermined number of consecutive unit periods, and the arrangement of groups in the next unit period of those unit periods is different from that before that, the abnormality determination unit 132f indicates that there is an abnormality in the measurement data. judge.

The abnormality determination unit 132f may treat the B group and the B'group together as one group. Further, the abnormality determination unit 132f may treat the C group and the C'group together as one group.

Next, the abnormality data detection process by the abnormality data detection unit 132 will be described in detail.
FIG. 15 is a flowchart showing an example of the procedure for abnormal data detection processing. Hereinafter, the process shown in FIG. 15 will be described along with the step numbers.

[Step S101] The correlation pair selection unit 132a acquires measurement data for one unit period scheduled to be input to the machine learning unit 140 by the measurement data input unit 131 from the measurement data input unit 131.

[Step S102] The correlation pair selection unit 132a selects a correlation pair for monitoring the correlation characteristics in the corresponding unit period. Details of the correlation pair selection process will be described later (see FIG. 16).
[Step S103] The correlation pair selection unit 132a selects one unselected correlation pair whose correlation characteristics have not been investigated from the selected plurality of correlation pairs. The correlation pair selection unit 132a transmits the information indicating the selected correlation pair and the time series data acquired from the two sensors corresponding to the correlation pair to the data smoothing processing unit 132b.

[Step S104] The data smoothing processing unit 132b, the rate of change calculation unit 132c, the correlation characteristic monitoring unit 132d, and the abnormality determination unit 132f are linked to perform the correlation characteristic investigation processing. The details of the correlation characteristic investigation process will be described later (see FIG. 17).

[Step S105] The abnormality determination unit 132f branches the process according to the result of the correlation characteristic investigation process. For example, if the abnormality determination unit 132f determines that there is an abnormality in the data, the process proceeds to step S106. If the abnormality determination unit 132f determines that there is no abnormality in the data, the abnormality determination unit 132f advances the process to step S108.

[Step S106] The abnormality determination unit 132f adds "1" to the number of abnormal pairs in the unit period to be investigated. The number of abnormal pairs is a variable whose initial value is "0", and is stored in, for example, the memory 102 during the abnormal data detection process.

[Step S107] The abnormality determination unit 132f adds "1" to the cumulative number of abnormalities corresponding to the selected correlation pair in the abnormality history storage unit 132g.
[Step S108] The abnormality determination unit 132f determines whether or not the correlation characteristics have been investigated for all the selected correlation pairs. If there is a correlation pair that has not been investigated, the abnormality determination unit 132f advances the process to step S103. Further, if the correlation characteristic investigation is completed for all the selected correlation pairs, the abnormality determination unit 132f proceeds to the process in step S109.

[Step S109] The abnormality determination unit 132f determines whether or not the number of abnormal pairs in the unit period to be investigated is equal to or greater than a predetermined threshold value. If the number of abnormal pairs is equal to or greater than the threshold value, the abnormality determination unit 132f advances the process to step S110. If the number of abnormal pairs is less than the threshold value, the abnormality determination unit 132f advances the process to step S111.

[Step S110] The abnormality data removing unit 132h removes the input data for the corresponding unit period. For example, the abnormality data removing unit 132h instructs the measurement data input unit 131 to suppress the input of the acquired measurement data for the unit period to the machine learning unit 140. The measurement data input unit 131 suppresses the input of the measurement data to the machine learning unit 140 in accordance with the instruction from the abnormality data removal unit 132h. After that, the abnormality data detection unit 132 ends the abnormality data detection process.

[Step S111] The abnormality data removing unit 132h instructs the measurement data input unit 131 to input the acquired measurement data to the machine learning unit 140. The measurement data input unit 131 inputs the acquired measurement data to the machine learning unit 140 as input data to the DNN in machine learning according to the instruction. After that, the abnormality data detection unit 132 ends the abnormality data detection process.

In this way, when it is determined that the time series data may be abnormal data in a predetermined number or more of the correlation pairs among the plurality of correlation pairs, the machine learning unit 140 of the measurement data of the corresponding unit period is reached. Input is suppressed. As a result, it is possible to prevent the ship 30 from being erroneously determined that an abnormality has occurred based on the measurement data including the abnormality data. Moreover, since the presence or absence of abnormality in the measurement data is determined based on the investigation result of the presence or absence of abnormality in a plurality of correlation pairs, it is highly reliable whether or not the measurement data may contain abnormality data. The judgment result can be obtained.

FIG. 16 is a flowchart showing an example of the procedure of the correlation pair selection process. Hereinafter, the process shown in FIG. 16 will be described along with the step numbers.
[Step S121] The correlation pair selection unit 132a refers to the abnormality history storage unit 132g, and selects the sensor group having the largest cumulative number of abnormalities among the sensor groups not selected as the correlation pair in the current abnormality data detection process. , Select as the correlation pair to be the target of this correlation characteristic survey.

[Step S122] The correlation pair selection unit 132a determines whether or not all the sensors are covered in the set of sensors already selected as the correlation pair. If there is a sensor that is not included in any of the selected correlation pairs, the correlation pair selection unit 132a advances the process to step S121. Further, the correlation pair selection unit 132a advances the process to step S123 when all the sensors are included in any of the selected correlation pairs.

[Step S123] The correlation pair selection unit 132a determines whether or not a predetermined number or more of correlation pairs have been selected. When the correlation pair selection unit 132a can select a predetermined number or more of the correlation pairs, the correlation pair selection process ends the correlation pair selection process. If the number of selected correlation pairs is less than a predetermined number, the correlation pair selection unit 132a advances the process to step S121.

In this way, the pair of sensors with the highest cumulative number of abnormalities is selected as a correlation pair in order. That is, the set of sensors determined to have the possibility of data abnormality in the past abnormality data detection processing is preferentially selected as the correlation pair in the current abnormality data detection processing. As a result, it is possible to prevent sensors that are extremely unlikely to output abnormal data from being included in a large number of correlation pairs, and it is possible to avoid processing with low effect. As a result, the processing efficiency of the abnormality data detection process is improved.

FIG. 17 is a flowchart showing an example of the procedure of the correlation characteristic investigation process. Hereinafter, the process shown in FIG. 17 will be described along with the step numbers.
[Step S131] The data smoothing processing unit 132b smoothes the time series data of each of the two sensors of the selected correlation pair. The data smoothing processing unit 132b removes measurement noise (power supply noise, etc.) by using, for example, filtering or a moving average method for data smoothing.

[Step S132] The rate of change calculation unit 132c determines the time interval ΔT used for calculating the rate of change. For example, the rate of change calculation unit 132c determines the time interval ΔT based on the sampling cycle of the measurement of the sensor of the correlation pair. Specifically, the rate of change calculation unit 132c sets the time interval ΔT as a value obtained by multiplying the sampling period of one sensor by a predetermined integer.

[Step S133] The change rate calculation unit 132c performs change rate calculation processing for each smoothed time-series data for each sensor, and calculates the change rate (change rate) of the measured value at each time of a predetermined time interval. .. For example, the rate of change calculation unit 132c divides the value obtained by subtracting the measured value at the start time from the measured value at the end time of the time zone of the time interval ΔT determined in step S132 by the time interval ΔT. Then, the rate of change calculation unit 132c sets the division result as the rate of change at the time within the corresponding time zone.

[Step S134] The correlation characteristic monitoring unit 132d uses one of the two sensors of the correlation pair selected as the investigation target as the first sensor and the other as the second sensor. Then, the correlation characteristic monitoring unit 132d includes the first sensor and the second sensor based on the change rate information of each of the first sensor and the second sensor (including the change rate of each time at a predetermined time interval). Generate a correlation diagram of the rate of change between.

[Step S135] The correlation characteristic monitoring unit 132d performs the correlation characteristic extraction process. For example, the correlation characteristic monitoring unit 132d divides the xy plane of the correlation diagram into a plurality of regions, and classifies the plotted points into a plurality of groups according to which region includes the plotted points. Then, the correlation characteristic monitoring unit 132d sets the appearance frequency of the points classified into each group as the correlation characteristic of the correlation pair to be investigated.

[Step S136] The abnormality determination unit 132f determines whether or not the correlation characteristic is broken. For example, the abnormality determination unit 132f acquires the correlation frequency table for a predetermined number of times generated in the past (before the unit period of the current processing target) for the selected correlation pair from the correlation frequency storage unit 132e. The abnormality determination unit 132f sorts each group by the frequency of appearance for each unit period based on the past correlation frequency table. When the order of the sorted groups for each unit period matches, the correlation frequency storage unit 132e sorts the groups according to the frequency of occurrence of each group of the correlation pair in the unit period of the current processing target. The correlation frequency storage unit 132e determines that the correlation characteristic has collapsed when the order of the groups based on the past correlation frequency table and the order of the groups based on the appearance frequency in the latest unit period are different.

When the correlation characteristic is broken, the abnormality determination unit 132f advances the process to step S138. Further, the abnormality determination unit 132f proceeds to step S137 if the correlation characteristic is not broken.

[Step S137] The abnormality determination unit 132f determines that the correlation characteristics of the correlation pair to be investigated are normal. After that, the abnormality determination unit 132f ends the correlation characteristic investigation process.

[Step S138] The abnormality determination unit 132f determines that the correlation characteristic of the correlation pair to be investigated is abnormal. After that, the abnormality determination unit 132f ends the correlation characteristic investigation process.

In this way, when the correlation characteristic in the unit period of the survey target is different from the previous one (when the correlation characteristic is broken), it is determined that the correlation characteristic is abnormal. As shown in FIG. 14, the collapse of the correlation characteristic can be determined by comparing the frequency of appearance of the plotted points in each group, so that it is not necessary to set a threshold value for determining an abnormality for each sensor in advance. Therefore, it is possible to detect a measurement abnormality due to an unexpected disturbance.

FIG. 18 is a flowchart showing an example of the procedure of the correlation characteristic extraction process. Hereinafter, the process shown in FIG. 18 will be described along with the step numbers.
[Step S141] The correlation characteristic monitoring unit 132d performs a classification process (A group classification process) for the points near the origin among the plotted points into the A group. Details of this process will be described later (see FIG. 19).

[Step S142] The correlation characteristic monitoring unit 132d performs a classification process (B group classification process) for points near the coordinate axes among the points not classified in the A group into the B group or the B'group. Details of this process will be described later (see FIG. 20).

[Step S143] The correlation characteristic monitoring unit 132d performs a classification process (C group classification process) for points that are not classified into either the A group or the B group into the C group or the C'group. Details of this process will be described later (see FIG. 21).

[Step S144] The correlation characteristic monitoring unit 132d performs a histogram generation process. For example, the correlation characteristic monitoring unit 132d counts the number of points included in the corresponding group for each group. The correlation characteristic monitoring unit 132d generates a histogram in which the counted number is used as the frequency of appearance of the corresponding group. The correlation characteristic monitoring unit 132d transmits the generated histogram to the abnormality determination unit 132f.

[Step S145] The correlation characteristic monitoring unit 132d stores the correlation frequency table in which the appearance frequency of points for each group is set in the correlation frequency storage unit 132e in association with the selected correlation pair and the unit time of the current investigation target. Store.

The points plotted in this way are classified and a histogram is generated. Next, the classification process for each group will be described in detail with reference to FIGS. 19 to 21.
First, the A group classification process will be described in detail.

FIG. 19 is a flowchart showing an example of the procedure of the A group classification process. Hereinafter, the process shown in FIG. 19 will be described along with the step numbers.
[Step S151] The correlation characteristic monitoring unit 132d has an x-coordinate value (rate of change obtained from the time-series data of the first sensor) and a y-coordinate value (from the time-series data of the second sensor) of each point on the correlation diagram. Obtain the obtained rate of change).

[Step S152] The correlation characteristic monitoring unit 132d calculates the variance value of the x-coordinate value.
[Step S153] The correlation characteristic monitoring unit 132d calculates the variance value of the y coordinate value.
[Step S154] The correlation characteristic monitoring unit 132d generates an elliptical formula showing the region 65a (see FIG. 10) of the A group. For example, the correlation characteristic monitoring unit 132d sets the length of the major axis or the minor axis to be twice the variance value of the x-coordinate value and the variance value of the y-coordinate value, respectively, and sets an ellipse in which the two focal points are on the x-axis or the y-axis. To generate.

[Step S155] The correlation characteristic monitoring unit 132d selects one unselected point on the correlation diagram.
[Step S156] The correlation characteristic monitoring unit 132d determines whether or not the selected point is within the region 65a of the A group. For example, when the coordinates of the selected point are within the ellipse generated in step S154, the correlation characteristic monitoring unit 132d determines that the point is within the region 65a of the A group. If the selected point is within the area 65a of the A group, the correlation characteristic monitoring unit 132d proceeds to step S157. If the selected point is not within the area 65a of the A group, the correlation characteristic monitoring unit 132d proceeds to step S158.

[Step S157] The correlation characteristic monitoring unit 132d classifies the selected points into group A. For example, the correlation characteristic monitoring unit 132d stores information indicating that the point belongs to the A group in the memory 102 in association with the identification number of the selected point.

[Step S158] The correlation characteristic monitoring unit 132d determines whether or not all the points on the correlation diagram have been selected. If all the points have been selected, the correlation characteristic monitoring unit 132d ends the A group classification process. If there is an unselected point, the correlation characteristic monitoring unit 132d advances the process to step S155.

FIG. 20 is a flowchart showing an example of the procedure of the B group classification process. Hereinafter, the process shown in FIG. 20 will be described along with the step numbers.
[Step S171] The correlation characteristic monitoring unit 132d acquires the x-coordinate value and the y-coordinate value of each point on the correlation diagram.

[Step S172] The correlation characteristic monitoring unit 132d calculates the variance value of the y coordinate value.
[Step S173] correlation characteristic monitor unit 132d determines the variance value of the y-coordinate values B group region 65b, the width L ^B of 65c (see FIG. 10).

[Step S174] The correlation characteristic monitoring unit 132d calculates the variance value of the x-coordinate value.
[Step S175] correlation characteristic monitor unit 132d determines the variance value of the x coordinate value B in the 'group of regions 65d, 65e (see FIG. 10) the width L ^B of ^the'.

[Step S176] The correlation characteristic monitoring unit 132d selects one of the points on the correlation diagram that has not been classified into any group and has not been selected in the B group classification process.
[Step S177] The correlation characteristic monitoring unit 132d determines whether or not the selected point is within the regions 65b and 65c of the B group. For example correlation characteristic monitor unit 132d includes, y coordinates of the selected point is -L ^B / 2 or more, and the case of L ^B / 2 or less, it is determined that point B Group regions 65b, to be within the 65c. If the selected point is within the regions 65b and 65c of the B group, the correlation characteristic monitoring unit 132d proceeds to step S178. If the selected point is not within the regions 65b and 65c of the B group, the correlation characteristic monitoring unit 132d proceeds to step S179.

[Step S178] The correlation characteristic monitoring unit 132d classifies the selected points into group B. For example, the correlation characteristic monitoring unit 132d stores information indicating that the point belongs to the B group in the memory 102 in association with the identification number of the selected point. The correlation characteristic monitoring unit 132d then proceeds to step S181.

[Step S179] The correlation characteristic monitoring unit 132d determines whether or not the selected point is within the regions 65d and 65e of the B'group. For example correlation characteristic monitor unit 132d is determined that the x-coordinate values of the selected point is -L ^B '/ 2 or more, and L ^B' if the / 2 or less, the area of the point B 'group 65d, are within 65e do. If the selected point is within the regions 65d and 65e of the B'group, the correlation characteristic monitoring unit 132d proceeds to step S180. If the selected point is not within the regions 65d and 65e of the B'group, the correlation characteristic monitoring unit 132d proceeds to step S181.

[Step S180] The correlation characteristic monitoring unit 132d classifies the selected points into the B'group. For example, the correlation characteristic monitoring unit 132d stores information indicating that the point belongs to the B'group in the memory 102 in association with the identification number of the selected point.

[Step S181] Whether or not the correlation characteristic monitoring unit 132d has not classified any of the points on the correlation diagram into any group and has selected all the unselected points in the B group classification process in step S176. To judge. If all the points have been selected, the correlation characteristic monitoring unit 132d ends the B group classification process. If there is an unselected point, the correlation characteristic monitoring unit 132d advances the process to step S176.

FIG. 21 is a flowchart showing an example of the procedure of the C group classification process. Hereinafter, the process shown in FIG. 21 will be described along with the step numbers.
[Step S191] The correlation characteristic monitoring unit 132d selects one point that has not been classified into any group among the points on the correlation diagram and has not been selected in the C group classification process.

[Step S192] The correlation characteristic monitoring unit 132d determines whether or not the selected point is within the first quadrant or the third quadrant. If the selected point is in the first quadrant or the third quadrant, the correlation characteristic monitoring unit 132d advances the process to step S193. If the selected point is not in the first quadrant or the third quadrant, the correlation characteristic monitoring unit 132d advances the process to step S194.

[Step S193] The correlation characteristic monitoring unit 132d classifies the selected points into the C group. For example, the correlation characteristic monitoring unit 132d stores information indicating that the point belongs to the C group in the memory 102 in association with the identification number of the selected point. The correlation characteristic monitoring unit 132d then proceeds to step S196.

[Step S194] The correlation characteristic monitoring unit 132d determines whether or not the selected point is in the second quadrant or the fourth quadrant. If the selected point is in the second quadrant or the fourth quadrant, the correlation characteristic monitoring unit 132d advances the process to step S195. If the selected point is not in the second quadrant or the fourth quadrant, the correlation characteristic monitoring unit 132d proceeds to step S196.

[Step S195] The correlation characteristic monitoring unit 132d classifies the selected points into the C'group. For example, the correlation characteristic monitoring unit 132d stores information indicating that the point belongs to the C'group in the memory 102 in association with the identification number of the selected point.

[Step S196] Whether or not the correlation characteristic monitoring unit 132d has not classified any of the points on the correlation diagram into any group and has selected all the unselected points in the C group classification process in step S191. To judge. If all the points have been selected, the correlation characteristic monitoring unit 132d ends the C group classification process. If there is an unselected point, the correlation characteristic monitoring unit 132d advances the process to step S191.

By the processing shown in FIGS. 19 to 21, all the points plotted in the correlation diagram are classified into one of the groups. The points plotted in the correlation diagram represent the rate of change of the measured values measured by each of the two sensors of the correlation pair at the corresponding time. In this way, the server 100 does not generate the correlation diagram with the measured values as they are, but generates the correlation diagram after calculating the rate of change. This makes it possible to accurately capture abnormal data when the plotted points are classified.

FIG. 22 is a diagram showing a difference in the correlation diagram depending on whether or not the rate of change is calculated. FIG. 22 shows an example of generating a correlation diagram when a pair of a sensor for measuring voltage and a sensor for measuring integrated current is selected as a correlation pair. In this case, a correlation diagram is generated based on the time series data 211 of the voltage and the time series data 212 of the integrated current amount. Spike noise appears in the voltage time series data 211.

When the correlation diagram 221 is generated without calculating the rate of change here, the distance from the position where the plotted points are concentrated to the point corresponding to the time when the spike noise occurs is short (distance is small). On the other hand, when the correlation diagram 222 is generated after the rate of change is calculated, the distance from the position where the plotted points are concentrated to the point corresponding to the time when the spike noise occurs is long (large distance). Therefore, it is difficult to detect an abnormality in the data in the correlation diagram 221 that has not undergone the calculation of the rate of change. On the other hand, in the correlation diagram 222 generated after calculating the rate of change, the points corresponding to the time when the spike noise occurred can be divided into a separate group from the other points, and can be detected as an abnormality of the data. It becomes.

When calculating the rate of change, the existence of abnormal data can be made clearer by appropriately setting the time interval ΔT when calculating the amount of change in the measured value.
FIG. 23 is a diagram showing an example of a method of setting the time interval ΔT in calculating the rate of change. In the example of FIG. 23, the difference in the rate of change is shown when the rate of change is calculated by changing the time interval ΔT in the calculation of the rate of change based on the time series data 231 of the shaft rotation speed. The sampling period of the time series data 231 is, for example, 1.5 seconds.

In the time series data 231, the shaft rotation speed fluctuates greatly at a specific time. When the rate of change is calculated with the time interval ΔT = 1 sec based on such time series data 231, the time series change of the calculated rate of change clearly shows the fluctuation status in the original time series data 231. It does not appear. When the rate of change is calculated with the time interval ΔT = 10 sec, the time-series change of the calculated rate of change shows the change in the original time-series data 231. When the rate of change is calculated with the time interval ΔT = 60 sec, the change in the original time-series data 231 appears more clearly in the time-series change of the calculated rate of change.

As described above, when the change of the waveform of the time series data 231 is instantaneous with respect to the sampling cycle, if the time interval ΔT is set to the same time as the sampling cycle, it is difficult for a large change of the measured value in the time series data 231 to appear. By setting the time interval ΔT to be several times or more the sampling cycle or more, a large change in the time-series data 231 of the measured value can be clearly shown in the time-series change of the change rate after the change rate is calculated.

In the server 100, for example, the time interval ΔT in calculating the rate of change is determined to be an integral multiple of the sampling cycle of the sensor having the longer sampling cycle among the sensors of the correlation pair. As a result, even when there are various sensors, the rate of change can be calculated by an appropriate time interval ΔT, and the characteristic change of the time series data 231 of the measured value can be accurately captured.

In addition, by calculating the rate of change of time-series data, it is possible to clarify the time when the measured value fluctuates significantly momentarily.
FIG. 24 is a diagram showing a comparison example of the time series data and the time change of the change rate after the change rate is calculated. In the example of FIG. 24, the time series data 121a represents the time change of the axial horsepower of the engine. The time series data 121b represents the time change of the speed (ship speed) of the ship 30. The time series data 121c represents the time change of the rotation speed of the engine shaft. The time series data 121d represents the time change of the CPP blade angle response value.

In FIG. 24, the time changes of the measured values of the time series data 121a to 121d and the time changes of the change rate for each time interval ΔT obtained by calculating the change rate are shown in graphs 241 to 244 with the time on the horizontal axis. Represents. In the graphs 241 to 244, the scale on the right side shows the value of the measured value before the calculation of the rate of change, and the scale on the left side shows the value of the rate of change after the calculation of the rate of change.

As can be seen from the graphs 241 to 244, before the rate of change is calculated, the magnitude and the fluctuation range of the measured values differ greatly depending on the measurement target. On the other hand, the rate of change after calculating the rate of change changes around 0, and becomes a large value at the time when the original measured value suddenly changes. Therefore, by generating a correlation diagram of a correlation pair based on the rate of change, a correlation diagram in which many points are plotted near the origin is generated.

FIG. 25 is a diagram showing an example of the generated correlation diagram. For example, from the time-series data 121a of the axial horsepower, the change rate information 251 indicating the time change of the change rate (kw / s) of the axial horsepower is generated. Further, from the time series data 121b of the ship speed, the change rate information 252 indicating the time change of the change rate (knots / s) of the ship speed is generated. From the time series data 121d of the CPP blade angle response value, the change rate information 253 indicating the time change of the change rate (deg / s) of the CPP blade angle response value is generated. From the time-series data 121c of the shaft rotation speed, the change rate information 254 indicating the time change of the change rate of the shaft rotation speed (shaft rotation speed / s) is generated.

When a correlation pair of the sensor for measuring the axial horsepower and the sensor for measuring the ship speed is selected, the correlation diagram 255 between the rate of change information 251 and the rate of change information 252 is generated. Correlation The horizontal axis of FIG. 255 is the rate of change in axial horsepower, and the vertical axis is the rate of change in ship speed. With reference to the correlation diagram 255, neither the axial horsepower nor the ship speed changes in many periods, and the points corresponding to the times within that period are plotted near the origin. In addition, during the period when either one of the axial horsepower and the ship speed changes, the other also changes at the same time, and some points corresponding to the time within the corresponding period are plotted in the first quadrant.

When a correlation pair of the sensor for measuring the axial horsepower and the sensor for measuring the CPP wing angle response value is selected, the correlation diagram 256 between the change rate information 251 and the change rate information 253 is generated. Correlation The horizontal axis of FIG. 256 is the rate of change in axial horsepower, and the vertical axis is the rate of change in the CPP blade angle response value. With reference to the correlation diagram 256, there is a period in which the CPP blade angle response value does not change even if the axial horsepower fluctuates, and the points corresponding to the times within the corresponding period are plotted on the horizontal axis. In addition, there is a period in which the CPP blade angle response value fluctuates according to the fluctuation in the axial horsepower, and the points corresponding to the times within the corresponding period are plotted in the first quadrant.

When a correlation pair of the sensor for measuring the shaft horsepower and the sensor for measuring the shaft rotation speed is selected, the correlation diagram 257 between the change rate information 251 and the change rate information 254 is generated. Correlation The horizontal axis of FIG. 257 is the rate of change in axial horsepower, and the vertical axis is the rate of change in axial speed. With reference to the correlation diagram 257, neither the axial horsepower nor the axial rotation speed changes in many periods, and the points corresponding to the times within that period are plotted near the origin. Further, in the period when either one of the shaft horsepower and the shaft rotation speed changes, the other also changes at the same time, and some points corresponding to the times within the corresponding period are plotted in the first quadrant.

In this way, the characteristics (correlation characteristics) of the correlation diagrams 255 to 257 are different for each correlation pair. If the correlation characteristic of the correlation pair collapses in a specific unit period, it is highly possible that the measured value contains abnormal data due to changes in the external environment (disturbance).

FIG. 26 is a diagram showing an example of a correlation diagram when the correlation characteristic is broken. In the example of FIG. 26, correlation diagrams 261 to 264 in unit periods P1 to P4 are shown for a particular correlation pair. During the unit periods P1 to P3, when the measured value of one sensor fluctuates, the measured value of the other data also fluctuates. When the rate of change of one increases, the rate of change of the other also increases, and when the rate of change of one decreases, the rate of change of the other also decreases, so that there is a positive correlation.

After such a positive correlation continues as a correlation characteristic over a plurality of unit periods P1 to P3, a state different from the positive correlation occurs in the unit period P4. For example, the points plotted in the portion surrounded by the dotted circle in the correlation diagram 264 deviate from the positive correlation. The occurrence of points that deviate from the positive correlation in this way can be detected by classifying each point by dividing it into regions as shown in FIG. That is, by grouping the plotted points, it is possible to detect the collapse of the correlation characteristic and correctly judge that the data is abnormal.

[Other embodiments]
The abnormality detection method shown in the first embodiment or the second embodiment can be applied to a system other than the system for monitoring the operation of the ship 30. For example, even in EV operation monitoring by a sensor attached to an electric vehicle (EV) or the like, the abnormality detection shown in the first embodiment or the second embodiment can be performed. In EV, measurement data such as position (latitude, longitude) and running motor power consumption can be obtained by a sensor. In addition, by applying the technology of the first or second embodiment to monitoring equipment having a large number of sensors such as construction vehicles, machine tools, and aircraft, erroneous judgment based on abnormal data is suppressed. Highly accurate abnormality detection is possible.

In the second embodiment, the average rate of change for each time interval ΔT is calculated as the rate of change of the measured value shown in the time series data acquired from the sensor, but the rate of change is differentiated for each predetermined time interval. A coefficient (instantaneous rate of change) can also be used. When a differential coefficient is used as the rate of change, the rate of change calculation unit 132c performs a differential operation of an equation indicating a change in the measured value shown in the time series data, and calculates the differential coefficient at a predetermined time interval.

Further, in the second embodiment, the variance value is used as an index of the degree of variation in the rate of change, but other indexes can also be used. For example, the standard deviation can be used as an index of the degree of variation instead of the variance value.

Although the embodiment has been illustrated above, the configuration of each part shown in the embodiment can be replaced with another having the same function. Further, any other components or processes may be added. Further, any two or more configurations (features) of the above-described embodiments may be combined.

1 1st sensor 2 2nd sensor 10 Information processing device 11 Storage unit 12 Processing unit

Claims (9)

Obtain the first plurality of values obtained based on the measurement of the first sensor and the second plurality of values obtained based on the measurement of the second sensor.
A plurality of sets in which each of the first plurality of values and each of the second plurality of values are combined are classified into a plurality of groups.
Anomalies are detected according to changes in the number of pairs included in each of the plurality of groups.
An anomaly detection program characterized by having a computer execute processing. In the process of acquiring the first plurality of values and the second plurality of values, the change rate of the measured values obtained by the measurement of the first sensor at a predetermined time interval is set to the first plurality of values. The process includes a process of acquiring as a value and acquiring the rate of change of the measured value obtained by the measurement of the second sensor as the second plurality of values at a predetermined time interval.
The abnormality detection program according to claim 1. The process of classifying the plurality of sets is performed on the plurality of groups on a coordinate system having a first coordinate axis corresponding to the first plurality of values and a second coordinate axis corresponding to the second plurality of values. A plurality of regions corresponding to each of the above are generated, and for each of the plurality of sets, the set is classified into a group corresponding to a region to which a point having two values included in the set as coordinate values belongs. ,
The abnormality detection program according to claim 1 or 2. The plurality of regions include a region including the origin of the coordinate system, a region including the first coordinate axis, a region including the second coordinate axis, and a region not including the first coordinate axis and the second coordinate axis. Including at least one of
3. The abnormality detection program according to claim 3. The process of acquiring the first plurality of values and the second plurality of values is performed for each unit period, based on the measurement of the first sensor within the unit period. Including a process of acquiring a plurality of values and the second plurality of values obtained based on the measurement of the second sensor within the unit period.
The process of classifying into the plurality of groups is executed for each unit period, and the process is executed.
In the process of detecting the abnormality, the plurality of groups are sorted by the number of pairs included in the group for each unit period, the order after sorting of the plurality of groups in the first unit period, and the second. Includes a process of determining that there is an abnormality when the sorted order of the plurality of groups in the unit period does not match.
The abnormality detection program according to any one of claims 1 to 4. A process of selecting two sensors from a plurality of sensors attached to a controlled object and generating a plurality of obtained sensor pairs is executed.
By using one of the sensor pairs as the first sensor and the other as the second sensor, a process of acquiring the first plurality of values and the second plurality of values for each of the sensor pairs. , The process of classifying into the plurality of groups, and the process of detecting the abnormality.
The abnormality detection program according to any one of claims 1 to 5, wherein the computer executes the process. Based on the ratio of the sensor pair in which the abnormality is detected among the plurality of generated sensor pairs, it is determined whether or not there is a measurement abnormality in at least a part of the plurality of sensors.
The abnormality detection program according to claim 6, wherein the computer executes the process. Obtain the first plurality of values obtained based on the measurement of the first sensor and the second plurality of values obtained based on the measurement of the second sensor.
A plurality of sets in which each of the first plurality of values and each of the second plurality of values are combined are classified into a plurality of groups.
Anomalies are detected according to changes in the number of pairs included in each of the plurality of groups.
An anomaly detection method characterized in that a computer executes processing. The first plurality of values obtained based on the measurement of the first sensor and the second plurality of values obtained based on the measurement of the second sensor are acquired, and the first plurality of values are obtained. A processing unit that classifies a plurality of sets that combine each of the second plurality of values into a plurality of groups and detects an abnormality according to a change in the number of pairs included in each of the plurality of groups.
An information processing device characterized by having.

Images