Monitoring device, method, and program

Abstract

To provide a monitoring device capable of suppressing erroneous detection in system abnormality detection, a method, and a program.SOLUTION: A monitoring device 100 comprises a model acquisition unit 103, a prediction unit 105, and an abnormality detection unit 106. When first data is entered which includes a measurement value of each sensor belonging to a first sensor assembly, the model acquisition unit 103 acquires a model for generating second data containing a prediction value of each sensor belonging to a second sensor assembly. When a control signal changes which causes a sudden change in a measurement value of a first sensor, the determination unit 104 generates a determination signal by determining the change of the control signal. The prediction unit 105 generates second data from the first data and the model. The abnormality detection unit 106 detects an abnormality of a sensor belonging to the system or the second sensor assembly on the basis of the measurement value, the prediction value, and the determination signal. When the determination signal represents the change of the control signal, the abnormality detection unit 106 makes it difficult to detect the abnormality.SELECTED DRAWING: Figure 1

Description

Technical Field

[0001] Embodiments of the present invention relate to a monitoring device, method, and program.

Background Art

[0002] Infrastructure systems used in power generation plants, water treatment plants, etc., and systems used in manufacturing equipment are composed of a plurality of devices. When the infrastructure system stops, it may have an adverse impact on social functions. Also, when the system of manufacturing equipment stops, it may cause economic losses. Therefore, it is important to keep these systems in a sound state.

[0003] Such a system is generally controlled using feedback control, feedforward control, or a derivative method of these control methods so that the measured value of the system output approaches the target value. Also, when the scale of the system is large, it is composed of a plurality of subsystems, and each subsystem is controlled by feedback control or feedforward control. Therefore, control becomes complicated in a large-scale system.

[0004] In order to prevent system failures or to restore the system as soon as possible after a failure, it is necessary to monitor the system. For system monitoring, a plurality of sensors are installed at various locations within the system. And by monitoring the values of the plurality of sensors installed in the system, the state of the system can be monitored. Also, when the scale of the system is large or the system is complicated, the number of sensors required for system monitoring increases. In this case, it is difficult to monitor all the sensors simultaneously with a limited number of people.

[0005] On the other hand, a monitoring device that assists or automates system monitoring using sensor measurement values is known. In such a monitoring device, it is required to suppress false detection in which the system is erroneously detected as abnormal even though it is normal.

Prior Art Documents

Patent Documents

[0006]

Patent Document 1

Patent Document 2

Summary of the Invention

Problems to be Solved by the Invention

[0007] The problem to be solved by the present invention is to provide a monitoring device, method, and program capable of suppressing false detection in the detection of system abnormalities.

Means for Solving the Problems

[0008] To solve such problems, the monitoring device according to the embodiment includes a measurement value acquisition unit, a control signal acquisition unit, a model acquisition unit, a prediction unit, and an abnormality detection unit. The measurement value acquisition unit acquires measurement values of a plurality of sensors installed in the system. The control signal acquisition unit acquires a control signal from the system. The model acquisition unit inputs first data including measurement values of each sensor belonging to a first sensor set including a first sensor in which a sudden change occurs in the measurement value when the control signal changes during a predetermined operation mode, and acquires a model that generates second data including predicted values of each sensor belonging to a second sensor set including a second sensor correlated with the first sensor. The determination unit generates a determination signal by determining a change in the control signal. The prediction unit generates the second data including predicted values of each sensor belonging to the second sensor set from the first data included in the measurement value and the model. The abnormality detection unit detects an abnormality of the system or an abnormality of at least one sensor belonging to the second sensor set based on the measurement values of each sensor belonging to the second sensor set, the second data, the determination signal, and a threshold value. Further, when the determination signal indicates that there is a change in the control signal, the abnormality detection unit makes it difficult to detect the abnormality.

Brief Description of the Drawings

[0009]

Figure 1

Figure 2

Figure 3

Figure 4

Figure 5

Figure 6

Figure 7

Figure 8

Figure 9

Figure 10

Figure 11

Figure 12

Figure 13

Figure 14

Figure 15

Figure 16

[0010] Hereinafter, embodiments of a monitoring device, a method, and a program will be described in detail with reference to the drawings. In the following description, components having substantially the same function and configuration are denoted by the same reference numerals, and duplicate description will be made only when necessary.

[0011] (First Embodiment) FIG. 1 is a diagram showing the configuration of a monitoring device 100 according to the first embodiment. The monitoring device 100 is connected to a system to be monitored via a network or the like. The monitoring device 100 acquires measurement values that are constantly measured at a plurality of sensors installed in the system to be monitored, and detects an abnormality in the system. Then, a signal indicating the abnormality detection result is output to the outside such as a display.

[0012] The network is, for example, a LAN (Local Area Network). Note that the connection to the network may be a wired connection or a wireless connection. Further, the network is not limited to a LAN, and may be the Internet, a public communication line, or the like.

[0013] The monitoring device 100 includes a processing circuit that controls the entire monitoring device 100 and a storage medium (memory). The processing circuit is a processor that executes the functions of the measurement value acquisition unit 101, the control signal acquisition unit 102, the model acquisition unit 103, the determination unit 104, the prediction unit 105, and the abnormality detection unit 106 by calling and executing a program in the storage medium. The processing circuit is formed from an integrated circuit including a CPU (Central Processing Unit), an ASIC (Application Specific Integrated Circuit), or an FPGA (Field Programmable Gate Array), etc. The processor may be formed from one integrated circuit or may be formed from a plurality of integrated circuits.

[0014] The storage medium stores a processing program used by the processor, and parameters, tables, etc. used in the operations of the processor. The storage medium is a storage device such as an HDD (Hard Disk Drive), an SSD (Solid State Drive), or an integrated circuit that stores various information. Also, the storage device may be a portable storage medium such as a CD (Compact Disc), a DVD (Digital Versatile Disc), or a flash memory in addition to an HDD or an SSD, etc., or may be a drive device that reads and writes various information between a semiconductor memory element such as a flash memory or a RAM (Random Access Memory).

[0015] Note that each function of the measurement value acquisition unit 101, the control signal acquisition unit 102, the model acquisition unit 103, the determination unit 104, the prediction unit 105, and the abnormality detection unit 106 may be realized by a single processing circuit, or a processing circuit may be configured by combining a plurality of independent processors, and each function may be realized by each processor executing a program. Also, each function of the measurement value acquisition unit 101, the control signal acquisition unit 102, the model acquisition unit 103, the determination unit 104, the prediction unit 105, and the abnormality detection unit 106 may be implemented as an individual hardware circuit.

[0016] The measurement value acquisition unit 101 acquires information (hereinafter referred to as measurement value data) including the measurement values of each sensor installed in the system to be monitored. The measurement value data is time-series data measured moment by moment by a plurality of sensors installed in the system to be monitored. The acquisition interval of the measurement value data is preferably determined according to the sampling interval of the sensor and the processing speed of the monitoring device 100. The acquisition interval of the measurement value data is, for example, 1 minute, 10 minutes, or the like.

[0017] The control signal acquisition unit 102 acquires the control signal of the system to be monitored in a predetermined operation mode. In this embodiment, the case of acquiring one specific control signal that affects the sensor value will be described, but a plurality of control signals may be acquired.

[0018] The operation mode is, for example, a mode for starting the system, a mode for stopping the system, a standby mode, a mode for performing a test run, a mode for operating at a predetermined load, a mode for performing work on the system, and the like.

[0019] As the control signal, for example, a feedback control signal, a feedforward control signal, a control signal whose correlation with the change in the operation mode is weak and whose change is irregular, etc. are used. Also, as the control signal, a signal that rarely changes during a predetermined operation mode may be used.

[0020] Specifically, as the control signal, for example, a signal of BIR (boiler input regulator), a signal of a soot blower, a signal related to the start-up and stop of a plurality of fuel devices provided in the system, a signal indicating the number of start-ups of a plurality of fuel devices provided in the system, a signal indicating the switching of the fuel type, etc. are used.

[0021] The signal of BIR (boiler input regulator) is an example of a feedforward control signal for compensating the response delay of the boiler included in the system.

[0022] The soot blower removes soot and dust from the equipment in the system. The soot blower is used, for example, in a boiler. The soot blower often operates regardless of the timing of the change in the operation mode. The signal of the soot blower is an example of a control signal with a weak correlation with the change in the operation mode and irregular changes.

[0023] The system may include a fuel device for operating the system. Also, the number of fuel devices is not necessarily one, and there may be a plurality of them. When the system is operating at a predetermined load, the number of fuel devices started may change. Also, even if the number of fuel devices started is the same, the fuel devices started may be different. The signal indicating the start and stop of a plurality of fuel devices is an example of a control signal with a weak correlation with the change in the operation mode and irregular changes. Also, the signal indicating the number of fuel devices started among a plurality of fuel devices is an example of a control signal with a weak correlation with the change in the operation mode and irregular changes.

[0024] The fuel for operating the system is not necessarily one. When there are a plurality of fuels, even if the load of the system is constant, the fuel may be switched. The signal for switching the type of fuel is an example of a control signal with a weak correlation with the change in the operation mode and irregular changes.

[0025] The model acquisition unit 103 acquires a model. The model receives the input of the measurement values of one or more sensors provided in the system, and outputs a predicted value (hereinafter referred to as a predicted value) of the sensor value normally obtained from at least one of the one or more sensors having a correlation with at least one of the input measurement values. Here, the data input to the model is called first data, and the set of sensors corresponding to the first data is called the first sensor set. The first data includes the measurement values of the first sensor set. Also, the data output from the model is called second data, and the set of sensors corresponding to the second data is called the second sensor set. The second data includes the predicted values of the second sensor set. That is, the model receives the input of the first data including the measurement values of each sensor belonging to the first sensor set, and generates and outputs the second data including the predicted values of each sensor belonging to the second sensor set.

[0026] The number of sensors belonging to the second sensor set may be the same as or different from the number of sensors belonging to the first sensor set. Also, the sensors belonging to the first sensor set and the sensors belonging to the second sensor set may or may not overlap.

[0027] In the system, after a certain control signal changes, the measured value of a specific sensor may change. For example, when a change (hereinafter referred to as a mutation) occurs in which the signal value of a certain control signal increases only for an instant and then returns to near the original value, or the signal value decreases and then returns to near the original value, a mutation may also occur in the measured value of a specific sensor. Among the sensors provided in the system, there are sensors (hereinafter referred to as first sensors) in which a mutation is likely to occur in the measured value due to a change in a specific control signal, and sensors in which a mutation is unlikely to occur in the measured value due to a change in a specific control signal. In the above description of the mutation, it is described as "instantaneously", but the time that can be regarded as "instantaneously" varies depending on the control method and response time of the system. The time that can be regarded as "instantaneously" may be, for example, a few seconds, a few minutes, or about 10 hours.

[0028] The sensors included in the first sensor set and the sensors included in the second sensor set are determined in advance according to the type of system to be monitored, the type of control signals that can be obtained, and the type of model to be used. Typically, the first sensor set includes sensors corresponding to the first sensors. In the present embodiment, the first sensor set is described as including the first sensors.

[0029] Among the sensors included in the second sensor set, when predicting sensor values using a model, there may be a sensor (hereinafter referred to as the second sensor) whose predicted value is likely to change in conjunction with a change in the measured value of the first sensor included in the first sensor set. The second sensor is a sensor having a correlation with the first sensor. Also, the second sensor is a sensor whose measured value is less likely to be affected by a change in the control signal compared to the first sensor. Typically, the second sensor set includes a sensor corresponding to the second sensor. In the present embodiment, the second sensor set will be described as including the second sensor.

[0030] Note that the first sensor set may include a sensor corresponding to the second sensor. Also, the second sensor set may include a sensor corresponding to the first sensor. Also, a specific sensor belonging to both the first sensor set and the second sensor set may correspond to both the first sensor and the second sensor. Also, a specific sensor corresponding to the first sensor or the second sensor may be included in both the first sensor set and the second sensor set.

[0031] The model is a machine learning model that learns the correlation between sensors using the measurement history of each sensor belonging to a group of sensors of the same type having a correlation among the sensors provided in the system. The model is generated by machine learning using the historical data of the measured values of each sensor belonging to the union of the first sensor set and the second sensor set.

[0032] That is, when the model inputs first data including the measured values of each sensor belonging to the first sensor set including at least the first sensor as an element, it generates second data including the predicted values of each sensor belonging to the second sensor set consisting of one or more of the plurality of sensors installed in the system.

[0033] The model is a function generated by machine learning. Let that function be represented by f. The function f is generated by executing Equation (1) by a machine learning method. E(g) in Equation (1) is represented by Equation (2).

[0034]

Equation

[0035]

Number

[0036] In formulas (1) and (2), T represents the set of times of the data used for training the model among the historical data. t represents a time belonging to T. x t represents the measured values of each sensor belonging to the first sensor set at time t, or a vector obtained by arranging the measured values after normalization or standardization. Since normalization and standardization are well-known methods, the description thereof is omitted. This embodiment is effective even when normalization or standardization is performed on the measured values. That is, x t corresponds to the first data. y t represents a vector obtained by arranging the measured values of each second sensor at time t. g is a function that, when input with a vector of the same dimension as x t , returns a vector of the same dimension as y t . ||·|| 2 represents the norm of a vector. x t and y t The set {x t, y t} t∈T is called training data.

[0037] As the machine learning method, any method may be adopted. The model is, for example, a neural network model including a deep neural network or an autoencoder. Alternatively, the model is a model such as linear regression, ridge regression, lasso regression, kernel regression, kernel ridge regression, support vector regression, decision tree regression, random forest regression, etc.

[0038] The determination unit 104 generates a determination signal based on a control signal. The determination unit 104 sends the generated determination signal to the abnormality detection unit 106. The determination signal is a signal indicating a determination result as to whether there has been a change in the control signal during a period (hereinafter referred to as the target period) from a time (hereinafter referred to as the start time) that is a predetermined time M before the time serving as the determination criterion (hereinafter referred to as the target time) to the target time. The predetermined time M corresponds to the length of the target period. The predetermined time M is, for example, stored in advance in a storage medium. The predetermined time M is set, for example, according to the time from when a change in the control signal occurs until a sudden change occurs in the measurement value of the first sensor. Specifically, the determination unit 104 determines, for example, whether the control signal has changed by more than a predetermined threshold value during the target period from the start time to the target time, and generates a determination signal based on the determination result.

[0039] The amount of change in the control signal is, for example, the absolute value of the difference in the values of the control signal between the target time and the start time. Alternatively, as the amount of change in the control signal, the cumulative value of the differential values of the control signal from the start time to the target time may be used. Alternatively, as the amount of change in the control signal, the cumulative value of the absolute values of the differential values of the control signal from the start time to the target time may be used. When a plurality of signals are used as the control signal, a determination signal may be generated for each of the plurality of signals included in the control signal.

[0040] The prediction unit 105 causes the model to output second data by inputting first data into the model. The second data includes predicted values of each sensor belonging to the second sensor set.

[0041] The second data is represented as in the following formula (3). In formula (3), the ^ attached to y t is a hat symbol. Hereinafter, a symbol X with a hat symbol will be represented as " ^X ". In formula (3), ^y t indicates the second data at time t. x t represents the first data at time t. f represents the function of the model.

[0042]

Equation

[0043] The abnormality detection unit 106 detects an abnormality of a sensor belonging to the second sensor set based on the measurement value, the determination signal, and the second data included in the measurement value data of each sensor belonging to the second sensor set. Then, the abnormality detection unit 106 outputs an abnormality detection signal indicating the detection result of the abnormality to the outside of the monitoring device 100.

[0044] Hereinafter, the abnormality detection unit 106 will be described in detail.

[0045] The abnormality detection unit 106 includes a difference calculation unit 113. The difference calculation unit 113 calculates the difference between the measurement value and the predicted value for at least one sensor belonging to the second sensor set.

[0046] The abnormality detection unit 106 determines whether there is an abnormality in the sensor by performing a threshold determination on the difference between the measurement value and the predicted value, or the absolute value thereof, for the sensors belonging to the second sensor set. When performing the threshold determination on the difference between the measurement value and the predicted value, the abnormality detection unit 106 prepares an upper threshold and a lower threshold.

[0047] Also, when performing the threshold determination, the abnormality detection unit 106 obtains the determination signal to obtain whether there is a change in the control signal during the period of interest. When there is a change in the control signal during the period of interest, the abnormality detection unit 106 controls so that it is difficult to detect an abnormality during the period of interest. For example, when there is a change in the control signal during the period of interest, the abnormality detection unit 106 determines that there is no abnormality during the period of interest and sets a signal indicating that there is no abnormality as the abnormality detection signal. Alternatively, the abnormality detection unit 106 controls so as not to detect an abnormality during the period of interest when there is a change in the control signal during the period of interest. Alternatively, the abnormality detection unit 106 sets the threshold value for the threshold determination during the period of interest to infinity only when the determination signal indicates that there has been a change in the control signal. Alternatively, the abnormality detection unit 106 makes the threshold value for the threshold determination during the period of interest larger than at other times only when the determination signal indicates that there has been a change in the control signal.

[0048] Next, the operations of the processes executed by the monitoring device 100 will be described. FIG. 2 is a flowchart showing an example of the procedure of the monitoring process. The monitoring process is a process of monitoring a system by detecting an abnormality of the system based on sensor values obtained from sensors installed in the system to be monitored. Note that the procedure in each process described below is merely an example, and each process can be changed as appropriate as much as possible. Also, regarding the procedure described below, steps can be omitted, replaced, and added as appropriate according to the embodiment.

[0049] (Monitoring Process) (Step S201) The measurement value acquisition unit 101 acquires measurement value data of sensors from each of a plurality of sensors installed in the system to be monitored. The measurement value data includes measurement values of each sensor belonging to the union of the first sensor set and the second sensor set. The measurement value acquisition unit 101 sends the acquired measurement value data to the prediction unit 105 and the abnormality detection unit 106.

[0050] (Step S202) The control signal acquisition unit 102 acquires the control signal of the system to be monitored. The control signal acquisition unit 102 sends the acquired control signal to the determination unit 104.

[0051] (Step S203) The model acquisition unit 103 acquires a model from a storage medium provided inside the monitoring device 100. The model acquisition unit 103 sends the acquired model to the prediction unit 105.

[0052] (Step S204) The determination unit 104 acquires the control signal and information regarding the target period. Next, the determination unit 104 calculates the amount of change in the control signal during the target period and determines whether the control signal has changed during the target period. If the amount of change in the control signal during the target period is greater than a predetermined value, the determination unit 104 sets the value of the determination signal to "1". On the other hand, if the amount of change is less than or equal to the predetermined value, the determination unit 104 sets the value of the determination signal to "0". The determination unit 104 sends the generated determination signal to the abnormality detection unit 106.

[0053] (Step S205) The prediction unit 105 acquires measurement value data and a model. The prediction unit 105 extracts the measurement values of the sensors belonging to the first sensor set from the measurement value data as first data, and inputs the extracted first data into the model. The model receives the input of the first data and outputs second data including the predicted values of the second sensors. The prediction unit 105 sends the second data output from the model to the abnormality detection unit 106.

[0054] (Step S206) The difference calculation unit 113 of the abnormality detection unit 106 acquires measurement value data and second data. The difference calculation unit 113 acquires the measurement values of the sensors belonging to the second sensor set from the measurement value data. Further, the difference calculation unit 113 extracts the predicted values of the sensors belonging to the second sensor set from the second data. Then, the difference calculation unit 113 calculates the difference between the measurement value and the predicted value for each sensor belonging to the second sensor set. The difference between the measurement value and the predicted value is calculated using, for example, the following formula (4) or formula (5).

[0055]

Equation

[0056]

Equation

[0057] In formula (4) and formula (5), y t,s represents the measurement value of the sensor. The subscript s represents the ID of the sensor. ^y t,s represents the predicted value of the sensor. d t,s represents the difference between the measurement value and the predicted value.

[0058] (Step S207) The abnormality detection unit 106 acquires the difference between the measured value and the predicted value and the determination signal. The abnormality detection unit 106 executes an abnormality detection process using the difference between the measured value and the predicted value. The abnormality detection process is a process for detecting the presence or absence of an abnormality for each sensor belonging to the second sensor set. The abnormality detection unit 106 outputs an abnormality detection signal indicating the detection result of the abnormality to a management system that manages the system or an external display.

[0059] Here, the abnormality detection process executed in step S207 will be described. FIG. 3 is a flowchart showing an example of the procedure of the abnormality detection process. Note that the procedure in each process described below is merely an example, and each process can be changed as appropriate as much as possible. Also, regarding the procedure described below, steps can be omitted, replaced, and added as appropriate according to the embodiment.

[0060] (Abnormality Detection Process) (Step S301) Based on the determination signal, the abnormality detection unit 106 determines whether there is a change in the control signal during the period of interest. For example, when "0" is acquired as the determination signal (step S301 - No), the abnormality detection unit 106 determines that the control signal has not changed during the period of interest. On the other hand, when "1" is acquired as the determination signal (step S301 - Yes), the abnormality detection unit 106 determines that the control signal has changed during the period of interest.

[0061] (Step S302) When the control signal changes at the time of interest (step S301 - Yes), the abnormality detection unit 106 changes the threshold value during the period of interest. For example, the abnormality detection unit 106 sets the threshold value during the period of interest to a value larger than the threshold value at other times. Thereby, when the control signal changes during the period of interest, it becomes difficult to detect an abnormality.

[0062] (Step S303) The abnormality detection unit 106 determines whether the absolute value of the difference between the measured value and the predicted value is greater than the threshold value for each sensor belonging to the second sensor set. For example, the absolute value of the difference between the measured value and the predicted value (|dt,s If |)| is greater than the threshold value, the abnormality detection unit 106 determines that there is an abnormality in the sensor. Also, the absolute value of the difference between the measured value and the predicted value (|d t,s |) is less than or equal to the threshold value, the abnormality detection unit 106 determines that there is no abnormality in the sensor.

[0063] The abnormality detection unit 106 outputs an abnormality detection signal indicating the detection result of the abnormality to the outside. After that, the monitoring device 100 ends the abnormality detection process, and the process proceeds to step S208.

[0064] (Step S208) The monitoring device 100 determines whether an instruction to stop monitoring the system has been input from an external system. If an instruction to stop monitoring the system has been input (Step S208 - Yes), the monitoring device 100 ends the monitoring process.

[0065] The monitoring device 100 repeatedly performs the processes of steps S201 to S207 until an instruction to stop monitoring the system is input, thereby acquiring the measured value data measured every moment in a plurality of sensors installed in the system to be monitored, and outputs an abnormality detection signal indicating the abnormality detection result to the outside.

[0066] (Effect of the First Embodiment) Hereinafter, the effects of the monitoring device 100 according to the present embodiment will be described.

[0067] When utilizing a model of a machine learning method that uses the correlation relationship between sensors belonging to the same system sensor group, false detections may occur where the predicted value of the second sensor follows the measured value of the first sensor in conjunction with a change in the control signal, or false detections may occur due to a relative shortage in the training data of a scene where a mutation occurs in the measured value of the first sensor in conjunction with a change in the control signal. Hereinafter, after explaining these two typical examples where false detections occur, the effects of the present embodiment will be described.

[0068] (False Detection Caused by the Predicted Value of the Second Sensor Following the Measured Value of the First Sensor) First, an explanation will be given of false detections that occur when the predicted values of the second sensors included in the second sensor set are linked to the measured values of the first sensors included in the first sensor set. Here, for simplicity of explanation, the first sensor shall refer to the first sensors included in the first sensor set, and the second sensor shall refer to the second sensors included in the second sensor set.

[0069] In the system, it is normal behavior for the control signal to change during a given operation mode. For example, during a given operation mode, a mutation may occur in which the control signal changes significantly in a short period and then returns to its original state.

[0070] FIG. 4 is a diagram for explaining how false detections occur when the predicted values of the second sensors are linked to the measured values of the first sensors. FIG. 4(a) is a diagram showing an example of a mutation occurring in the control signal. The horizontal axis in FIG. 4(a) indicates time. The vertical axis in FIG. 4(a) indicates the value of the control signal. In FIG. 4(a), a mutation occurs in the control signal near time t1. The control signal increases near time t1 and then returns to its value before the change near time t2.

[0071] Also, in the system, it is normal behavior for the measured value of a sensor to change after the control signal changes. Although control is often performed to suppress sudden changes in the measured value of the sensor of the device to be controlled as much as possible by feedback control, feedforward control, etc., it cannot always be completely suppressed. Also, when the system is constructed of multiple subsystems, even if a sudden change in the measured value of a sensor within a certain subsystem is suppressed to some extent, a sudden change in the measured value of a sensor not included in that subsystem is not necessarily suppressed. For example, when the control signal of an upstream subsystem changes, even if a sudden change in the measured value of a sensor within that subsystem is suppressed to some extent, a sudden change may occur in the measured value of a sensor within a subsystem downstream of that subsystem. Therefore, when the system is large-scale and complex, there are often many sensors whose measured values suddenly change due to a change in the control signal. The first sensor is a sensor whose measured value suddenly changes due to a change in the control signal. Therefore, the measured value of the first sensor changes in conjunction with the change in the control signal.

[0072] FIG. 4(b) is a diagram showing an example of the measured value of the first sensor when a sudden change occurs in the control signal as shown in FIG. 4(a). The horizontal axis in FIG. 4(b) indicates the time synchronized with the horizontal axis in FIG. 4(a). The vertical axis in FIG. 4(b) indicates the value of the first sensor. In FIG. 4(b), the measured value of the first sensor increases near time t2 after time t1 and then returns to the value before the change. That is, in conjunction with the change in the control signal, a sudden change also occurs in the measured value of the first sensor.

[0073] The predicted value of the second sensor included in the second sensor set is generated with reference to the measured value of the first sensor included in the first sensor set. As described above, the second data including the predicted value of the second sensor included in the second sensor set is calculated, for example, using Equation (3). As can be seen from Equation (3), the measured value of the first sensor is included in the variables of the equation for calculating the predicted value of the second sensor. Therefore, when using the model utilizing Equation (3), the predicted value of the second sensor included in the second sensor set changes in conjunction with the change in the measured value of the first sensor included in the first sensor set. On the other hand, the second sensor is a sensor that is less affected by the change in the control signal compared to the first sensor. For this reason, the measured value of the second sensor does not change significantly in conjunction with the change in the control signal compared to the measured value of the first sensor.

[0074] FIG. 4(c) is a diagram showing an example of the predicted value and the measured value of the second sensor included in the second sensor set when the measured value of the first sensor changes as shown in FIG. 4(b). The dashed line in FIG. 4(c) indicates the predicted value of the second sensor. The solid line in FIG. 4(c) indicates the measured value of the second sensor. The horizontal axis in FIG. 4(c) indicates the time synchronized with the horizontal axes in FIGS. 4(a) and 4(b). The vertical axis in FIG. 4(c) indicates the value of the second sensor. In FIG. 4(c), the amount of change in the measured value of the second sensor near time t2 is smaller than the amount of change in the measured value of the first sensor at time t2. That is, the measured value of the second sensor does not change significantly even when a mutation occurs in the control signal. On the other hand, the predicted value of the second sensor increases near time t2 after time t1 and then returns to the value before the change. Therefore, a mutation also occurs in the predicted value of the second sensor in conjunction with the change in the measured value of the first sensor. That is, a mutation also occurs in the predicted value of the second sensor in conjunction with the change in the control signal.

[0075] FIG. 4(d) is a diagram showing an example of the difference between the predicted value and the measured value of the second sensor when the predicted value and the measured value of the second sensor change as shown in FIG. 4(c). The horizontal axis of FIG. 4(d) indicates the time synchronized with the horizontal axes of FIGS. 4(a), 4(b), and 4(c). The vertical axis of FIG. 4(d) indicates the value of the difference between the predicted value and the measured value of the second sensor. This difference is an example of the difference represented by Equation (4). In FIG. 4(d), after the difference between the predicted value and the measured value of the second sensor increases near time t2, it returns to the value before the change. Therefore, in conjunction with the change in the control signal, a mutation also occurs in the difference between the predicted value and the measured value of the second sensor.

[0076] As described above, due to the occurrence of a mutation in the control signal, a mutation occurs in the measured value of the first sensor, and in conjunction with the change in the measured value of the first sensor, a mutation occurs in the predicted value of the second sensor. Therefore, a change also occurs in the predicted value of the second sensor due to the change in the control signal. On the other hand, the measured value of the second sensor does not change significantly due to the change in the control signal and the change in the measured value of the first sensor. For this reason, a mutation also occurs in the difference between the measured value and the predicted value of the second sensor due to the change in the control signal.

[0077] When the absolute value of the difference between the predicted value and the measured value of the second sensor exceeds the threshold value, in the abnormality detection process, it is determined that there is an abnormality in the second sensor. Therefore, when a mutation occurs in the control signal, even if the behavior of the measured value of the second sensor is normal, there is a high possibility of false detection in which the second sensor is detected as abnormal. Thus, when performing abnormality detection by utilizing a model that utilizes the correlation relationship between sensors, due to the linkage between the measured value of the first sensor and the predicted value of the second sensor, when the control signal changes, a mutation occurs in the predicted value of the second sensor, resulting in a case where false detection occurs in which the second sensor or its sensing target is determined to be abnormal even though it is normal.

[0078] (False detection due to insufficient training data during model training) Next, false detection due to insufficient training data during model training will be described. When the first sensor is included in the second sensor set, false detection may occur due to insufficient training data during model training.

[0079] In a system where stability is required, scenes where the control signal changes or the measured value of the first sensor has a mutation do not occur relatively frequently. Therefore, in the training data of the model, data at the time when a mutation occurs in the measured value of the first sensor due to a change in the control signal is relatively scarce. Also, in the machine learning of the model, in order to minimize the aforementioned E(g), data with a low occurrence frequency is relatively neglected. For this reason, the model tends to be trained to generate predicted values similar to data with a high occurrence frequency.

[0080] Therefore, when calculating the predicted value of each sensor belonging to the second sensor set including the first sensor using the model, the predicted value of the first sensor often does not change in conjunction with the change in the control signal compared to the measured value. For this reason, the change width of the predicted value of the first sensor at the time when the control signal changes is often smaller than the change width of the measured value of the first sensor.

[0081] FIG. 5 is a diagram for explaining how false detection occurs due to insufficient training data during model training. FIG. 5(a) is a diagram showing an example of a situation where a mutation occurs in the control signal. The horizontal axis in FIG. 5(a) indicates time. The vertical axis in FIG. 5(a) indicates the value of the control signal. In FIG. 5(a), a mutation occurs in the control signal near time t1. The control signal increases near time t1 and then returns to the value before the change near time t2.

[0082] FIG. 5(b) is a diagram showing an example of the measured value and predicted value of the first sensor when a mutation occurs in the control signal as shown in FIG. 5(a). The dashed line in FIG. 5(b) indicates the predicted value of the first sensor. The solid line in FIG. 5(b) indicates the measured value of the first sensor. The horizontal axis in FIG. 5(b) indicates the time synchronized with the horizontal axis in FIG. 5(a). The vertical axis in FIG. 5(b) indicates the value of the first sensor. In FIG. 5(b), a mutation also occurs in the measured value of the first sensor in conjunction with the change in the control signal. On the other hand, the predicted value of the first sensor does not change significantly near time t2 compared to the measured value.

[0083] Thus, a mutation occurs in the measured value of the first sensor due to a mutation in the control signal. On the other hand, the predicted value of the first sensor does not change significantly even when a change occurs in the control signal. Therefore, regardless of whether there is an abnormality in the first sensor, the absolute value of the difference between the measured value and the predicted value of the first sensor near the time when the control signal changes becomes larger than at other times. When the difference between the predicted value and the measured value of the first sensor exceeds a threshold, in the abnormality detection process, it is determined that there is an abnormality in the first sensor. For this reason, when a mutation occurs in the control signal, even if the behavior of the measured value of the first sensor is normal, there is a high possibility of a false detection in which the first sensor is detected as abnormal. As described above, when performing abnormality detection by utilizing a model that utilizes the correlation relationship between sensors, due to a relatively insufficient amount of training data, the absolute value of the difference between the predicted value and the measured value of the first sensor becomes large when the control signal changes, resulting in a false detection in which the first sensor or its sensing target is determined to be abnormal even though it is normal.

[0084] As described above, false detections may occur due to the interlocking of the predicted value of the second sensor with respect to the measured value of the first sensor and the relative lack of training data. Against such false detections, for example, in a method of setting a threshold for each operation mode determined according to the load, it is not possible to suppress false detections caused by changes in the control signal occurring during the operation mode. Also, in a method of controlling the threshold according to the number of false detections and non-detections within a certain period, it is not possible to suppress the aforementioned false detections.

[0085] On the one hand, the monitoring device 100 according to this embodiment acquires measurement value data from a plurality of sensors including a first sensor and a second sensor, acquires a control signal from the system that causes a mutation in the measurement value of the first sensor when it changes in a predetermined operation mode, and inputs first data including the measurement values of each sensor belonging to a first sensor set including the first sensor, and acquires a model that generates second data including the predicted values of each sensor belonging to a second sensor set including the second sensor as an element, generates a determination signal by determining the change in the control signal during a period of interest, generates the second data from the first data included in the measurement value data and the model, and can detect an abnormality in at least one sensor belonging to the second sensor set based on the measurement values of each sensor belonging to the second sensor set, the second data, the determination signal, and a predetermined threshold value. Further, when the determination signal indicates that there is a change in the control signal, the monitoring device 100 can make it less likely to detect an abnormality than when the determination signal indicates no change.

[0086] The control signal includes, for example, any one of a feedback control signal, a feedforward control signal, and a control signal whose correlation with the change in the operation mode is weak and whose change is irregular. Further, the control signal includes, for example, any one of a signal of a boiler input regulator, a signal of a soot blower, a signal related to the start or stop of a fuel device, a signal indicating the number of starts of a fuel device, and a signal indicating the switching of the type of fuel.

[0087] When the monitoring device 100 determines that there is no abnormality, for example, when the determination signal indicates a change in the control signal, it becomes difficult to detect an abnormality. Alternatively, the monitoring device 100 changes the threshold value according to the value of the determination signal, making it difficult to detect an abnormality. In this case, for example, by increasing the threshold value during the period when the control signal changes, it becomes difficult to detect an abnormality. The control for making it difficult to detect an abnormality is preferably applied to a predetermined sensor. The predetermined sensor may be all the sensors included in the second sensor set, or may be limited to the second sensor among the sensors included in the second sensor set. The predetermined sensor may be limited to, for example, the first sensor included in the second sensor set. The predetermined sensor may be limited to, for example, the first sensor and the second sensor included in the second sensor set.

[0088] FIG. 6 is a diagram for explaining a method of suppressing false detection caused by the interlocking between the measured value of the second sensor and the predicted value of the first sensor by making it difficult to detect an abnormality during the period when the control signal changes. FIG. 6(a) is a diagram showing an example of a state where a mutation has occurred in the control signal. The horizontal axis in FIG. 6(a) indicates time. The vertical axis in FIG. 6(a) indicates the value of the control signal. Here, the attention time is represented by t C and the start time, which is a predetermined time M back in the past from the attention time t C , is represented by t C-M . In FIG. 6(a), the control signal is changing during the attention period between the attention time t C and the start time t C-M .

[0089] FIG. 6(b) is a diagram showing an example of the determination signal when a mutation occurs in the control signal as shown in FIG. 6(a). The horizontal axis in FIG. 6(b) indicates the time synchronized with the horizontal axis in FIG. 6(a). The vertical axis in FIG. 6(b) indicates the value of the determination signal. In FIG. 6(b), the determination signal is "0" at times other than the attention period. Also, during the attention period, it is determined that the control signal has changed, and the determination signal is "1".

[0090] FIG. 6(c) is a diagram showing an example of the measured value and predicted value of the second sensor when a mutation occurs in the control signal as shown in FIG. 6(a). The horizontal axis of FIG. 6(c) indicates the time synchronized with the horizontal axes of FIGS. 6(a) and 6(b). The vertical axis of FIG. 6(c) indicates the value of the second sensor. The dashed line in FIG. 6(c) indicates the predicted value of the second sensor. The solid line in FIG. 6(c) indicates the measured value of the second sensor. Further, FIG. 6(d) is a diagram showing an example of the difference between the measured value and predicted value of the second sensor when a mutation occurs in the control signal as shown in FIG. 6(a). The difference indicated by the dashed line in FIG. 6(d) is an example of the difference represented by Equation (4). In FIG. 6(c), during the period of interest, a mutation occurs in the predicted value of the second sensor in conjunction with the change in the control signal. Therefore, in FIG. 6(d), a mutation also occurs in the difference between the measured value and predicted value of the second sensor during the period of interest.

[0091] Also, the dashed-dotted line in FIG. 6(d) indicates the upper threshold and lower threshold for the difference between the measured value and predicted value of the second sensor. As shown in FIG. 6(d), based on the determination signal in the period of interest being "1", the monitoring device 100 determines that there has been a change in the control signal during the period of interest, makes the upper threshold during the period of interest larger than at other times, and makes the lower threshold smaller than at other times. By increasing the upper threshold and decreasing the lower threshold during the period of interest, even when a mutation occurs in the difference between the measured value and predicted value of the second sensor despite the second sensor and its sensing target being normal, it becomes difficult for the mutation in the difference between the measured value and predicted value of the second sensor to exceed the upper threshold or fall below the lower threshold. Therefore, it is possible to suppress the occurrence of false detection that is regarded as an abnormality when a mutation occurs in the difference between the measured value and predicted value of the second sensor despite the second sensor and its sensing target being normal.

[0092] Further, FIG. 7 is a diagram for explaining a method of suppressing false detection caused by insufficient training data during model training by making it difficult to detect an abnormality during a period when the control signal changes. FIG. 7(a) is a diagram showing an example of how a mutation occurs in the control signal. The horizontal axis of FIG. 7(a) indicates time. The vertical axis of FIG. 7(a) indicates the value of the control signal. In FIG. 7(a), at the time of interest t C and the start time is tC-M During the period of interest between them, the control signal is changing.

[0093] FIG. 7(b) is a diagram showing an example of a determination signal when a mutation occurs in the control signal as shown in FIG. 7(a). The horizontal axis of FIG. 7(b) indicates the time synchronized with the horizontal axis of FIG. 7(a). The vertical axis of FIG. 7(b) indicates the value of the determination signal. In FIG. 7(b), the determination signal is "0" at times other than the period of interest. Also, in the period of interest, it is determined that the control signal has changed, and the determination signal is "1".

[0094] FIG. 7(c) is a diagram showing an example of the measured value and predicted value of the first sensor when a mutation occurs in the control signal as shown in FIG. 7(a). The horizontal axis of FIG. 7(c) indicates the time synchronized with the horizontal axes of FIGS. 7(a) and 7(b). The vertical axis of FIG. 7(c) indicates the value of the first sensor. The dashed line in FIG. 7(c) indicates the predicted value of the first sensor. The solid line in FIG. 7(c) indicates the measured value of the first sensor. In FIG. 7(c), during the period of interest, a mutation has occurred in the measured value of the first sensor in conjunction with the change in the control signal.

[0095] Also, FIG. 7(d) is a diagram showing an example of the difference between the measured value and predicted value of the first sensor when a mutation occurs in the control signal as shown in FIG. 7(a). The dashed line in FIG. 7(d) represents the difference between the measured value and predicted value of the first sensor. This difference is an example of the difference represented by Equation (4). In FIG. 7(d), during the period of interest, a mutation has also occurred in the difference between the measured value and predicted value of the first sensor.

[0096] The dashed-dotted line in Fig. 7(d) represents the upper threshold and the lower threshold for the difference between the measured value and the predicted value of the first sensor. As shown in Fig. 7(d), based on the determination signal being "1" during the period of interest, the monitoring device 100 determines that there has been a change in the control signal during the period of interest, and sets the upper threshold during the period of interest to be larger than at other times and the lower threshold to be smaller than at other times. Thereby, even when a mutation occurs in the measured value of the first sensor although the first sensor and its sensing target are normal, it becomes difficult for the difference between the measured value and the predicted value of the first sensor to exceed the upper threshold or fall below the lower threshold. For this reason, it is possible to suppress the occurrence of false detection that is regarded as an abnormality when a mutation occurs in the measured value of the first sensor although the first sensor and its sensing target are normal.

[0097] As described above, according to the monitoring device 100 according to the present embodiment, when a mutation occurs due to a change in the control signal, a determination signal indicating that the control signal has changed is generated, and control is performed so that it becomes difficult to detect an abnormality during the period in which the control signal has changed. Thereby, even when the absolute value of the difference between the measured value and the predicted value increases in the first sensor and the second sensor due to a change in the control signal although each sensor and its sensing target are normal, it is possible to suppress the occurrence of false detection in which an abnormality is erroneously detected by being controlled so that it becomes difficult to detect an abnormality when the determination signal changes.

[0098] Note that, as the determination signal, instead of a signal indicating the presence or absence of a change in the control signal, a signal indicating the amount of change in the control signal may be used. In this case, abnormality detection can be performed more flexibly. For example, the larger the amount of change in the control signal, the larger the threshold value used for abnormality determination using the absolute value of the difference between the measured value and the predicted value of each sensor. Thereby, it is possible to make it more difficult to detect an abnormality as the amount of change in the control signal increases.

[0099] In addition, in this embodiment, for all the sensors included in the second sensor set, control is performed so that it is difficult to detect an abnormality when the control signal changes. However, control may be performed so that it is difficult to detect an abnormality only for some of the sensors included in the second sensor set when the control signal changes. In this case, for example, when the control signal changes, the abnormality detection unit 106 makes it difficult to detect an abnormality only for each sensor belonging to the third sensor set, which is a subset of the second sensor set. The third sensor set is composed of only the sensors corresponding to the first sensor and the sensors corresponding to the second sensor among the sensors belonging to the second sensor set. Here, the first sensor and the second sensor are sensors in which a mutation occurs in the measured value or the predicted value in conjunction with a change in the control signal, and are sensors that may cause false detection. On the other hand, among the sensors belonging to the second sensor set, sensors that do not correspond to either the first sensor or the second sensor do not change the measured value or the predicted value in conjunction with a change in the control signal, and thus are less likely to cause false detection. In this modified example, by performing control to make it difficult to detect an abnormality only for sensors that are likely to cause false detection, false detection can be effectively suppressed.

[0100] Also, in this embodiment, the case where one type of signal is used as the control signal has been described. However, a plurality of types of control signals may be used. In this case, for each control signal, a set of sensors that may cause false detection due to a change in the control signal is preset. The control signal acquisition unit 102 acquires a plurality of control signals. The determination unit 104 determines whether there is a change during a predetermined operation mode for each of the plurality of control signals. Then, the abnormality detection unit 106 controls so that it is difficult to detect an abnormality only for the sensors associated with the changed control signal. Thereby, abnormality can be detected more accurately.

[0101] (First Modified Example) A description will be given of the first modification example. This modification example is obtained by modifying the configuration of the first embodiment as follows. Descriptions of the same configurations, operations, and effects as those of the first embodiment will be omitted. The monitoring device 100 of this modification example differs from the first embodiment in that it controls the detection of abnormalities using the continuous time of threshold exceedance.

[0102] FIG. 8 is a diagram showing the configuration of the monitoring device 100 according to this modification example. The abnormality detection unit 106 further includes a provisional detection unit 801 and a continuous time acquisition unit 802.

[0103] The provisional detection unit 801 provisionally detects the exceedance of a threshold for at least one sensor belonging to the second sensor set based on the measured values of the sensors belonging to the second sensor set included in the measured value data and the second data. For example, the provisional detection unit 801 detects the exceedance of the threshold by comparing the absolute value of the difference between the measured value and the predicted value of the sensor belonging to the second sensor set with the threshold.

[0104] The continuous time acquisition unit 802 acquires the continuous time of threshold exceedance for at least one sensor belonging to the second sensor set. For example, when the processes of steps S201 to S207 in the above-described monitoring process are executed over time at equal intervals, the continuous time of threshold exceedance is acquired by counting the number of times the absolute value of the difference between the measured value and the predicted value of the sensor belonging to the second sensor set continuously exceeds the threshold.

[0105] The abnormality detection unit 106 detects an abnormality in at least one sensor belonging to the second sensor set based on the measured values of the sensors belonging to the second sensor set included in the measurement value data, the second data, the determination signal, the detection result of threshold value exceedance by the provisional detection unit 801, and the continuous time of threshold value exceedance acquired by the continuous time acquisition unit 802. The detection at this time follows, for example, the case classification in FIG. 9. FIG. 9 is a diagram showing an example of the case classification executed in the control of abnormality detection. In the cases of FIGS. 9(a) and 9(c), that is, when the control signal has not changed regardless of the continuous time of threshold value exceedance, the abnormality detection unit 106 determines that the threshold value exceedance is not due to the change in the control signal and there is a possibility of an abnormality, and does not perform control to make it difficult to detect the abnormality. In the case of FIG. 9(b), that is, when the control signal has changed and the continuous time of threshold value exceedance is short, the abnormality detection unit 106 determines that the threshold value exceedance may be a normal behavior linked to a sudden change in the control signal, and performs control to make it difficult to detect the abnormality. In the case of FIG. 9(d), that is, when the control signal has changed and the continuous time of threshold value exceedance is long, the abnormality detection unit 106 determines that the threshold value exceedance is not a normal behavior linked to a sudden change in the control signal and may be due to an abnormality to be detected, and does not perform control to make it difficult to detect the abnormality. In FIG. 9, whether the time of threshold value exceedance is long or short is determined, for example, by whether the continuous time of threshold value exceedance is equal to or greater than a predetermined time R. The predetermined time R is, for example, stored in advance in a storage medium. The predetermined time R is set, for example, according to the length of the mutation occurring in the signal acquired as the control signal by the control signal acquisition unit 102.

[0106] Next, the operation of the processing executed by the monitoring device 100 will be described. FIG. 10 is a flowchart showing an example of the procedure of the abnormality detection process when the abnormality detection unit 106 follows FIG. 9 and the method of control for making it difficult to detect an abnormality is the change of the threshold value. For each sensor belonging to the second sensor set, an abnormality detection signal is created according to the detection procedure shown in this flowchart. Hereinafter, the sensor to be processed will be referred to as the target sensor. Note that the processing procedures in each of the processes described below are merely examples, and each process can be changed as appropriate as much as possible. Also, regarding the processing procedures described below, steps can be omitted, replaced, and added as appropriate according to the embodiment.

[0107] (Abnormality Detection Process) (Step S1001) The provisional detection unit 801 tentatively determines whether the absolute value of the difference between the measured value and the predicted value of the target sensor is greater than the threshold value. The provisional detection unit 801 sends the provisional detection result to the continuous time acquisition unit 802 as the provisional detection result.

[0108] (Step S1002) The continuous time acquisition unit 802 acquires the provisional detection result regarding the threshold value exceeding of the target sensor. For example, the process of step S1001 is repeatedly executed at regular time intervals until the monitoring is stopped. Therefore, the continuous time acquisition unit 802 can calculate the continuous time of the threshold value exceeding of the target sensor by counting the number of times the threshold value has been exceeded in the process of step S1001.

[0109] (Step S1003) Based on the determination signal, the abnormality detection unit 106 determines whether there is a change in the control signal during the target period. For example, when "0" is obtained as the determination signal (step S1003 - No), the abnormality detection unit 106 determines that the control signal has not changed during the target period. On the other hand, when "1" is obtained as the determination signal (step S1003 - Yes), the abnormality detection unit 106 determines that the control signal has changed during the target period. The processing in this step is common regardless of the target sensor as long as the target period is the same. Therefore, when the processing of step S1003 is executed for a certain target sensor, the processing result can be reused for other target sensors.

[0110] (Step S1004) When the control signal changes at the target time (step S1003 - Yes), the abnormality detection unit 106 determines whether the continuous time of the threshold exceedance of the target sensor is less than the threshold. When the continuous time of the threshold exceedance of the target sensor is less than the threshold (step S1004 - Yes), the abnormality detection unit 106 determines that the threshold exceedance may be caused by a sudden change in the control signal. On the other hand, when the continuous time of the threshold exceedance is equal to or greater than the threshold (step S1004 - No), the abnormality detection unit 106 determines that the threshold exceedance is not caused by a sudden change in the control signal.

[0111] (Step S1005) When the continuous time of the threshold exceedance is less than the threshold (step S1004 - Yes), the abnormality detection unit 106 determines that the threshold exceedance of the target sensor is caused by a sudden change in the control signal, and sets the threshold of the target sensor to infinity during the target period. As a result, no abnormality is detected by the target sensor during the target period. At this time, instead of setting the threshold to infinity, it can also be set to a large value. In this case, it becomes difficult to detect an abnormality by the target sensor.

[0112] (Step S1006) The abnormality detection unit 106 determines whether the absolute value of the difference between the measured value and the predicted value of the target sensor exceeds a threshold value, generates an abnormality detection signal, and outputs it to the outside of the monitoring device 100. Here, the threshold value is controlled in step S1005. Therefore, if both branches in steps S1003 and S1004 are Yes, and the threshold value is set to infinity for the sensor in step S1005, no abnormality will be detected. If the threshold value is not infinity but is set to a large value in step S1005, it will be difficult to detect an abnormality for the sensor with the threshold value set to a large value.

[0113] Hereinafter, the effects of the monitoring device 100 according to this modification example will be described.

[0114] If an excess of the threshold value for the second sensor included in the second sensor set occurs due to being linked to a change in the control signal, the excess time of the threshold value is likely to be short. On the other hand, even when the control signal changes, if the excess time of the threshold value for the second sensor included in the second sensor set is long, it is less likely to be due to being linked to a change in the control signal, and more likely to be due to the abnormality to be detected. If it becomes difficult to detect an abnormality based on the change in the control signal, undetected cases where the abnormality to be detected is missed may occur.

[0115] The monitoring device 100 according to this modification example tentatively detects an excess of the threshold value for at least one sensor belonging to the second sensor set based on the measured value of each sensor belonging to the second sensor set, the second data, and a predetermined threshold value, obtains the continuous time of the excess of the threshold value, and when the continuous time is shorter than the predetermined time and the determination signal indicates that there is a change in the control signal, it is possible to make it more difficult to detect an abnormality compared to the case where the determination signal indicates no change.

[0116] With the above configuration, according to the monitoring device 100 according to this modification example, by controlling so that an abnormality is only detected with difficulty when there is a change in the control signal and the excess time of the threshold value is short, it is possible to suppress undetected cases where the abnormality to be detected is missed.

[0117] In the system, there are more control signals that cause mutations in the measured values as the system is larger. It is not always possible for the control signal acquisition unit 102 to acquire all of them as control signals. For example, when an external disturbance signal with a shorter mutation than the acquirable control signal cannot be acquired as a control signal, false detections caused by mutations in the first sensor in conjunction with changes in the external disturbance signal, or false detections caused by the linkage of the predicted value of the second sensor to the mutation of the first sensor may not be suppressed.

[0118] By using a plurality of values as the threshold for the threshold exceeding time, such false detections can be suppressed. For example, in addition to a predetermined time R, a predetermined time R' shorter than time R is used. In this case, the abnormality detection unit 106 follows, for example, FIG. 11 instead of FIG. 9. FIG. 11 is a diagram showing an example of case division executed in the control of abnormality detection. In the cases of FIGS. 11(a) and 11(b), that is, when the continuous time of threshold exceeding is less than time R', the abnormality detection unit 106 controls so that it is difficult to detect an abnormality regardless of the presence or absence of a change in the control signal. In the case of FIG. 11(c), that is, when the continuous time of threshold exceeding is equal to or greater than time R', less than time R, and the determination signal indicates that there is no change in the control signal, the abnormality detection unit 106 does not control so that it is difficult to detect an abnormality. In the case of FIG. 11(d), that is, when the continuous time of threshold exceeding is equal to or greater than time R', less than time R, and the determination signal indicates that there is a change in the control signal, the abnormality detection unit 106 controls so that it is difficult to detect an abnormality. In the cases of FIGS. 11(e) and 11(f), that is, when the continuous time of threshold exceeding is equal to or greater than time R, the abnormality detection unit 106 does not control so that it is difficult to detect an abnormality regardless of the presence or absence of a change in the control signal. By performing such control, when the continuous time of a temporary detection is shorter than the predetermined time R', it becomes difficult to detect regardless of the determination signal. As a result, false detections caused by changes in external disturbance signals and the like that cannot be acquired as control signals can also be suppressed.

[0119] (Second Modification Example) A second modification will be described. This modification is obtained by modifying the configuration of the first embodiment as follows. Regarding the configuration, operation, and effects similar to those of the first embodiment, the description will be omitted. The monitoring device 100 of this modification is different from the first embodiment in that it calculates the degree of abnormality of the system using the measured values of the sensors belonging to the second sensor set and the second data, and determines the presence or absence of an abnormality in the system using the calculated degree of abnormality.

[0120] FIG. 12 is a diagram showing the configuration of the monitoring device 100 according to this modification. The processing circuit of the monitoring device 100 further includes an abnormality degree calculation unit 1201.

[0121] The abnormality degree calculation unit 1201 calculates the degree of abnormality of the system based on the measured value data and the second data. Specifically, the abnormality degree calculation unit 1201 calculates the degree of abnormality of the system based on the measured values of the sensors belonging to the second sensor set included in the measured value data and the predicted values of the sensors belonging to the second sensor set included in the second data.

[0122] The abnormality detection unit 106 detects an abnormality in the system by comparing the magnitude relationship between the degree of abnormality of the system and a predetermined threshold. For example, when the degree of abnormality is greater than the threshold, the abnormality detection unit 106 detects that there is an abnormality in the system. On the other hand, when the degree of abnormality is less than or equal to the threshold, the abnormality detection unit 106 assumes that there is no abnormality in the system. Also, similar to the first embodiment, when the determination signal indicates a change, the abnormality detection unit 106 makes it difficult to detect an abnormality.

[0123] Next, the operation of the processing executed by the monitoring device 100 will be described. FIG. 13 is a flowchart showing an example of the procedure of the monitoring process. The processes of steps S1301 - S1305 are the same as the processes of steps S201 - S205 of the first embodiment, so the description will be omitted. Note that the processing procedures in each of the processes described below are merely examples, and each process can be changed as appropriate as much as possible. Also, regarding the processing procedures described below, depending on the embodiment, steps can be omitted, replaced, and added as appropriate.

[0124] (Monitoring Process) (Step S1306) The abnormality degree calculation unit 1201 of the abnormality detection unit 106 acquires measurement value data and second data. The abnormality degree calculation unit 1201 acquires the measurement values of each sensor belonging to the second sensor set from the measurement value data. Also, the abnormality degree calculation unit 1201 extracts the predicted values of each sensor belonging to the second sensor set from the second data.

[0125] The abnormality degree calculation unit 1201 calculates the abnormality degree of the system using, for example, the following formula (6).

[0126] [Number]

[0127] In formula (6), A t is the abnormality degree of the system at time t. ^y t represents a vector obtained by arranging the predicted values of each sensor belonging to the second sensor set included in the second data. y t represents a vector obtained by arranging the measurement values of each sensor belonging to the second sensor set included in the measurement value data at time t in the same order as the second data.

[0128] (Step S1307) The abnormality detection unit 106 acquires the abnormality degree of the system and a determination signal. The abnormality detection unit 106 executes an abnormality detection process using the abnormality degree of the system. In the abnormality detection process, when the abnormality degree of the system is greater than a predetermined threshold value, the abnormality detection unit 106 determines that there is an abnormality in the system. Also, when the abnormality degree of the system is less than or equal to the threshold value, the abnormality detection unit 106 determines that there is no abnormality in the system. Also, similar to the first embodiment, when the control signal changes at the attention time, the abnormality detection unit 106 makes it difficult to detect an abnormality.

[0129] Hereinafter, the effects of the monitoring device 100 according to this modification will be described.

[0130] The abnormality degree of the system is generated by referring to the predicted values and measured values of each sensor belonging to the second sensor set included in the second data. Therefore, the abnormality degree of the system changes in conjunction with the change in the measured value of the first sensor included in the second sensor set. Also, the measured value of the first sensor included in the second sensor set changes in conjunction with the mutation of the control signal. That is, a mutation occurs in the abnormality degree of the system in conjunction with the change in the control signal. As described above, in the system, it is normal behavior for the measured value of the first sensor to change after the control signal changes.

[0131] When the abnormality degree of the system exceeds the threshold value, in the abnormality detection process, it is determined that there is an abnormality in the system. For this reason, when a mutation occurs in the control signal, even if the behavior of the measured value of the first sensor is normal, there is a high possibility of a false detection in which the system is detected as abnormal. Thus, when performing abnormality detection by utilizing a model that utilizes the correlation relationship between sensors, a false detection may occur in which the system is determined to be abnormal even though it is normal, due to the linkage between the change in the control signal and the measured value of the first sensor.

[0132] The monitoring device 100 according to this modification example can calculate the abnormality degree of the system based on the measured values of each sensor belonging to the second sensor set and the second data, and detect the abnormality of the system based on the abnormality degree and the threshold value.

[0133] With the above configuration, according to the monitoring device 100 according to this modification example, even when the abnormality degree of the system increases due to the change in the control signal although the system is normal, it is controlled so that it becomes difficult to detect an abnormality when the determination signal changes, thereby suppressing the occurrence of a false detection in which an abnormality is erroneously detected.

[0134] FIG. 14 is a diagram for explaining a method of suppressing false detection by the monitoring device 100 according to this modified example. FIG. 14(a) is a diagram showing an example of the time-series change of a control signal in a predetermined operation mode. The horizontal axis of FIG. 14(a) indicates time. The vertical axis of FIG. 14(a) indicates the value of the control signal. FIG. 14(b) represents an example of a time-series graph of a determination signal generated as a determination result for the control signal shown in FIG. 14(a). The horizontal axis of FIG. 14(b) indicates time. The vertical axis of FIG. 14(b) represents the value of the determination signal. The horizontal axis of FIG. 14(b) is synchronized with the horizontal axis of FIG. 14(a).

[0135] FIG. 14(c) is a diagram showing an example of the time-series change of the abnormality degree of the system. The horizontal axis of FIG. 14(c) indicates the time synchronized with the horizontal axes of FIGS. 14(a) and 14(b). The vertical axis of FIG. 14(c) indicates the value of the abnormality degree of the system. The solid line in FIG. 14(c) indicates the abnormality degree. In FIG. 14(c), a sudden change also occurs in the abnormality degree in conjunction with the change of the control signal. Also, the dashed-dotted line in FIG. 14(c) indicates the threshold value for the abnormality degree. In FIGS. 14(b) and 14(c), it can be seen that when the determination signal changes during the period of interest, the control signal during the period of interest changes and the threshold value during the period of interest becomes larger than the threshold value at other times.

[0136] (Third Modified Example) Note that the information used in the monitoring process and the abnormality detection process and the detection result of the abnormality may be displayed on a display. In this case, the monitoring device 100 includes a display control unit that controls the screen to be displayed on the display. The display control unit, for example, arranges or overlaps a first graph representing the time-series change of the measured value, the predicted value, or the difference between the measured value and the predicted value and a second graph representing the time-series change of the control signal or the determination signal with the time-axis scales aligned, and displays them on the display. The display may be provided inside the monitoring device 100 or outside the monitoring device 100. By checking the information displayed on the display, the user can know the reason why an abnormality was detected or the reason why no abnormality was detected.

[0137] For example, the display control unit displays a time series graph of the control signal on the display as shown in Fig. 6(a). Fig. 6(a) is an example of the second graph. In this case, the user can easily check the timing at which the control signal changes.

[0138] Alternatively, the display control unit may display a time series graph of the determination signal on the display, as shown in Fig. 6(b). Fig. 6(b) is an example of a second graph. In this case, the user can easily confirm the timing at which the determination signal changes and control is performed to make it difficult to detect an abnormality.

[0139] Alternatively, the display control unit may display a time series graph of the measured values and predicted values of the second sensor on the display, as shown in Figure 6(c). Figure 6(c) is an example of the first graph. In this case, the user can easily check the changes in the measured values and predicted values of the second sensor.

[0140] Alternatively, the display control unit may display on the display a time series graph of the difference between the measured value and the predicted value of the second sensor included in the second sensor set, as shown in Figure 6(d). Figure 6(d) is an example of the first graph. In this case, the user can easily confirm the change in the difference between the measured value and the predicted value of the second sensor included in the second sensor set and the change in the threshold controlled in response to the determination signal.

[0141] The display control unit may also display the time series graphs shown in Figures 6(a) to 6(d) side by side on the display. In this case, the user can easily confirm the timing at which the determination signal increased in magnitude due to a change in the control signal. The user can also easily confirm whether the reason why the absolute value of the difference between the measured value and the predicted value of the second sensor was not considered to be an abnormality despite the large absolute value of the difference is because the sensor was controlled to make the abnormality difficult to detect. The user can also easily confirm whether the increase in the absolute value of the difference between the measured value and the predicted value is due to a sudden change in the measured value of the first sensor caused by a change in the control signal, or due to a correlation between the measured value of the first sensor and the predicted value of the second sensor.

[0142] Also, as shown in FIGS. 6(a) - 6(d), a vertical line representing a specific time may be displayed on the display. In this case, the times in FIGS. 6(a) - 6(d) can be synchronized and are easier to view. It is preferable to enable a user to set the time of the vertical line via a user interface such as a mouse.

[0143] Also, each of the time series graphs shown in FIGS. 6(a) - 6(d) may be overlaid and displayed within a common area sharing the vertical axis. In this case, the display space can be reduced.

[0144] Also, the display control unit may display, side by side or overlaid, on the display, a second graph representing the time series change of a control signal or a determination signal and a third graph representing the time series change of the abnormality degree of the system, with the time axis scales aligned.

[0145] For example, the display control unit may display on the display a time series graph of the abnormality degree as shown in FIG. 14(c) of the second modification example. The graph shown in FIG. 14(c) is an example of the third graph. In this case, the user can easily confirm the change in the abnormality degree and the change in the threshold value controlled according to the determination signal.

[0146] Also, the time series graphs shown in FIGS. 14(a) - 14(c) of the second modification example may be displayed side by side on the display. The graphs shown in FIGS. 14(a) and 14(b) are examples of the second graph. In this case, the user can easily confirm whether the timing when the abnormality degree increases is after the timing when the control signal changes, etc.

[0147] Also, as shown in FIGS. 14(a) - 14(c), a vertical line representing a specific time may be displayed. In this case, the times in FIGS. 14(a) - 14(c) can be synchronized and are easier to view. It is preferable to enable a user to set the time of the vertical line via a user interface such as a mouse.

[0148] Also, each of the time series graphs shown in FIGS. 14(a) to 14(c) may be displayed by overlapping them within a common area that shares the vertical axis. In this case, the display space can be reduced.

[0149] (Fourth Modification Example) Also, the predetermined time M, which is the length of the period of interest, may be changed as appropriate. In this case, for example, the determination unit 104 sets the predetermined time M, which is the determination time of the change in the control signal, according to the type of the control signal, the type of the first sensor, the type of the second sensor, or the type of the system. For example, a table in which the types of the above-described devices or systems and the set values of time M are associated is stored in a storage medium, and an appropriate set value is set using this table. Also, the set value of the predetermined time M may be input from the outside of the monitoring device 100 by a user or the like.

[0150] The time from when a change in the control signal occurs until a sudden change occurs in the measured value of the first sensor varies depending on the control signal and the type of the first sensor. Also, when the state of the system changes, the time from when a change occurs in the control signal until a sudden change occurs in the measured value of the first sensor may change. Even in such a case, by appropriately setting the predetermined time M, it is possible to cope with the change in the time from when a change in the control signal occurs until a sudden change occurs in the measured value of the first sensor.

[0151] Also, when the monitoring device 100 is used to monitor a plurality of different systems, the time from when a change in the control signal occurs until a sudden change occurs in the measured value of the first sensor varies between the systems. Even in this case, by appropriately setting the predetermined time M according to the system to be monitored, it is possible to cope with a plurality of different systems.

[0152] In addition, the time M, which is the length of the attention period, may be displayed on the display. In this case, for example, the time M may be displayed in a dialog box. Alternatively, the time M may be displayed on a graph showing the time-series change of the determination signal. FIG. 15 is a diagram showing an example of a display screen for displaying the time M. FIG. 15(a) is a diagram showing an example of the time-series change of a control signal in a predetermined operation mode. The horizontal axis in FIG. 15(a) indicates time. The vertical axis in FIG. 15(a) indicates the value of the control signal. FIG. 15(b) is a diagram showing a state where a predetermined time M is displayed on the time-series graph of the determination signal. The horizontal axis in FIG. 15(b) indicates time. The vertical axis in FIG. 15(b) indicates the value of the determination signal. By displaying the predetermined time M on the display, the user can easily confirm the setting of the time M. In particular, when the time M is changed according to the types of the control signal and the first sensor and the change of the system to be monitored, the significance of being able to confirm the setting of the time M is great.

[0153] Also, as shown in FIG. 15, the time M may be displayed by showing auxiliary lines in the time-series graphs of the control signal and the determination signal. In this case, it is more preferable to enable the user to move the horizontal position of the auxiliary line via a user interface such as a mouse.

[0154] (Fifth Modification Example) Note that the predetermined time R used as the threshold value in the second modification example may be changed as appropriate. In this case, for example, the abnormality detection unit 106 sets a predetermined time R, which is the threshold value for the continuous time exceeding the threshold value, according to the type of the control signal, the type of the first sensor, the type of the second sensor, or the type of the system. At this time, the time R may be set for each sensor. For example, a table in which the types of the above-described devices and systems, or sensor IDs and sensor names are associated with the set values of the time R is stored in a storage medium, and appropriate set values are set using this table. Further, the set value of the predetermined time R may be input from outside the monitoring device 100 by the user or the like.

[0155] The duration of the mutation occurring in the measurement value of the first sensor due to the change in the control signal, and the duration of the mutation occurring in the predicted value of the second sensor in conjunction with the mutation in the measurement value of the first sensor, vary depending on the types and combinations of the control signal, the first sensor, and the second sensor. Also, the durations of these mutations vary depending on the state of the system. Even in such cases, by appropriately setting a predetermined time R, it is possible to cope with changes in the duration of the mutation.

[0156] Also, when the monitoring device 100 is used to monitor a plurality of different systems, the duration of the mutation varies between systems. Also in this case, by appropriately setting the predetermined time R according to the system to be monitored, it is possible to cope with a plurality of different systems.

[0157] Also, the predetermined time R may be displayed on the display. In this case, for example, the time R may be displayed in a dialog box. Alternatively, the time R may be displayed on a graph showing the time-series change in the difference between the measured value and the predicted value of the second sensor. FIG. 16 is a diagram showing an example of a display screen on which the time R, which is the threshold for the excess time, is displayed. The horizontal axis in FIG. 16 indicates time. The vertical axis in FIG. 16 indicates the value of the second sensor. The dashed line in FIG. 16 indicates the difference between the measured value and the predicted value of the second sensor. The one-dot chain line in FIG. 16 indicates the threshold. By displaying the predetermined time R on the display, the user can easily confirm the setting of the time R. In particular, when the time R is changed according to changes in the types of the control signal, the first sensor, and the second sensor, and the system to be monitored, the significance of being able to confirm the setting of the time R is great. Also, the user can easily confirm the relationship between the time during which the difference between the measured value and the predicted value of the second sensor continuously exceeds the threshold and the time R.

[0158] Also, as shown in FIG. 16, the time R may be displayed by showing auxiliary lines in the time-series graph of the difference between the measured value and the predicted value of the second sensor. In this case, it is more preferable to enable the user to move the horizontal position of the auxiliary line via a user interface such as a mouse.

[0159] (Sixth Modification Example) In the first embodiment, as the first data (x t ) input to the model, an example was described in which a vector obtained by arranging measurement values of each sensor belonging to the first sensor set at time t, or values obtained by normalizing or standardizing the measurement values, was used. However, a vector obtained by arranging measurement values of each sensor belonging to the first sensor set at a single time before time t, or values obtained by normalizing or standardizing the measurement values, may be used as the first data (x t ).

[0160] For example, when the first data (x t ) is a vector obtained by arranging measurement values of a single time before time t, or values obtained by normalizing or standardizing the measurement values, the second data (^y t ) at time t will be predicted from the first data (x t ) consisting of data of a single time before time t. Also, when the first data (x t ) is a vector obtained by arranging measurement values of a plurality of times before time t, or values obtained by normalizing or standardizing the measurement values, the second data (^y t ) at time t will be predicted from the first data (x t ) consisting of data of a plurality of times before time t. Also, when the first data (x t ) is a vector obtained by arranging measurement values of time t and times before time t, or values obtained by normalizing or standardizing the measurement values, the second data (^y t ) at time t will be predicted from the first data (x t ) consisting of data of a plurality of times before time t.

[0161] Since this modification example relates to a modification example of the input / output of the model, attention is required. When the first data (x t ) is a vector obtained by arranging measurement values of time t and times before time t, or values obtained by normalizing or standardizing the measurement values, y t representing a vector obtained by arranging measurement values of each sensor included in the second sensor set at time t or the second data (^y tRegarding as well, after using the data at time t and the data at times before time t, train the machine learning model, and the predicted value at time t may be generated by extracting the portion at time t from the second data (^y t ) output by the model (f).

[0162] In this modification, although the content of the first data (x t ) is changed, the same effect as in the first embodiment can be obtained.

[0163] (Seventh Modification) The first data may include a control signal in addition to the measurement values of the sensors belonging to the first sensor set including the first sensor as an element. In this case, the model is trained using the first data including the control signal. By using the model trained by the prediction unit 105 using the first data including the control signal, the linkage between the measurement value of the first sensor in the predicted value of the second sensor is reduced according to the value of the control signal, and the reproducibility of the predicted value of the second sensor with respect to the measurement value of the second sensor is enhanced. Therefore, false detection caused by the linkage of the predicted value of the second sensor with respect to the measurement value of the first sensor can be suppressed.

[0164] Thus, according to any of the above-described embodiments, it is possible to provide a monitoring device, method, and program that can suppress false detection in the abnormality detection of the system.

[0165] Although some embodiments of the present invention have been described, these embodiments are presented as examples and are not intended to limit the scope of the invention. These embodiments can be implemented in various other forms, and various omissions, replacements, and changes can be made without departing from the gist of the invention. These embodiments and their modifications are included in the scope and gist of the invention, and are also included in the invention described in the claims and its equivalent scope.

Explanation of Reference Numerals

[0166] 100... Monitoring device, 101... Measured value acquisition unit, 102... Control signal acquisition unit, 103... Model acquisition unit, 104... Determination unit, 105... Prediction unit, 106... Abnormality detection unit, 113... Difference calculation unit, 801... Temporary detection unit, 802... Continuous time acquisition unit, 1001... Abnormality degree calculation unit, t1, t2... Time, t C ... Time of interest, t C-M ... Start time, M... Time, R... Threshold value.

Claims

1. A measured value acquisition unit that acquires measured values of a plurality of sensors installed in the system, A control signal acquisition unit that acquires a control signal from the system, When the first data including the measured values of each sensor belonging to the first sensor set including the first sensor in which a mutation occurs in the measured value when the control signal changes during a predetermined operation mode is input, a model acquisition unit that acquires a model that generates second data including predicted values of each sensor belonging to the second sensor set including the second sensor correlated with the first sensor, A determination unit that generates a determination signal by determining a change in the control signal, A prediction unit that generates the second data including the predicted values of each sensor belonging to the second sensor set from the first data included in the measured values and the model, An abnormality detection unit that detects an abnormality in the system or an abnormality in at least one sensor belonging to the second sensor set based on the measured values of each sensor belonging to the second sensor set, the second data, the determination signal, and a threshold value, Comprising, When the determination signal indicates that there is a change in the control signal during the predetermined operation mode, the abnormality detection unit makes it difficult to detect the abnormality, Monitoring device.

2. The control signal includes any one of a feedback control signal, a feedforward control signal, and a control signal having a weak correlation with a change in the operation mode and changing irregularly. The monitoring device according to claim 1.

3. The control signal includes any one of a signal of a BIR (boiler input regulator), a signal of a soot blower, a signal related to startup or stop of a fuel device, a signal indicating the number of startups of a fuel device, and a signal indicating a change in fuel type. The monitoring device according to claim 1.

4. When the determination signal indicates that there is a change in the control signal, the abnormality detection unit does not perform detection of the abnormality. The monitoring device according to any one of claims 1 to 3.

5. The abnormality detection unit changes the threshold value according to the determination signal. The monitoring device according to any one of claims 1 to 3.

6. The abnormality detection unit, Based on the measured values of each sensor belonging to the second sensor set, the second data, and a predetermined threshold value, a preliminary detection unit that preliminarily detects an excess of the threshold value related to the system or at least one sensor belonging to the second sensor set. a continuous time acquisition unit that acquires the continuous time of the exceeding of the threshold value; when the continuous time is shorter than a predetermined time and the determination signal indicates that there is a change in the control signal, it is less likely to detect the abnormality compared to the case where the determination signal indicates no change; The monitoring device according to any one of claims 1 to 5.

7. The abnormality detection unit further includes an abnormality degree calculation unit that calculates the abnormality degree of the system based on the measurement values of the sensors belonging to the second sensor set and the second data; detects the abnormality of the system based on the abnormality degree and the threshold value; The monitoring device according to any one of claims 1 to 6.

8. The abnormality detection unit has a difference calculation unit that calculates the difference between the measurement value and the predicted value of the sensor belonging to the second sensor set; detects the abnormality of the sensors belonging to the second sensor set based on the difference and the threshold value; The monitoring device according to any one of claims 1 to 6.

9. When the determination signal indicates that there is a change in the control signal, the abnormality detection unit makes it less likely to detect the abnormality only for the sensors belonging to the third sensor set that includes only the first sensor and the second sensor among the sensors belonging to the second sensor set; The monitoring device according to any one of claims 1 to 8.

10. further includes a display control unit that arranges or overlaps a first graph representing the time-series change of the measurement value, the predicted value, or the difference between the measurement value and the predicted value and a second graph representing the time-series change of the control signal or the determination signal with the time-axis scales aligned and displays them on a display; The monitoring device according to any one of claims 1 to 9.

11. further includes a display control unit that arranges or overlaps a second graph representing the time-series change of the control signal or the determination signal and a third graph representing the time-series change of the abnormality degree of the system with the time-axis scales aligned and displays them on a display; The monitoring device according to claim 7.

12. The determination unit sets the determination time of the change of the control signal according to the type of the control signal, the type of the first sensor, the type of the second sensor, or the type of the system; The monitoring device according to any one of claims 1 to 11.

13. The abnormality detection unit sets the predetermined time according to the type of the control signal, the type of the first sensor, the type of the second sensor, or the type of the system. The monitoring device according to claim 6.

14. Obtaining measurement values of a plurality of sensors installed in the system; Obtaining a control signal from the system; When inputting first data including measurement values of each sensor belonging to a first sensor set including as an element a first sensor in which a mutation occurs in the measurement value when the control signal changes during a predetermined operation mode, obtaining a model that generates second data including predicted values of each sensor belonging to a second sensor set including as an element a second sensor correlated with the first sensor; Generating a determination signal by determining a change in the control signal; Generating the second data including predicted values of each sensor belonging to the second sensor set from the first data included in the measurement values and the model; Detecting an abnormality of the system or an abnormality of at least one sensor belonging to the second sensor set based on the measurement values of each sensor belonging to the second sensor set, the second data, the determination signal, and a threshold value; When the determination signal indicates that there is a change in the control signal during the predetermined operation mode, making it difficult to detect the abnormality; A method comprising the steps of:

15. On a computer, a function of obtaining measurement values of a plurality of sensors installed in the system; a function of obtaining a control signal from the system; a function of obtaining a model that generates second data including predicted values of each sensor belonging to a second sensor set including as an element a second sensor correlated with the first sensor when inputting first data including measurement values of each sensor belonging to a first sensor set including as an element a first sensor in which a mutation occurs in the measurement value when the control signal changes during a predetermined operation mode; a function of generating a determination signal by determining a change in the control signal; a function of generating the second data including predicted values of each sensor belonging to the second sensor set from the first data included in the measurement values and the model; a function of detecting an abnormality of the system or an abnormality of at least one sensor belonging to the second sensor set based on the measurement values of each sensor belonging to the second sensor set, the second data, the determination signal, and a threshold value; When the determination signal indicates that there is a change in the control signal during the predetermined operation mode, a function that makes it difficult to detect the abnormality, A program for realizing the above.

Images