JP7225447B2 - Monitoring device, monitoring method and monitoring program - Google Patents

Description

The present invention relates to a monitoring device, a monitoring method and a monitoring program.

Conventionally, methods of predicting or detecting anomalies based on process data collected in the past have been proposed as techniques related to prediction of anomalies using process data from sensors in factories, plants, and the like.

In general, process data expressed as multivariate time-series data is clipped in a window of a certain period of time, and calculations are performed to calculate values related to anomaly prediction and anomaly detection. For example, if an abnormality is detected, an alarm is issued based on the calculated value to notify the occurrence of an abnormality or a forecast of the abnormality.

JP 2017-142654 A

The conventional method as described above has a problem that it is not possible to dynamically and easily confirm which input value is important among a plurality of input values. For example, when using the output of a model to predict or monitor a specific sensor, conventional multivariate analysis extracts, on average and statically, which inputs are considered important for the created model. Although it can be done, it was not possible to dynamically extract the importance of which input is emphasized for the input at a specific time. For this reason, it is difficult to take out important sensors based on the state of a system whose state changes from moment to moment.

For example, in the case of an anomaly detection system, the process for identifying the cause and recovering after an anomaly is notified can only extract a static degree of importance with conventional multivariate analysis. Since the effect of the input cannot be presented, experts who are familiar with the state of the system can only make judgments by manually looking at the input data at that time, and how to operate to avoid anomalies. I didn't know until

Furthermore, for example, when using a neural network as a machine learning method, several methods have been proposed for extracting important inputs, but the computational complexity is high in order to perform calculations for all combinations of inputs. It was difficult to put it into practical use because it would be huge. Therefore, it was difficult to extract important sensors in a system using a neural network.

In order to solve the above-described problems and achieve the object, the monitoring device of the present invention is a neural network for predicting the state of the monitored equipment by inputting a plurality of data acquired by the monitored equipment. Using a prediction means that outputs a predetermined output value using a trained model, each data at each time input to the trained model, and a predetermined output value output from the trained model, Calculating the degree of importance for the predetermined output value for each input item at each time, the degree of importance at each time changing with time indicating how much each of the data contributed to the output value of the trained model. and calculating means for calculating an output value from an input value in the learned model, using a partial differential value or an approximate value thereof with respect to each input value of the output value, at each time The importance is calculated for each input item.

Further, the monitoring method of the present invention is a monitoring method executed by a monitoring device, and is a neural network for predicting the state of the monitoring target equipment by inputting a plurality of data acquired by the monitoring target equipment. Using a prediction step of outputting a predetermined output value using a trained model, each data at each time input to the trained model, and a predetermined output value output from the trained model, Calculating the degree of importance for the predetermined output value for each input item at each time, the degree of importance at each time changing with time indicating how much each of the data contributed to the output value of the trained model. and a calculation step of calculating an output value from an input value in the learned model, using a partial differential value or an approximate value thereof with respect to each input value of the output value, the input at each time The importance is calculated for each item.

In addition, the monitoring program of the present invention receives a plurality of data acquired by the equipment to be monitored, uses a trained model that is a neural network for predicting the state of the equipment to be monitored, and outputs a predetermined output value. The predetermined output for each input item at each time using the prediction step to be output, each data at each time input to the learned model, and a predetermined output value output from the learned model a calculating step of calculating the importance of the value at each time that changes with time indicating how much each of the data contributed to the output value of the trained model; In the step, in the trained model that calculates the output value from the input value, the importance is calculated for each input item at each time using a partial differential value or an approximate value thereof with respect to each input value of the output value. characterized by

ADVANTAGE OF THE INVENTION According to this invention, it is effective in being able to confirm dynamically and easily which input value is important among several input values.

FIG. 1 is a block diagram showing a configuration example of a monitoring device according to the first embodiment. FIG. 2 is a diagram for explaining the abnormality prediction evaluation value calculation process executed by the abnormality prediction unit. FIG. 3 is a diagram illustrating an example of the degree of importance of each sensor calculated by the calculator. FIG. 4 is a diagram for explaining importance calculation processing using a saliency map executed by the calculation unit. FIG. 5 is a diagram for explaining noise removal processing executed by the calculator. FIG. 6 is a diagram explaining an outline of anomaly prediction processing executed by the monitoring device. FIG. 7 is a flowchart showing an example of the flow of abnormality prediction processing in the monitoring device according to the first embodiment. FIG. 8 is a flowchart showing an example of the flow of importance calculation processing in the monitoring device according to the first embodiment. FIG. 9 is a flowchart showing an example of the flow of graph display processing in the monitoring device according to the first embodiment. FIG. 10 is a diagram explaining an outline of the sensor value prediction process. FIG. 11 is a block diagram showing a configuration example of a monitoring system according to the second embodiment. FIG. 12 is a diagram showing a computer executing a monitoring program or a display program.

Embodiments of a monitoring device, a monitoring method, a monitoring program, a display device, a display method, and a display program according to the present application will be described below in detail with reference to the drawings. Note that the monitoring device, monitoring method, monitoring program, display device, display method, and display program according to the present application are not limited by this embodiment.

[First embodiment]
In the following embodiments, the configuration of the monitoring device 10 according to the first embodiment and the processing flow of the monitoring device 10 will be described in order, and finally the effects of the first embodiment will be described.

[Configuration of monitoring device]
First, the configuration of the monitoring device 10 will be described with reference to FIG. FIG. 1 is a block diagram showing a configuration example of a monitoring device according to the first embodiment. The monitoring device 10, for example, collects a plurality of data acquired by sensors installed in monitoring target equipment such as factories and plants, and uses the collected data as input to predict an abnormality in the monitoring target equipment. using the trained model, the frame abnormality evaluation value, which is the degree of abnormality of the equipment to be monitored, is output as an output value. In addition, the monitoring device 10 uses each data at each time input to the learned model and the degree of abnormality output from the learned model to calculate the importance of the output value for each input item at each time. . Here, the degree of importance indicates how much each input contributes to the output, and the larger the absolute value of the degree of importance, the higher the degree of influence the input has on the output.

As shown in FIG. 1, the monitoring device 10 includes a collection unit 11, an abnormality prediction unit 12, an output visualization unit 13, a calculation unit 14, an importance visualization unit 15, a process data buffer 16, an analysis frame buffer 17, and a frame abnormality evaluation unit. It has a value buffer 18 . Processing of each unit of the monitoring device 10 will be described below.

The collection unit 11, the abnormality prediction unit 12, the output visualization unit 13, the calculation unit 14, and the importance visualization unit 15 are, for example, CPU (Central Processing Unit), MPU (Micro Processing Unit), GPU (Graphical Processing Unit), and other electronic It is a circuit, an integrated circuit such as an ASIC (Application Specific Integrated Circuit) or an FPGA (Field Programmable Gate Array). The process data buffer 16, the analysis frame buffer 17, the frame abnormality evaluation value buffer 18, and the model buffer 19 are storage devices such as semiconductor memory devices such as RAM (Random Access Memory) and flash memory. .

The collection unit 11 collects a plurality of data acquired by the monitored equipment. For example, the collection unit 11 periodically (for example, every minute) receives data from sensors installed in facilities to be monitored such as factories and plants, and stores the data in the process data buffer 16 . Here, the data acquired by the sensor is, for example, various data such as temperature, pressure, sound, vibration, etc. about the equipment and reactors in the factory, which is the equipment to be monitored. Note that the time-series data acquired by the sensor is hereinafter referred to as process data as appropriate. The data collected by the collection unit 11 is not limited to the data acquired by the sensor, and may be, for example, numerical data input manually, label data such as the type and brand of material, and the like.

The abnormality prediction unit 12 receives a plurality of data collected by the collection unit 11, uses a learned model for predicting the state of the equipment to be monitored, and outputs a predetermined output value. For example, the abnormality prediction unit 12 uses process data and a learned model (identification function) to predict whether or not an abnormality will occur after a predetermined period of time. The abnormality prediction unit 12 has an analysis frame extraction unit 12a and a frame abnormality evaluation value calculation unit 12b. It should be noted that the abnormality prediction processing described below is an example, and the present invention is not limited to this.

The analysis frame extraction unit 12a extracts process data included in an analysis frame having a predetermined time width Wa from the process data collected by the collection unit 11 . Specifically, the analysis frame extraction unit 12a extracts and reads from the process data buffer 16 the process data for the frame of the time width Wa including the process data at a predetermined point in time, and stores the read process data in the analysis frame buffer 17. Store.

For example, when the collecting unit 11 collects process data every minute and the time width Wa is "5", the analysis frame extracting unit 12a extracts the process data acquired at the current time t, the current time 1 Process data acquired 2 minutes before the current time Process data acquired 3 minutes before the current time Process data acquired 4 minutes before the current time Extract and read from buffer 16 .

The process data buffer 16 stores process data collected by the collecting unit 11 . The process data buffer 16 stores at least the latest process data for frames of the time width Wa as process data. The analysis frame buffer 17 stores the process data extracted by the analysis frame extractor 12a.

The frame abnormality evaluation value calculation unit 12b receives as input the process data extracted by the analysis frame extraction unit 12a, uses a learned model for predicting an abnormality in the equipment to be monitored, and calculates the value after a predetermined time from a predetermined time. A frame abnormality evaluation value, which is an evaluation value representing the probability that an abnormality will occur, is calculated. The predetermined time may be the current time, or may be a preset time such as 10 seconds before the current time. Further, the frame abnormality evaluation value calculation unit 12b calculates the frame abnormality evaluation value using a neural network as a trained model. Note that the trained model is not limited to a neural network, and may be one to which machine learning algorithms such as logistic regression, support vector machine, and discriminant analysis are applied.

Specifically, the frame abnormality evaluation value calculator 12b receives the process data extracted from the analysis frame extractor 12a. Then, the frame abnormality evaluation value calculation unit 12b receives the received process data as an input and uses a learned model for predicting an abnormality in the monitored equipment to evaluate the frame abnormality at a point in time after a predetermined time from the current time t. A value is calculated and stored in the frame abnormality evaluation value buffer 18 . Note that the frame abnormality evaluation value may be calculated at time t instead of after a predetermined time. Further, the frame abnormality evaluation value is, for example, a probability value that an abnormality occurs in the equipment to be monitored, and may be a numerical value represented by "0" to "1". In this case, for example, if the probability of occurrence of an abnormality in the equipment to be monitored at a certain time is predicted to be "40%", the frame abnormality evaluation value will be "0.4". Further, the frame abnormality evaluation value is not limited to this. It may be a numerical value expressed by Moreover, when there is time-series data such as specific sensor values that serve as indicators of abnormality, it is also possible to regard this specific time-series data itself as the frame abnormality evaluation value.

The frame abnormality evaluation value buffer 18 stores the frame abnormality evaluation value calculated by the frame abnormality evaluation value calculation section 12b. The frame abnormality evaluation value buffer 18 stores at least the latest frame abnormality evaluation values for the frames of the time width Wb as the frame abnormality evaluation values. The model buffer 19 is a learned model used by the frame abnormality evaluation value calculation unit 12b described above, and stores a learned model for predicting an abnormality in the equipment to be monitored.

Here, a series of abnormality prediction evaluation value calculation processing executed by the abnormality prediction unit 12 will be described with reference to FIG. 2 . FIG. 2 is a diagram for explaining the abnormality prediction evaluation value calculation process executed by the abnormality prediction unit. (A) in FIG. 2 shows a plot of the process data acquired from the sensor for each item x _1t to x _nt , and an analysis frame having a time width Wa is generated each time an abnormality prediction evaluation value is calculated. It moves by the movement width L.

As illustrated in FIGS. 2A and 2B, for example, it is assumed that the collection unit 11 collects process data every minute and the time width Wa is "5". In this case, the frame abnormality evaluation value calculation unit 12b receives process data for a time width Wa of “5” as input, uses a learned model for predicting an abnormality in the equipment to be monitored, and calculates a predetermined value from the current time t. A frame anomaly evaluation value is calculated for the point in time after the hour.

Returning to the description of FIG. 1, the output visualization unit 13 outputs a warning sign based on the abnormality prediction evaluation value, and outputs time-series data of the abnormality prediction evaluation value as a chart screen. The output visualization unit 13 has an abnormality determination unit 13a and a chart display unit 13b.

The abnormality determination unit 13a determines whether or not the frame abnormality evaluation value calculated by the frame abnormality evaluation value calculation unit 12b is equal to or greater than a predetermined threshold. Outputs warnings about abnormal occurrences. For example, when the abnormality prediction evaluation value is equal to or greater than a predetermined threshold value, the abnormality determination unit 13a outputs a warning message indicating that an abnormality may occur in the equipment to be monitored after a certain period of time as a warning sign. Alternatively, a sound for notifying the warning may be output.

The chart display unit 13b displays the time-series data of the frame abnormality evaluation values calculated by the frame abnormality evaluation value calculation unit 12b as a chart screen. For example, when receiving a display request from the user, the chart display unit 13b displays the time-series data of the abnormality prediction evaluation values as a chart screen.

The calculation unit 14 uses each data at each time input to the learned model and the predetermined output value output from the learned model to calculate the importance of the predetermined output value for each input item at each time. calculate. The calculator 14 has an importance calculator 14a and an importance noise remover 14b.

The importance calculation unit 14a reads from the model buffer 19 the learned model used by the frame abnormality evaluation value calculation unit 12b. Then, the importance calculation unit 14a reads the process data input to the learned model from the analysis frame buffer 17, reads the frame abnormality evaluation value output from the learned model from the frame abnormality evaluation value buffer 18, and calculates the process data. , the frame abnormality evaluation value, and the learned model are used to calculate the degree of importance for a predetermined output value for each input item at each time. For example, as illustrated in FIG. 3, the importance calculator 14a reads the process data and the frame abnormality evaluation value every minute, and calculates the importance of a predetermined output value for each sensor at each time. FIG. 3 is a diagram illustrating an example of the degree of importance of each sensor calculated by the calculator.

In the example of FIG. 3, the importance calculation unit 14a calculates the importance of sensor A as "20", the importance of sensor B as "31", and the importance of sensor C as "17" at "time 12:00". , the importance of sensor D is calculated as "15", and the importance of sensor E is calculated as "24". Further, at “time 12:01”, the importance calculation unit 14a calculates the importance of sensor A as “21”, the importance of sensor B as “32”, the importance of sensor C as “16”, and the importance of sensor D as “21”. "13" is calculated as the importance, and "26" as the importance of the sensor E is calculated. That is, in the example of FIG. 3, at "time 12:00" and "time 12:01", the importance of sensor B is the highest, and sensor B process data contributed most as input values.

A specific example of calculating the degree of importance will now be described. For example, the importance calculation unit 14a calculates the importance of each sensor at each time using a partial differential value or an approximate value thereof with respect to each input value of the output value in a trained model that calculates an output value from an input value. calculate. As an example, the importance calculator 14a calculates the importance of each sensor at each time using the saliency map. The saliency map is a technique used in neural network image classification, and is a technique for extracting a partial differential value for each input of the neural network output as a degree of importance that contributes to the output.

Here, the importance degree calculation process using the saliency map executed by the calculation unit 14 will be described using the example of FIG. 4 . FIG. 4 is a diagram for explaining importance calculation processing using a saliency map executed by the calculation unit. As exemplified in FIG. 4, the calculation unit 14 has learned the process data for the frame of the time width (the time width from t to t+N in the example of FIG. 4) collected from M sensors as input I. When inputting to the model and obtaining the output S(I) from the trained model as an output value, ∂S(I)/∂I is calculated as the value of importance. Thereby, the calculation unit 14 can calculate the importance of each of the M sensors at each time.

It should be noted that the importance may be calculated by a method other than the saliency map, for example, the method of obtaining the importance of the input by propagating the output value to each layer of the neural network (see reference 1), or the above improved version , a method that considers nonlinearity and the sign of importance (see reference 2), or a method that approximates a neural network to a simple model (decision tree, etc.) to determine the importance of inputs to data ( See reference 3).
Reference 1: Bach Sebastian, et al. "On pixel-wise explanations for non-linear classifier decisions by layer-wise relevance propagation." PloS one 10.7 (2015): e0130140.
Reference 2: Shrikumar Avanti, Peyton Greenside, and Anshul Kundaje. "Learning important features through propagating activation differences." arXiv preprint arXiv:1704.02685 (2017).
Reference 3: Ribeiro Marco Tulio, Sameer Singh, and Carlos Guestrin. "Why should i trust you?: Explaining the predictions of any classifier." Proceedings of the 22nd ACM SIGKDD International Conference on Knowledge Discovery and Data Mining. ACM, 2016.

The importance noise removal unit 14b smoothes the importance calculated by the importance calculation unit 14a for a predetermined time width to remove noise. For example, the importance noise removal unit 14b may smooth the importance using a Smooth Grad technique as a technique for removing noise. Smooth Grad is a method of removing noise by creating dozens of duplicates by applying random Gaussian noise to an input image and averaging the degrees of importance extracted from each duplicate. Note that the process of removing the noise described above may be omitted, and in that case, the monitoring device 10 may not have the function of the importance level noise removing unit 14b.

Here, the noise removal processing executed by the calculation unit 14 will be described using the example of FIG. 5 . FIG. 5 is a diagram for explaining noise removal processing executed by the calculator. As exemplified in FIG. 5, the calculation unit 14 calculates the importance from the input values and the output values of the process data for frames of a predetermined time width collected from M sensors, and calculates the importance of each sensor. Smoothing is performed by averaging degrees in the time direction to remove noise.

In this manner, the calculation unit 14 applies the above-described Saliency Map and Smooth Grad to time-series data. Since it is known that time-series data contains Gaussian noise at each time, the same principle as Smooth Grad can be obtained simply by performing smoothing for averaging at each time.

The importance visualization unit 15 displays a graph showing changes in the importance of each sensor data, and notifies important sensors in synchronization with the determination result of the abnormality determination unit 13a. The importance visualization unit 15 has an acquisition unit 15a, a creation unit 15b, an importance display unit 15c, and a notification unit 15d.

The acquisition unit 15a acquires the degree of importance for a predetermined output value for each input item at each time input to the trained model. For example, the acquisition unit 15a acquires the importance of each sensor from which noise has been removed by the importance noise removal unit 14b.

The creation unit 15b creates a graph showing the transition of the importance of each input item using the importance acquired by the acquisition unit 15a. The importance display unit 15c displays the graph created by the creation unit 15b.

The notification unit 15d notifies an input item having a degree of importance equal to or higher than a predetermined threshold among the degrees of importance of the input items acquired by the acquisition unit 15a when an abnormality is detected in the equipment to be monitored. For example, the notification unit 15d acquires the determination result by the abnormality determination unit 13a, and when an abnormality in the monitored equipment is detected, that is, when the abnormality prediction evaluation value is equal to or greater than a predetermined threshold, an abnormality is detected. Of the importance of each sensor corresponding to the specified time, the sensor whose importance is equal to or higher than a predetermined threshold is notified. This makes it possible to notify which sensor was important when an abnormality occurred.

Here, it is a figure explaining the outline|summary of the abnormality prediction process performed by the monitoring apparatus 10 using FIG. FIG. 6 is a diagram explaining an outline of anomaly prediction processing executed by the monitoring device.

FIG. 6 illustrates that sensors and devices for collecting signals for operation are attached to reactors and devices in the plant, and data are collected at regular time intervals. FIG. 6 shows the transition of the process data collected from the sensors A to E by the collection unit 11 (see FIG. 6A), and the data in the portion surrounded by the frame is extracted, and the abnormality prediction unit 12 calculates an abnormality prediction evaluation value based on the process data, and estimates an abnormality after a certain period of time (see (B) in FIG. 6). Then, the output visualization unit 13 outputs the time-series data of the calculated abnormality prediction evaluation value as a chart screen (see (C) in FIG. 6).

Further, the calculation unit 14 uses the process data input to the trained model and the frame abnormality evaluation value output from the trained model to calculate the importance of a predetermined output value for each sensor at each time. (See (D) in FIG. 6). Then, the importance visualization unit 15 displays a graph showing changes in the importance of each sensor data (see (E) in FIG. 6).

[Processing procedure of the monitoring device]
Next, an example of a processing procedure by the monitoring device 10 according to the first embodiment will be described with reference to FIGS. 7 to 9. FIG. FIG. 7 is a flowchart showing an example of the flow of abnormality prediction processing in the monitoring device according to the first embodiment. FIG. 8 is a flowchart showing an example of the flow of importance calculation processing in the monitoring device according to the first embodiment. FIG. 9 is a flowchart showing an example of the flow of graph display processing in the monitoring device according to the first embodiment.

First, the flow of abnormality prediction processing by the monitoring device 10 will be described with reference to FIG. As illustrated in FIG. 7, when the collecting unit 11 acquires process data (Yes at step S101), the process data in the analysis frame having a predetermined time width Wa is extracted from the process data collected by the collecting unit 11. (Step S102).

Then, the frame abnormality evaluation value calculation unit 12b receives the process data extracted by the analysis frame extraction unit 12a as an input, uses a learned model for predicting an abnormality in the equipment to be monitored, and calculates the value after a predetermined time from a predetermined point in time. , the frame abnormality evaluation value is calculated (step S103).

After that, the abnormality determination unit 13a of the output visualization unit 13 determines whether or not the frame abnormality evaluation value calculated by the frame abnormality evaluation value calculation unit 12b is equal to or greater than a predetermined threshold (step S104). As a result, when the frame abnormality evaluation value is less than the predetermined threshold value (No at step S104), the abnormality determination unit 13a terminates the process. Further, when the frame abnormality evaluation value is equal to or greater than the predetermined threshold value (Yes at step S104), the abnormality determination unit 13a outputs a warning sign (step S105).

Next, the flow of importance calculation processing by the monitoring device 10 will be described with reference to FIG. As illustrated in FIG. 8, the importance calculation unit 14a obtains the process data input to the trained model from the analysis frame buffer 17, and converts the frame abnormality evaluation value output from the trained model into a frame abnormality evaluation value. When it is obtained from the buffer 18 (Yes at step S201), the process data and the frame abnormality evaluation value are used to calculate the degree of importance for a predetermined output value for each sensor at each time (step S202). For example, the importance calculator 14a reads the process data and the frame abnormality evaluation value every minute, and calculates the importance of a predetermined output value for each sensor at each time.

Then, the importance noise removal unit 14b smoothes the importance calculated by the importance calculation unit 14a for a predetermined time width to remove noise (step S203). For example, the importance noise removal unit 14b smoothes the importance using a Smooth Grad technique as a technique for removing noise.

Next, the flow of importance calculation processing by the monitoring device 10 will be described with reference to FIG. As illustrated in FIG. 9, when the obtaining unit 15a obtains the degree of importance for a predetermined output value for each sensor at each time input to the trained model (Yes in step S301), the creating unit 15b obtains the obtaining unit 15a A graph showing the transition of the importance of each sensor is created using the importance obtained by (step S302). Then, the importance display unit 15c displays the graph created by the creation unit 15b (step S303).

[Effects of the first embodiment]
The monitoring device 10 according to the first embodiment collects a plurality of sensor data acquired by the monitoring target equipment, uses the collected sensor data as an input, and has learned to predict an abnormality in the monitoring target equipment. Using the model, the frame abnormality evaluation value, which is the degree of abnormality of the equipment to be monitored, is output as an output value. In addition, the monitoring device 10 uses each sensor data input to the learned model at each time and the degree of anomaly output from the learned model to calculate the importance of the output value for each input item at each time. do. Therefore, in the monitoring device 10, it is possible to dynamically and easily confirm which input value is important among the plurality of input values.

In addition, the monitoring device 10 uses the saliency map to calculate the importance of each input item at each time. Therefore, the monitoring device 10 can easily calculate which input value is important using the saliency map.

In other words, in the monitoring apparatus 10 according to the first embodiment, it is possible to extract and present important inputs for output along with their importance at each time for each data input. As a result, it became possible to present important sensors for the occasional state, and it became possible to deal with a system whose state changes. In addition, even if you are not an expert who is familiar with the state, it is easy to specify which item to operate for that state. In addition, as a model that detects anomalies and predicts values, it can be applied to neural networks, where it was difficult to extract important features in the past. Become.

With the monitoring device 10 according to the first embodiment, sensor prediction and abnormality detection are performed by passing the input values of sensors and the like through a learned model, and which input values are important for the output in that state. can be dynamically checked. As a result, it becomes easier to determine which input value should be manipulated according to the output value. Also, even when a neural network is used as a model, important sensors can be identified.

In the above description, the case where the monitoring device 10 outputs the degree of abnormality as an output value using a learned model for predicting an abnormality in the equipment to be monitored has been described, but it is not limited to this. do not have. For example, a predicted sensor value may be output as an output value using a learned model for predicting the sensor value of the data acquired by the sensor. In this case, the monitoring device collects a plurality of sensor data acquired by the sensors installed in the monitoring target equipment, uses the collected sensor data as input, and predicts the sensor value of the data acquired by the sensor. A predicted sensor value is output as an output value using a trained model for In addition, the monitoring device calculates the importance of the output value for each input item at each time using each sensor data input to the trained model at each time and the predicted sensor value output from the trained model. do.

Here, an overview of the sensor value prediction processing will be described with reference to FIG. 10 . FIG. 10 is a diagram explaining an outline of the sensor value prediction process. FIG. 10 illustrates that sensors and devices for collecting signals for operation are attached to the reactors and devices in the plant, and data are collected at regular time intervals. FIG. 10 shows changes in process data collected from sensors A to E (see FIG. 10A). is predicted (see (B) of FIG. 10). For example, the monitoring device predicts the sensor value N minutes after the current time t. Then, the monitoring device outputs the predicted time-series data of the sensor values as a chart screen (see (C) in FIG. 10).

In addition, the monitoring device uses the sensor data input to the learned model to calculate the degree of importance for a predetermined output value for each sensor at each time (see (D) in FIG. 10). Then, the monitoring device displays a graph showing changes in the degree of importance of each sensor data (see (E) in FIG. 10).

In this way, when the monitoring device 10 uses the learned model to predict the sensor value of the data acquired by the sensor and outputs the predicted sensor value as an output value, a plurality of inputs to the learned model It is possible to dynamically and easily ascertain which input values are important among the values.

[Second embodiment]
In the above-described first embodiment, the case where the monitoring device 10 has the importance visualization unit 15 and displays a graph showing changes in the importance of each sensor data has been described, but the present invention is not limited to this. . For example, the display device 20, which is a device different from the monitoring device 10, may display a graph showing changes in the degree of importance of each sensor data.

Therefore, in the following description, the monitoring system 100 according to the second embodiment has a monitoring device 10A and a display device 20, and the display device 20 displays a graph showing changes in the degree of importance of each sensor data. A case will be described where an important sensor is notified in synchronization with the determination result of the determination unit 13a. Note that description of the same configuration and processing as those of the monitoring apparatus 10 according to the first embodiment will be omitted.

First, a monitoring system 100 according to the second embodiment will be described with reference to FIG. 11 . FIG. 11 is a block diagram showing a configuration example of a monitoring system according to the second embodiment. As shown in FIG. 11, the monitoring system 100 has a monitoring device 10A and a display device 20. As shown in FIG. Also, in the monitoring system 100, the monitoring device 10A and the display device 20 are connected via a network (not shown).

Further, as shown in FIG. 11, the monitoring device 10A according to the second embodiment has an importance level visualization unit 15 compared to the monitoring device 10 according to the first embodiment shown in FIG. The difference is that there is no

The display device 20 has an acquisition unit 21 , a creation unit 22 , an importance display unit 23 and a notification unit 24 . The display device 20 is a device that displays important sensors so that an observer can check them, and is, for example, a terminal such as a PC (Personal Computer) or a mobile terminal such as a smart phone or a tablet terminal. Note that the display device 20 may have the function of the output visualization unit 13 instead of the monitoring device 10 .

The acquisition unit 21 inputs a plurality of data acquired by the monitoring target equipment, uses a learned model for predicting the state of the monitoring target equipment, and outputs a predetermined output value from the external monitoring device 10A, Obtain the degree of importance for a predetermined output value for each input item at each time input to the trained model. For example, the acquisition unit 21 acquires the importance of each sensor from which noise has been removed by the importance noise removal unit 14b.

The creation unit 22 creates a graph showing the transition of the importance of each input item using the importance acquired by the acquisition unit 21 . The importance display section 23 displays the graph created by the creation section 22 .

The notification unit 24 notifies an input item having a degree of importance equal to or higher than a predetermined threshold among the degrees of importance of the input items acquired by the acquisition unit 21 when an abnormality in the equipment to be monitored is detected. For example, the notification unit 24 acquires the determination result by the abnormality determination unit 13a, and when an abnormality in the monitored equipment is detected, that is, when the abnormality prediction evaluation value is equal to or greater than a predetermined threshold, an abnormality is detected. Of the importance of each sensor corresponding to the specified time, the sensor whose importance is equal to or higher than a predetermined threshold is notified.

[Effect of Second Embodiment]
The display device 20 according to the second embodiment receives a plurality of data acquired by the monitored equipment, uses a learned model for predicting the state of the monitored equipment, and outputs a predetermined output value. From the external monitoring device 10A, the degree of importance for a predetermined output value for each input item at each time input to the trained model is acquired. Then, the display device 20 creates a graph showing changes in the importance of each input item using the acquired importance, and displays the created graph. Therefore, with the display device 20, it is possible to dynamically and easily confirm which input value is important among the plurality of input values.

In addition, when an abnormality is detected in the equipment to be monitored, the display device 20 notifies the input items whose importance is equal to or higher than a predetermined threshold among the obtained degrees of importance of the respective input items. Therefore, the display device 20 can notify which sensor was important when an abnormality occurs.

[System configuration, etc.]
Also, each component of each device illustrated is functionally conceptual, and does not necessarily need to be physically configured as illustrated. In other words, the specific form of distribution and integration of each device is not limited to the one shown in the figure, and all or part of them can be functionally or physically distributed and integrated in arbitrary units according to various loads and usage conditions. Can be integrated and configured. Furthermore, all or any part of each processing function performed by each device is realized by a CPU or GPU and a program analyzed and executed by the CPU or GPU, or as hardware by wired logic can be realized.

In addition, among the processes described in the present embodiment, all or part of the processes described as being performed automatically can be performed manually, or the processes described as being performed manually can be performed manually. can also be performed automatically by known methods. In addition, information including processing procedures, control procedures, specific names, and various data and parameters shown in the above documents and drawings can be arbitrarily changed unless otherwise specified.

[program]
It is also possible to create a program in which the processes executed by the monitoring device and the display device described in the above embodiments are described in a computer-executable language. For example, it is also possible to create a monitoring program and a display program in which the processing executed by the monitoring device 10, 10A or the display device 20 according to the embodiment is described in a computer-executable language. In this case, the computer executes the monitoring program and the display program to obtain the same effect as the above embodiment. Further, by recording the monitoring program and the display program in a computer-readable recording medium and causing the computer to read and execute the monitoring program and the display program recorded in the recording medium, the same processing as in the above embodiment can be performed. may be realized.

FIG. 12 is a diagram showing a computer executing a monitoring program or a display program. As illustrated in FIG. 12, computer 1000 includes, for example, memory 1010, CPU 1020, hard disk drive interface 1030, disk drive interface 1040, serial port interface 1050, video adapter 1060, and network interface 1070. , and these units are connected by a bus 1080 .

The memory 1010 includes a ROM (Read Only Memory) 1011 and a RAM 1012 as illustrated in FIG. The ROM 1011 stores a boot program such as BIOS (Basic Input Output System). Hard disk drive interface 1030 is connected to hard disk drive 1090 as illustrated in FIG. Disk drive interface 1040 is connected to disk drive 1100 as illustrated in FIG. A removable storage medium such as a magnetic disk or optical disk is inserted into the disk drive 1100 . The serial port interface 1050 is connected to, for example, a mouse 1110 and a keyboard 1120, as illustrated in FIG. Video adapter 1060 is connected to display 1130, for example, as illustrated in FIG.

Here, as illustrated in FIG. 12, the hard disk drive 1090 stores an OS 1091, application programs 1092, program modules 1093, and program data 1094, for example. That is, the monitoring program or display program described above is stored, for example, in the hard disk drive 1090 as a program module in which instructions to be executed by the computer 1000 are described.

Various data described in the above embodiments are stored as program data in the memory 1010 or the hard disk drive 1090, for example. Then, the CPU 1020 reads the program modules 1093 and program data 1094 stored in the memory 1010 and the hard disk drive 1090 to the RAM 1012 as necessary, and executes various processing procedures.

Note that the program module 1093 and program data 1094 related to the monitoring program or display program are not limited to being stored in the hard disk drive 1090. For example, they are stored in a removable storage medium and read by the CPU 1020 via a disk drive or the like. may be Alternatively, the program module 1093 and program data 1094 related to the monitoring program or display program are stored in another computer connected via a network (LAN (Local Area Network), WAN (Wide Area Network), etc.) and are stored in a network interface. It may be read by CPU 1020 via 1070 .

The above-described embodiments and modifications thereof are included in the scope of the invention described in the claims and their equivalents, as well as in the technology disclosed in the present application.

10, 10A monitoring device 11 collection unit 12 abnormality prediction unit 12a analysis frame extraction unit 12b frame abnormality evaluation value calculation unit 13 output visualization unit 13a abnormality determination unit 13b chart display unit 14 calculation unit 14a importance calculation unit 14b importance noise removal unit 15 importance visualization unit 15a, 21 acquisition unit 15b, 22 creation unit 15c, 23 importance display unit 15d, 24 notification unit 16 process data buffer 17 analysis frame buffer 18 frame abnormality evaluation value buffer 20 display device 100 monitoring system

Claims (3)

prediction means for outputting a predetermined output value using a trained model, which is a neural network for predicting the state of the monitored equipment, with a plurality of data acquired by the monitored equipment as inputs;
The degree of importance for the predetermined output value for each input item at each time, using each data at each time input to the learned model and a predetermined output value output from the learned model. and calculating means for calculating the degree of importance at each time that changes with time indicating how much each of the data has contributed to the output value of the trained model,
The calculation means calculates the importance level for each input item at each time using a partial differential value or an approximate value thereof with respect to each input value of the output value in the learned model that calculates the output value from the input value A monitoring device characterized by calculating A monitoring method performed by a monitoring device, comprising:
a prediction step of outputting a predetermined output value using a trained model, which is a neural network for predicting the state of the monitored equipment, with a plurality of data acquired by the monitored equipment as input;
The degree of importance for the predetermined output value for each input item at each time, using each data at each time input to the learned model and a predetermined output value output from the learned model. a calculation step of calculating a time-varying degree of importance at each time indicating how much each of the data contributed to the output value of the trained model,
The calculating step, in the learned model that calculates the output value from the input value, uses a partial differential value or an approximate value thereof with respect to each input value of the output value, and calculates the importance level for each input item at each time A monitoring method characterized by calculating a prediction step of outputting a predetermined output value using a trained model, which is a neural network for predicting the state of the monitored equipment, with a plurality of data acquired by the monitored equipment as inputs;
The degree of importance for the predetermined output value for each input item at each time, using each data at each time input to the learned model and a predetermined output value output from the learned model. and causing a computer to perform a calculation step of calculating a time-varying degree of importance at each time indicating how much each data contributed to the output value of the trained model,
The calculating step, in the learned model that calculates an output value from an input value, uses a partial derivative value or an approximate value thereof with respect to each input value of the output value, and calculates the importance level for each input item at each time. A monitoring program characterized by calculating

Images