JP2019179400A - Monitoring device, monitoring method and monitoring program - Google Patents

Abstract

To provide a device allowing a person to dynamically and easily confirm the significance of each input data to output data for deciding and controlling the item related to a sensor to be controlled in order to prevent abnormality.SOLUTION: A monitoring device 10 includes: collection means 11 for collecting a plurality of pieces of data about a monitoring object; detection means 12 for detecting a state of the monitoring object with the plurality of pieces of data collected by the collection means as input data by using a self-encoder; and calculation means 14 for calculating the significance of each input data inputted to the self-encoder to output data by using each input data inputted to the self-encoder and the output data outputted from the self-encoder.SELECTED DRAWING: Figure 1

Description

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

  A self-encoder, which is one type of neural network, is a model that learns to reproduce input data as teacher data. This self-encoder is characterized in that the number of nodes in the input layer and the output layer is the same, and the number of nodes in the intermediate layer is smaller than that in the input / output layer. There are several application methods of the self-encoder. For example, data features may be extracted from the output of the intermediate layer of the model after learning as in principal component analysis. Also, as a factory abnormality detection, it is used to detect an abnormality by comparing the difference between input data and output data with respect to a model in which normal sensor data and images are learned.

  As an application example of abnormality detection, first, a model in which only normal data such as plant sensors is learned is created. Since this model can reproduce data for learned data, that is, normal data, the reproduction error is reduced. On the other hand, since abnormal data that has not been learned cannot be reproduced, a reproduction error becomes large. Thus, abnormality detection can be performed by looking at the reproduction error.

  Further, clustering or the like exists as a method for performing abnormality detection or state estimation without using a neural network. In these methods, an abnormality is extracted from the data distance between normal and abnormal.

JP 2017-142654 A

  The conventional method as described above has a problem that the importance of each input data with respect to the output data cannot be confirmed dynamically and easily. For example, in the method of extracting an abnormality from the data distance between normal and abnormal as described above, an important feature can be obtained indirectly from the data distance between normal and abnormal, but the relationship between features, etc. It is difficult to obtain detailed features that also take into account. In addition, as the deep learning technology is called a black box, it was difficult to extract how the model predicted and judged. The self-encoder is no exception, and it is difficult to confirm which input has affected the output.

  For example, in the case of abnormality detection, it is difficult to specify which input has an influence on abnormal data and causes a reproduction error. Therefore, in order to prevent abnormality, an item relating to which sensor may be controlled. I couldn't connect to such actions.

  In order to solve the above-described problems and achieve the object, the monitoring apparatus of the present invention includes a collection unit that collects a plurality of data related to a monitoring target, and a plurality of data collected by the collection unit as input data. Using an encoder, detection means for detecting the state of the monitoring target, each input data input to the self-encoder, and output data output from the self-encoder, And calculating means for calculating the importance of each input data input to the self-encoder for the output data.

  Further, the monitoring method of the present invention is a monitoring method executed by a monitoring device, which collects a plurality of data relating to a monitoring target, and uses a plurality of data collected by the collecting step as input data. Using an encoder, a detection step of detecting the state of the monitoring target, each input data input to the self-encoder, and output data output from the self-encoder, A calculation step of calculating the importance of each input data input to the self-encoder with respect to the output data.

  The monitoring program of the present invention includes a collection step for collecting a plurality of data acquired at a monitoring target facility, a collection step for collecting a plurality of data related to the monitoring target, and a plurality of data collected by the collection step. As the input data, using a self-encoder, a detection step of detecting the state of the monitoring target, each input data input to the self-encoder, and output data output from the self-encoder And calculating a degree of importance of each piece of input data input to the self-encoder with respect to output data.

  According to the present invention, it is possible to dynamically and easily confirm the importance of each input data with respect to output data.

FIG. 1 is a block diagram illustrating a configuration example of a monitoring apparatus according to the first embodiment. FIG. 2 is a diagram for explaining the learning of the self-encoder. FIG. 3 is a diagram for explaining processing for determining abnormality from the reproduction error of input data and output data. FIG. 4 is a diagram for explaining processing for determining abnormality from reproduction errors of input data and output data. FIG. 5 is a diagram for explaining an example of importance calculation processing for the self-encoder. FIG. 6 is a diagram for explaining an example in the case of calculating N × N importance. FIG. 7 is a diagram illustrating an example of processing for narrowing down the importance calculation target according to the reproduction error. FIG. 8 is a diagram for explaining an example of processing for narrowing down intermediate layer nodes as importance calculation targets in a neural network. FIG. 9 is a diagram for explaining an overview of the entire processing executed by the monitoring apparatus. FIG. 10 is a flowchart illustrating an example of a flow of abnormality prediction processing in the monitoring device according to the first embodiment. FIG. 11 is a flowchart illustrating an example of the flow of importance calculation processing in the monitoring apparatus according to the first embodiment. FIG. 12 is a flowchart illustrating an example of the flow of graph display processing in the monitoring apparatus according to the first embodiment. FIG. 13 is a diagram illustrating a computer that executes a monitoring program.

  Hereinafter, embodiments of a monitoring device, a monitoring method, and a monitoring program according to the present application will be described in detail with reference to the drawings. Note that the monitoring device, the monitoring method, and the monitoring 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 flow of processing of the monitoring device 10 will be described in order, and finally the effects of the first embodiment will be described. In the following, an example will be described in which the monitoring device 10 collects data from monitored equipment such as a factory or a plant, and calculates the degree of abnormality of the monitored equipment using a self-encoder.

[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 illustrating a configuration example of a monitoring device according to the first embodiment. For example, the monitoring device 10 collects a plurality of data acquired by sensors installed in a monitoring target facility such as a factory or a plant, and uses the collected plurality of data as input data using a self-encoder. Output data reproducing the input data, comparing the input data with the output data, detecting a reproduction error of the output data with respect to the input data, and depending on the reproduction error, the degree of abnormality of the monitored equipment Is calculated.

  In addition, the monitoring device 10 uses each input data input to the self-encoder and output data output from the self-encoder to output data of each input data input to the self-encoder. Calculate importance. The importance here indicates how much each input contributed to the output, in other words, how important each input was in reproducing the output, and the absolute value of the importance. The larger the value, the higher the influence of the input on the output.

  As shown in FIG. 1, this monitoring apparatus 10 includes a collection unit 11, a detection unit 12, an output visualization unit 13, a calculation unit 14, an importance level visualization unit 15, a process data buffer 16, an analysis frame buffer 17, a frame abnormality evaluation value. It has a buffer 18 and a model buffer 19. Hereinafter, processing of each unit included in the monitoring device 10 will be described.

  The collection unit 11, the detection unit 12, the output visualization unit 13, the calculation unit 14, and the importance visualization unit 15 are electronic circuits such as a CPU (Central Processing Unit), an MPU (Micro Processing Unit), and a GPU (Graphical Processing Unit), for example. And an integrated circuit such as an application specific integrated circuit (ASIC) or a field programmable gate array (FPGA). 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 elements such as a RAM (Random Access Memory) and a flash memory (Flash Memory), for example. .

  The collection unit 11 collects a plurality of data related to the monitoring target acquired by the monitoring target facility. For example, the collection unit 11 periodically receives data from a sensor installed in a monitoring target facility such as a factory or a plant (for example, every minute) 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, and the like regarding a factory, a device in the plant, and a reactor that are monitoring target facilities. 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 numerical data input manually, label data such as material types and brands, and the like.

  The detection unit 12 detects the state of the monitoring target facility using the self-encoder with the plurality of data collected by the collection unit 11 as input data. Specifically, the detection unit 12 outputs output data that reproduces each input data, compares the input data with the output data, and detects a reproduction error of the output data with respect to the input data. For example, the detection unit 12 detects the reproduction error of the output data with respect to the input data input to the neural network of the self-encoder using the process data and the self-encoder, and according to the reproduction error, Calculate the degree of abnormality of the monitored equipment. In the following, a description will be given by taking as an example a case where the reproduction error is detected using a self-encoder and the state of the monitored equipment is detected by calculating the degree of abnormality of the monitored equipment according to the reproduction error. However, the present invention is not limited to this, and the state of the monitoring target equipment may be detected by using a self-encoder. For example, the state of the monitoring target facility may be detected by visualizing the intermediate layer after learning with the self-encoder. The detection unit 12 includes an analysis frame extraction unit 12a and a frame abnormality evaluation value calculation unit 12b. In addition, the abnormality degree calculation process demonstrated below is an example, and is not limited to this.

  The analysis frame extraction unit 12a extracts process data included in an analysis frame having a predetermined time point or a predetermined time width from the process data collected by the collection unit 11. Specifically, the analysis frame extraction unit 12 a extracts and reads out the process data for the frame from the process data buffer 16 and stores the read process data in the analysis frame buffer 17. The predetermined time point may be the current time point or a preset time point such as 10 seconds before the current time point.

  For example, in the analysis frame extraction unit 12a, the collection unit 11 collects process data every minute, extracts the process data acquired at the current time t from the process data buffer 16, and reads it. Further, for example, when the time width is “5”, the analysis frame extraction unit 12a, the process data acquired at the current time t, the process data acquired one minute before the current time, the current time 2 The process data acquired minutes before, the process data acquired three minutes before the current time, and the process data acquired four minutes before the current time are extracted from the process data buffer 16 and read out.

  The process data buffer 16 stores process data collected by the collection unit 11. The process data buffer 16 stores at least the latest process data for a frame at a predetermined time point or a predetermined time width as process data. The analysis frame buffer 17 stores the process data extracted by the analysis frame extraction unit 12a.

  The frame abnormality evaluation value calculation unit 12b receives the process data extracted by the analysis frame extraction unit 12a as input, and uses the self encoder to output data for input data input to the neural network of the self encoder. The reproduction error is detected, and the degree of abnormality of the monitored equipment is calculated according to the reproduction error.

  For example, the frame abnormality evaluation value calculation unit 12b receives data of the sensor A, sensor B, sensor C, sensor D, and sensor E as process data at time t extracted from the analysis frame extraction unit 12a. Then, the frame abnormality evaluation value calculation unit 12b inputs the data of the sensor A, sensor B, sensor C, sensor D, and sensor E at the time t to the self-encoder, and reproduces the reproduction error between the input data and the output data. The frame abnormality evaluation value that is the degree of abnormality of the monitoring target equipment at time t is calculated according to the reproduction error.

  Further, the frame abnormality evaluation value is, for example, a probability value that abnormality of the monitoring target facility has occurred, and may be a numerical value expressed by “0” to “1”. In this case, for example, when the probability that abnormality of the monitoring target facility has occurred at a certain time point is predicted to be “40%”, the frame abnormality evaluation value is “0.4”. Further, the frame abnormality evaluation value is not limited to this. For example, “0” or “1” is used as a value indicating whether or not there is a possibility that the abnormality of the monitoring target facility has occurred over a certain level. The numerical value expressed by either of these may be sufficient.

  The frame abnormality evaluation value buffer 18 stores the output data output from the self-encoder, the output data reproduction error with respect to the input data detected by the frame abnormality evaluation value calculation unit 12b, and the calculated frame abnormality evaluation value. To do. The frame abnormality evaluation value buffer 18 stores at least the latest frame abnormality evaluation value for a frame at a predetermined time point or a predetermined time width as a frame abnormality evaluation value. The model buffer 19 stores a learned model for detecting an abnormality of the self-encoder used by the frame abnormality evaluation value calculation unit 12b, that is, the monitoring target equipment.

  Here, it is a figure explaining the learning of a self-encoder using FIG. As illustrated in FIG. 2, in the neural network of the self-encoder, the number of nodes in the input layer and the output layer is the same, and the number of intermediate nodes is smaller than that in the input / output layer. When input data is input to the self-encoder, the input data is learned to be reproduced as teacher data. Here, the data to be learned by the self-encoder is only the data acquired by the sensor when the monitoring target equipment is normal (hereinafter referred to as “normal data” as appropriate). That is, by learning only normal data as input data, it is learned so that it can be reproduced well only when the input data is normal data, and cannot be reproduced well when the input data is abnormal data.

  In addition, by performing learning using only normal data in this way, normal data is overwhelming compared to abnormal data because monitored equipment is unlikely to generate abnormalities at normal times, such as devices in factories and plants. Even when there are many abnormal data, it is possible to learn appropriately. Although the monitoring device 10 will be described as an example in which a learned self-encoder is stored in the model buffer 19 in advance, the monitoring device 10 may perform a learning process.

  Next, processing for determining an abnormality from the reproduction error of input data and output data will be described with reference to FIG. FIG. 3 and FIG. 4 are diagrams for explaining processing for determining abnormality from reproduction errors of input data and output data. In the example of FIG. 3, normal data is input as input data, and the example of FIG. 4 illustrates a case where abnormal data is input as input data. As illustrated in FIG. 3, for example, the detection unit 12 inputs normal data as input data to the self-encoder and outputs output data. Since the self-encoder learns so that only normal data can be reproduced, the reproduction error between the input data and the output data is small.

  On the other hand, as illustrated in FIG. 4, for example, the detection unit 12 inputs abnormal data as input data to the self-encoder and outputs output data. Since the self-encoder learns so that only normal data can be reproduced, the reproduction error between the input data and the output data is large.

  That is, the detection unit 12 detects a reproduction error of the output data with respect to the input data, and calculates so that the degree of abnormality of the monitoring target facility increases as the reproduction error increases, and the monitoring target decreases as the reproduction error decreases. Calculate so that the degree of abnormality of the equipment is reduced.

  Returning to the description of FIG. 1, the output visualization unit 13 outputs a warning sign based on the abnormality prediction evaluation value or outputs time series data of the abnormality prediction evaluation value as a chart screen. The output visualization unit 13 includes 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. If the abnormality prediction evaluation value is equal to or greater than the predetermined threshold, Output a warning about abnormal occurrence. For example, the abnormality determination unit 13a outputs a warning message indicating that an abnormality may occur in the monitored equipment after a certain time as a warning sign when the abnormality prediction evaluation value is equal to or greater than a predetermined threshold. Alternatively, a sound for notifying a warning may be output.

  The chart display unit 13b displays the time series data of the frame abnormality evaluation value 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 time series data of the abnormality prediction evaluation value as a chart screen.

  The calculation unit 14 uses the input data input to the self-encoder and the output data output from the self-encoder to determine the importance of the input data input to the self-encoder to the output data. Is calculated. For example, the calculation unit 14 calculates the importance of each input data with respect to the output data by using an importance feature method such as Saliency Map. The calculation unit 14 includes an importance level calculation unit 14a and an importance level noise removal unit 14b.

  The importance calculation unit 14a reads the input data (process data) input to the self-encoder from the analysis frame buffer 17, and outputs the output data output from the self-encoder and the reproduction error of the output data with respect to the input data. And the frame anomaly evaluation value are read from the frame anomaly evaluation value buffer 18, and each input data input to the self-encoder and the output data output from the self-encoder are used for the self-encoder. Calculate the importance of each input data to the output data.

  Here, a specific example of calculating the importance will be described. For example, in the learned model for calculating the output value from the input value, the importance level calculation unit 14a calculates the importance level for each sensor at each time using a partial differential value or an approximate value of each input value of the output value. calculate. As an example, the importance level calculation unit 14a calculates the importance level for each sensor at each time by using the Saliency Map. The Saliency Map is a technique used in image classification of a neural network, and is a technique for extracting a partial differential value regarding each input of the output of the neural network as an importance that contributes to the output.

The importance may be calculated by a method other than Saliency Map. For example, a method for obtaining the importance of an input by propagating an output value to each layer of the neural network (see Reference 1) or a simple neural network. It may be a technique for obtaining the importance of input to data by approximating a simple model (such as a decision tree) (see Reference 2).
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).

  Here, an example of importance calculation processing for the self-encoder will be described using the example of FIG. FIG. 5 is a diagram for explaining an example of importance calculation processing for the self-encoder. As illustrated in FIG. 5, when the input data from x1 to xn is input to the self-encoder, the importance calculation unit 14a outputs output data from y1 to yn. Here, paying attention to y2 in the output data, the importance calculation unit 14a calculates the importance of each of the input data x1 to xn important for reproduction of y2.

  Here, the importance calculation unit 14a may calculate the importance of each of the input data x1 to xn for all output data. In this case, as illustrated in FIG. 6, when N pieces of input data are input, N pieces of output data are obtained, and therefore N × N importance is calculated. FIG. 6 is a diagram for explaining an example in the case of calculating N × N importance. For this reason, under a situation where the input dimension is large, the processing load of the monitoring device 10 is heavy, and there are cases where the number is too large for humans to perform analysis and control determination.

  Therefore, the importance calculation unit 14a extracts output data in which the reproduction error detected by the detection unit 12 is equal to or greater than a predetermined threshold, and uses each input data input to the self-encoder and the extracted output data. Thus, the importance of each input data with respect to the extracted output data may be calculated. For example, using the example of FIG. 5 described above, the importance calculation unit 14a is configured to reproduce the reproduction error between the input data x1 and the output data y1, the reproduction error between the input data x2 and the output data y2,. It is determined whether or not the reproduction error between xn and the output data yn is greater than or equal to a predetermined threshold value. As a result, the importance calculation unit 14a extracts only the output data for which the reproduction error is determined to be equal to or greater than a predetermined threshold, and calculates the importance of each input data with respect to the extracted output data.

  The process of narrowing down the importance calculation target according to the reproduction error will be described using the example of FIG. FIG. 7 is a diagram illustrating an example of processing for narrowing down the importance calculation target according to the reproduction error. As illustrated in FIG. 7, when N pieces of input data are input, the number of output data is also N. However, among the output data, only output data with a large reproduction error is subjected to importance calculation. When three pieces of output data are subjected to importance calculation, 3 × N importance levels are calculated. Compared to calculating N × N importance levels, the processing of the monitoring device 10 is performed. It reduces the load and makes it easier for humans to make analysis and control decisions.

  The importance calculation unit 14a may calculate the importance of each input data for each intermediate layer in the neural network. For example, the importance calculation unit 14a uses each input data input to the self-encoder and output data of a node in the intermediate layer in the neural network of the self-encoder to input each input to the output data of the intermediate layer. Calculate the importance of the data.

  An example of processing for narrowing down intermediate layer nodes in the neural network as importance calculation targets will be described using the example of FIG. FIG. 8 is a diagram for explaining an example of processing for narrowing down intermediate layer nodes as importance calculation targets in a neural network. As illustrated in FIG. 8, when N pieces of input data are input, the number of output data is N, but the intermediate layer node having a smaller number of nodes than the input / output layer is the target of importance calculation. As a result, when the output data of the nodes in the three middle tiers are subject to importance calculation, 3 × N importance levels are calculated, compared with calculating N × N importance levels. The processing load on the monitoring device 10 is reduced, and it is easy for humans to perform analysis and control determination.

  The importance noise removing unit 14b removes noise by smoothing the importance calculated by the importance calculating unit 14a with a predetermined time width. For example, the importance level noise removal unit 14b may smooth the importance level by using a method of Smooth Grad as a technique for removing noise. Smooth Grad is a method for removing noise by creating dozens of duplicates obtained by applying random Gaussian noise to an input image and averaging the importance extracted from each of them. In addition, about the process which removes the above-mentioned noise, you may abbreviate | omit and the monitoring apparatus 10 does not need to have the function part of the importance noise removal part 14b in that case.

  The importance level visualization unit 15 displays a graph indicating the transition of the importance level of each sensor data, or notifies an important sensor in synchronization with the determination result of the abnormality determination unit 13a. The importance level visualization unit 15 includes an acquisition unit 15a, a creation unit 15b, an importance level display unit 15c, and a notification unit 15d. Note that each function of the importance level visualization unit 15 may be provided in a display device different from the monitoring device 10.

  The acquisition unit 15a acquires the importance of the input data input to the self encoder with respect to the output data. 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 indicating the transition of the importance of each input item using the importance acquired by the acquisition unit 15a. The importance level display unit 15c displays the graph created by the creation unit 15b.

  When the abnormality of the monitoring target facility is detected, the notification unit 15d notifies an input item having an importance level equal to or higher than a predetermined threshold among the respective importance levels acquired by the acquisition unit 15a. For example, the notification unit 15d acquires the determination result by the abnormality determination unit 13a, and when an abnormality of the monitoring target facility is detected, that is, when the abnormality prediction evaluation value is equal to or greater than a predetermined threshold, the abnormality is detected. Of the importance levels of the sensors corresponding to the designated time, a sensor having an importance level equal to or higher than a predetermined threshold is notified. Thereby, when abnormality occurs, it is possible to notify the relationship between sensors having particularly high importance.

  Here, it is a figure explaining the outline | summary of the abnormality prediction process performed by the monitoring apparatus 10 using FIG. FIG. 9 is a diagram for explaining an outline of the abnormality prediction process executed by the monitoring apparatus.

  FIG. 9 illustrates that a sensor or a device for collecting an operation signal is attached to a reaction furnace or apparatus in a plant, and data is collected at regular intervals. FIG. 9 shows the transition of the process data collected from the sensors A to E by the collection unit 11 (see FIG. 9A). The detection unit 12 Using the self-encoder, it detects the output data reproduction error with respect to the input data input to the self-encoder neural network, and calculates the abnormal prediction evaluation value of the monitored equipment according to the reproduction error. (Refer to FIG. 9B). And the output visualization part 13 outputs the time series data of the calculated abnormality prediction evaluation value as a chart screen (see FIG. 9C).

  Further, the calculation unit 14 uses each input data input to the self-encoder and output data output from the self-encoder to output data of each input data input to the self-encoder. The importance is calculated (see (D) of FIG. 9). And the importance visualization part 15 displays the graph which shows transition of importance of each sensor data (refer (E) of FIG. 9). In the example of FIG. 9, a graph showing the transition of the importance of the input data for each of the sensors A to E with respect to the output data for the sensor A is displayed.

[Processing procedure of monitoring device]
Next, an example of a processing procedure performed by the monitoring apparatus 10 according to the first embodiment will be described with reference to FIGS. FIG. 10 is a flowchart illustrating an example of a flow of abnormality prediction processing in the monitoring device according to the first embodiment. FIG. 11 is a flowchart illustrating an example of the flow of importance calculation processing in the monitoring apparatus according to the first embodiment. FIG. 12 is a flowchart illustrating an example of the flow of graph display processing in the monitoring apparatus according to the first embodiment. Note that the process illustrated in FIG. 11 is a process in the case of narrowing down only input items (sensors) having a large reproduction error as importance calculation targets.

  First, the flow of abnormality prediction processing by the monitoring apparatus 10 will be described with reference to FIG. As illustrated in FIG. 10, when the collection unit 11 collects process data (Yes in step S101), process data in an analysis frame having a predetermined time point or a predetermined time width is obtained from the process data collected by the collection unit 11. Extract (step S102).

  Then, the frame abnormality evaluation value calculation unit 12b reproduces input data using an auto encoder (self-encoder) (step S103), and outputs output data for input data input to the neural network of the auto encoder. A reproduction error is detected, and a frame abnormality evaluation value that is the degree of abnormality of the monitored equipment is calculated according to the reproduction error (step S104).

  Thereafter, 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 S105). As a result, when the abnormality prediction evaluation value is less than the predetermined threshold (No at Step S105), the abnormality determination unit 13a ends the process as it is. Moreover, the abnormality determination part 13a outputs a warning sign, when an abnormality prediction evaluation value is more than a predetermined threshold value (Yes at Step S105) (Step S106).

  Next, the flow of importance calculation processing by the monitoring apparatus 10 will be described with reference to FIG. As illustrated in FIG. 11, the importance calculation unit 14 a acquires process data input to the learned model from the analysis frame buffer 17, and uses the frame abnormality evaluation value output from the learned model as a frame abnormality evaluation value. When it is obtained from the buffer 18 (Yes at Step S201), an input item (sensor) having a large reproduction error is extracted as an object of importance calculation (Step S202).

  Then, the importance calculation unit 14a calculates the importance of each input data with respect to the output data of the item extracted as the importance calculation target (step S203). Then, the importance noise removal unit 14b removes noise by smoothing the importance calculated by the importance calculation unit 14a with a predetermined time width (step S204). For example, the importance noise removing unit 14b smoothes the importance by using the method of Smooth Grad as a technique for removing noise.

  Next, the flow of graph display processing by the monitoring apparatus 10 will be described with reference to FIG. As illustrated in FIG. 12, when the acquisition unit 15a acquires the importance of the input data input to the self-encoder for the output data (Yes in step S301), the creation unit 15b is acquired by the acquisition unit 15a. A graph showing the transition of the importance of each sensor is created using the importance (step S302). Then, the importance level display unit 15c displays the graph created by the creation unit 15b (step S303).

(Effects of the first embodiment)
The monitoring apparatus 10 according to the first embodiment collects a plurality of data related to the monitoring target facility, and uses the collected plurality of data as input data to output data that reproduces each input data using a self-encoder. And the input data is compared with the output data to detect a reproduction error of the output data with respect to the input data. Then, the monitoring apparatus 10 uses each input data input to the self-encoder and output data output from the self-encoder to output data of each input data input to the self-encoder. Calculate importance. Therefore, the monitoring device 10 can dynamically and easily check the importance of each input data with respect to the output data.

  That is, in the monitoring apparatus 10 according to the first embodiment, normal abnormality detection or state determination is performed by passing a learned model of a self-encoder from an input value of a sensor or the like, and output in that state It is possible to dynamically confirm which input values are important. As a result, it becomes easy to determine which input value should be operated according to the output value.

  For example, the monitoring apparatus 10 according to the first embodiment can extract and present the input important for the output of the self-encoder together with the importance for each time and each sample input. This makes it possible to present an important sensor for the current state, and to deal with a system whose state changes. In addition, it is easier for a person who is not an expert who knows the state to specify which item is operated for that state.

  In the monitoring apparatus 10 according to the first embodiment, for example, when only output data items having a large reproduction error are targeted for importance calculation, the processing load is reduced and the processing speed is increased. Input can be identified efficiently.

(System configuration etc.)
Further, each component of each illustrated apparatus is functionally conceptual, and does not necessarily need to be physically configured as illustrated. In other words, the specific form of distribution / integration of each device is not limited to that shown in the figure, and all or a part thereof may be functionally or physically distributed or arbitrarily distributed in arbitrary units according to various loads or usage conditions. Can be integrated and configured. Furthermore, each processing function performed in each device is realized by a CPU or GPU and a program that is analyzed and executed by the CPU or GPU, or as hardware by wired logic. Can be realized.

  In addition, among the processes described in this embodiment, all or part of the processes described as being automatically performed can be manually performed, or the processes described as being manually performed All or a part of the above can be automatically performed by a known method. In addition, the processing procedure, control procedure, specific name, and information including various data and parameters shown in the above-described document and drawings can be arbitrarily changed unless otherwise specified.

(program)
It is also possible to create a program in which the processing executed by the monitoring device described in the above embodiment is described in a language that can be executed by a computer. For example, a monitoring program in which processing executed by the monitoring apparatus 10 according to the embodiment is described in a language that can be executed by a computer can be created. In this case, when the computer executes the monitoring program, the same effect as in the above embodiment can be obtained. Furthermore, the monitoring program may be recorded on a computer-readable recording medium, and the monitoring program recorded on the recording medium may be read by the computer and executed to execute the same processing as in the above embodiment.

  FIG. 13 is a diagram illustrating a computer that executes a monitoring program. As illustrated in FIG. 13, the computer 1000 includes, for example, a memory 1010, a CPU 1020, a hard disk drive interface 1030, a disk drive interface 1040, a serial port interface 1050, a video adapter 1060, and a network interface 1070. 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). The hard disk drive interface 1030 is connected to the hard disk drive 1090 as illustrated in FIG. The disk drive interface 1040 is connected to the disk drive 1100 as illustrated in FIG. For example, a removable storage medium such as a magnetic disk or an 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. The video adapter 1060 is connected to a display 1130, for example, as illustrated in FIG.

  Here, as illustrated in FIG. 13, the hard disk drive 1090 stores, for example, an OS 1091, an application program 1092, a program module 1093, and program data 1094. That is, the above monitoring program is stored in, for example, the hard disk drive 1090 as a program module in which a command executed by the computer 1000 is described.

  In addition, various data described in the above embodiment is stored as program data in, for example, the memory 1010 or the hard disk drive 1090. Then, the CPU 1020 reads the program module 1093 and the 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 the program data 1094 related to the monitoring program are not limited to being stored in the hard disk drive 1090, but may be stored in, for example, a removable storage medium and read out by the CPU 1020 via the disk drive or the like. Good. Alternatively, the program module 1093 and the program data 1094 related to the monitoring program are stored in another computer connected via a network (LAN (Local Area Network), WAN (Wide Area Network), etc.), and via the network interface 1070. May be read by the CPU 1020.

  The above embodiments and modifications thereof are included in the invention disclosed in the claims and equivalents thereof as well as included in the technology disclosed in the present application.

DESCRIPTION OF SYMBOLS 10 Monitoring apparatus 11 Collection part 12 Detection part 12a Analysis frame extraction part 12b Frame abnormality evaluation value calculation part 13 Output visualization part 13a Abnormality determination part 13b Chart display part 14 Calculation part 14a Importance calculation part 14b Importance noise removal part 15 Importance Visualization unit 15a Acquisition unit 15b Creation unit 15c Importance display unit 15d Notification unit 16 Process data buffer 17 Analysis frame buffer 18 Frame abnormality evaluation value buffer

Claims (8)

A collection means for collecting a plurality of data related to the monitoring target;
A plurality of data collected by the collecting means as input data, using a self-encoder, detecting means for detecting the state of the monitoring target;
Using each input data input to the self-encoder and output data output from the self-encoder, the importance of the input data input to the self-encoder is calculated for the output data And a calculating device.   The detection means outputs a plurality of data collected by the collection means as input data, outputs output data that reproduces each input data using the self-encoder, and outputs the input data and the output data. The monitoring apparatus according to claim 1, wherein a comparison error of the output data with respect to the input data is detected by comparison.   The monitoring device according to claim 2, wherein the detection unit detects a reproduction error of the output data with respect to the input data, and calculates an abnormality degree of the monitoring target according to the reproduction error.   The calculation means extracts output data in which the reproduction error detected by the detection means is equal to or greater than a predetermined threshold, and uses each input data input to the self-encoder and the extracted output data to extract The monitoring device according to claim 2, wherein importance of each input data with respect to the output data is calculated.   The calculating means uses the input data input to the self-encoder and the output data of the nodes of the intermediate layer in the neural network of the self-encoder to input the input data for the output data of the intermediate layer. The monitoring apparatus according to claim 1, wherein the importance level of the metric is calculated. Using the importance calculated by the calculation means, creating means for creating a graph showing the transition of the importance for the output data of each input data;
The monitoring apparatus according to claim 1, further comprising display means for displaying a graph created by the creating means. A monitoring method executed by a monitoring device,
A collection process for collecting a plurality of data related to the monitoring target;
A plurality of data collected by the collecting step as input data, using a self-encoder, a detection step of detecting the state of the monitoring target,
Using each input data input to the self-encoder and output data output from the self-encoder, the importance of each input data input to the self-encoder is calculated. A monitoring method characterized by including a calculating step. A collection step for collecting a plurality of data acquired in the monitored facility;
A collection step to collect multiple data about the monitored object;
A plurality of data collected by the collecting step as input data, using a self-encoder to detect the state of the monitoring target; and
Using each input data input to the self-encoder and output data output from the self-encoder, the importance of each input data input to the self-encoder is calculated. A monitoring program that causes a computer to execute the calculating step.

Images