Plant monitoring support device, method, and program

Abstract

To provide a plant monitoring support technology that can efficiently narrow down process values linked to abnormal events with high reliability.SOLUTION: A plant monitoring support device 10 includes an acquisition unit 15 that acquires a process value P of the monitored object 20, a reproduction unit 16 that outputs a reproduced value R obtained by compressing / reconstructing the process value P acquired in real time based on a model 28, a first calculation unit 11 that calculates a deviation D between the process value P and the reproduced value R, a detection unit 17 that detects signs of abnormality in the monitored object 20 based on a set threshold S and the deviation D, a second calculation unit 12 that calculates a fluctuation amount V of the deviation D based on the timing T at which the sign of abnormality is detected, and a selection unit 18 that selects a process value P of another attribute that is linked to an abnormality based on the fluctuation amount V.SELECTED DRAWING: Figure 1

Description

[Technical Field]

[0001] An embodiment of the present invention relates to a plant monitoring support technology that utilizes artificial intelligence (AI) to detect signs of abnormality. [Background technology]

[0002] When introducing plant monitoring support technology that puts AI-based anomaly detection into practical use, it is important not only to detect anomalies early, but also to be able to analyze the causes and take action quickly.Furthermore, there is a need for such plant monitoring support technology to visualize process values ​​related to detected anomalies.

[0003] Process values ​​output by numerous sensors and monitoring control devices installed to detect the status of plant component equipment are centrally managed in a central monitoring room or similar location. In power plants, a wide variety of process values ​​are monitored in real time to confirm the plant's operating status. When a sign of an abnormality is detected in a specific process value, other closely related process values ​​are also closely monitored. The cause of the detected abnormality and its impact on the plant's operating status are then investigated.

[0004] However, since identifying such highly correlated process values ​​is difficult for human judgment due to the vast number of targets, mechanical methods of narrowing them down are being considered. A known technique involves using two pieces of data that are assumed to have a causal relationship as input data and output data, and narrowing down the candidates based on the response characteristics of the latter to the former. In this case, the response characteristics of the output data to non-steady changes such as increases, decreases, and vibrations in the input data are narrowed down for each type of change.

[0005] Furthermore, there is a publicly known technology that uses AI to detect when the deviation between a "predicted value calculated from normal data" and a monitored process value exceeds a threshold, as a sign of an abnormality. In this case, it is also disclosed that the trend of deviation fluctuations can be visualized, and the simple rank correlation coefficient between process values ​​can be ranked and displayed. [Prior art documents] [Patent documents]

[0006] [Patent Document 1] International Application Publication No. 2014 / 091955 [Patent Document 2] Patent No. 5669553 [Patent Document 3] Patent No. 7309548 [Patent Document 4] Patent No. 7391765 Summary of the Invention [Problem to be solved by the invention]

[0007] However, the above-mentioned known techniques have the following problems when extracting a process value related to an abnormal event or a process value linked to an abnormal event due to low accuracy.

[0008] Specifically, there were the following issues: 1) even if the trend in deviation fluctuations is visualized, it is difficult to extract process values ​​that are linked to abnormalities from a large number of process values; 2) when simple rank correlation coefficients are displayed in a ranking, process values ​​that have large steady fluctuations or deviations are displayed at the top of the ranking even though they are unrelated to abnormalities; 3) the ranking fluctuates due to the influence of small fluctuations in the process values ​​(signal noise, pump pulsation, etc.); and 4) it is difficult to take into account the fact that process values ​​related to abnormalities fluctuate with a time delay.

[0009] The embodiments of the present invention have been made in consideration of the above circumstances, and have an object to provide a plant monitoring support technology that can efficiently narrow down process values ​​linked to abnormal events with high reliability. [Means for solving the problem]

[0010] The plant monitoring support device according to the embodiment is characterized by comprising an acquisition unit that acquires a process value of a monitored object, a reproduction unit that outputs a reproduced value obtained by compressing / reconstructing the process value acquired in real time based on a model, a first calculation unit that calculates a deviation between the process value and the reproduced value, a detection unit that detects a sign of abnormality in the monitored object based on a set threshold and the deviation, a second calculation unit that calculates a fluctuation amount of the deviation based on the timing at which the sign of abnormality is detected, and a selection unit that selects the process value of another attribute that is linked to an abnormality based on the fluctuation amount. [Effects of the Invention]

[0011] According to an embodiment of the present invention, a plant monitoring support technology is provided that can efficiently narrow down process values ​​associated with abnormal events with high reliability. [Brief explanation of the drawings]

[0012] [Figure 1] 1 is a configuration diagram of a plant monitoring support device according to a first embodiment of the present invention. [Figure 2] 4 is a graph showing a time series of deviations output by a first calculation unit. [Figure 3] A screen that displays the attributes and fluctuation amounts of process values ​​that are abnormally linked to each other. [Figure 4] FIG. 10 is a configuration diagram of a plant monitoring support device according to a second embodiment. [Figure 5] FIG. 10 is a configuration diagram of a plant monitoring support device according to a third embodiment. [Figure 6] 3 is a flowchart illustrating the steps of a plant monitoring support method and the algorithm of a plant monitoring support program. DETAILED DESCRIPTION OF THE INVENTION

[0013] (First embodiment) Hereinafter, an embodiment of the present invention will be described with reference to the accompanying drawings. Fig. 1 is a configuration diagram of a plant monitoring support device 10A (10) according to a first embodiment of the present invention. As described above, the plant monitoring support device 10A includes an acquisition unit 15 that acquires a process value P of a monitored object 20, a reproduction unit 16 that compresses the process value P acquired in real time based on a model 28 and then restores the compressed data (hereinafter referred to as data compression / restoration) to output a reproduced value R, a first calculation unit 11 that calculates a deviation D between the process value P and the reproduced value R, a detection unit 17 that detects a sign of abnormality in the monitored object 20 based on a set threshold S and the deviation D, a second calculation unit 12 that calculates a fluctuation amount V of the deviation D based on a timing T (Fig. 2) at which the sign of abnormality is detected, and a selection unit 18 that selects a process value P of another attribute that is linked to an abnormality based on the fluctuation amount V.

[0014] A plant as the monitored object 20 is provided with many sensors 25 that detect the status of its constituent equipment. Specific examples of the process values ​​P output by the sensors 25 include temperature, pressure, water level (displacement), flow velocity (speed), and acceleration. Furthermore, in a complex monitored object 20 such as a nuclear power plant, the process values ​​P are also input and output to many installed control devices 26. Examples of such control devices 26 include power pumps, power cylinders, and valves.

[0015] The operating status of the plant is monitored by consolidating and managing these process values ​​P in a central monitoring room. In a power plant, process values ​​P with a wide variety of attributes are monitored in real time, and when a sign of an abnormality is detected in a specific process value P, other closely related process values ​​P are also closely monitored at the same time to confirm the cause of the detection and the impact of the abnormality sign on the plant.

[0016] The acquisition unit 15 acquires a plurality of process values ​​P of the monitored object 20 in real time as time-series data. Note that analog signals among the process values ​​P acquired from the sensor 25 or the control device 26 are converted into digital data. The first accumulation unit 21 accumulates the acquired process values ​​P together with at least ID information of the sensor 25 or the control device 26 that is the output source and acquisition time information.

[0017] The learning unit 27 constructs a model 28 that is trained by compressing normal accumulated process values ​​P as input data and restoring the compressed output data to match the input data. An example of the model 28 is an autoencoder. Alternatively, a pre-constructed model 28 may be used, and the learning unit 27 is not an essential component of the plant monitoring support device 10.

[0018] Here, an autoencoder is a type of neural network. An autoencoder is an algorithm that compresses input data, retains only important features, and then restores the original dimensions as output data. This autoencoder receives input data at the nodes in the input layer and compresses it in the hidden layer. At this time, the input data is weighted according to its importance, and data with low scores is excluded (encoded). When moving to the output layer, the data is again weighted, and the sum of the data received by the node from multiple edges becomes the final value as output data (decoded). The learning unit 27 constructs a model 28 (autoencoder) so that this output data matches the input data.

[0019] The reproduction unit 16 outputs a reproduced value R (output data) obtained by compressing / reconstructing a process value P (input data) acquired in real time from the monitored object 20 based on a model 28. Then, the first calculation unit 11 calculates the difference between the process value P before input to the reproduction unit 16 and the reproduced value R after output from the reproduction unit 16, and outputs the difference as a deviation D. The output deviation D is accumulated in the second accumulation unit 22 in time series so as to correspond to each of the multiple process values ​​P. Preferably, the deviation D is also accumulated in the second accumulation unit 22 using common ID information and time information so as to be linked to the process values ​​P accumulated in the first accumulation unit 21.

[0020] The detection unit 17 detects signs of abnormality in the monitored object 20 based on the set threshold value S and deviation D. The threshold value S is set for each of a plurality of process values ​​P and is registered in advance in a registration unit (not shown). When the deviation D exceeds the threshold value S in the process value P acquired in real time, it is determined that a sign of abnormality has been detected.

[0021] FIG. 2 is a graph showing a time series of deviation D of the process value. The second calculation unit 12 (FIG. 1) calculates the deviation D by calculating the deviation D from a threshold value S (S A ) and an abnormality sign is detected, the deviation D(D A ) fluctuation V(V A Here, the second calculation unit 12 divides the time series data of the deviation D based on the timing T. Then, the amount of fluctuation V is calculated by taking the difference between an arbitrary value of the time series data of the deviation D after division and an arbitrary value of the time series data before division.

[0022] The calculated variation amount V may be stored in a storage unit (not shown) using common ID information and time information for each piece of time series data of the divided process value P. The time series data of the deviation D may be divided by arbitrarily adjusting the time width before and after the timing T at which a sign of an abnormality is detected. Furthermore, this time width may be set separately depending on the attribute of the process value P.

[0023] As shown in Figure 2, the amount of fluctuation VA The amount of fluctuation V of the process value linked to the abnormality has a time delay. B is observed. The selection unit 18 selects process values ​​P that are abnormally linked with each other and have different attributes 37 based on the fluctuation amounts V of the plurality of process values.

[0024] 3 is a screen of the display unit 29 that displays the attributes 37 and fluctuation amounts D of process values ​​P that are abnormally linked to one another (including process values ​​in which a sign of an abnormality has been detected). The selection unit 18 can arbitrarily set the number of process values ​​P that exceed a threshold value S to be displayed on the display unit 29, for example, or it may set process values ​​P with large fluctuation amounts regardless of the threshold value S (TOP 10 shown in the figure). Alternatively, the attributes 37 of process values ​​that are related to one another may be linked to one another in advance, and when a sign of an abnormality is detected in the attribute 37 of a certain process value, the attribute 37 of the linked process value may be displayed as a candidate for abnormal linkage, and the fluctuation amount V may be observed.

[0025] 3, the display unit 29 displays the attribute 37 of the process values ​​that are abnormally linked together, accompanied by the amount of fluctuation D, but it may also display the process value P, deviation D, threshold value S, and timing T. Furthermore, although not shown, it may also display the simple rank correlation coefficient between the attributes 37 of the process values.

[0026] In this way, when a sign of an abnormality is detected in any one of the process values ​​P, the display unit 29 displays not only that process value P but also other process values ​​P that are linked to the abnormality. This makes it possible to give the supervisor a stricter warning. Furthermore, by eliminating the influence of a monitored process value that has a constantly large deviation D, the supervisor can accurately recognize the process value that is linked to the abnormality.

[0027] (Second embodiment) Next, a second embodiment of the present invention will be described with reference to Fig. 4. Fig. 4 is a configuration diagram of a plant monitoring support device 10B (10) according to the second embodiment. The plant monitoring support device 10B of the second embodiment has a configuration in which a first processing unit 31 and a second processing unit 32 are further added to the configuration of the first embodiment described above. In Fig. 4, parts having the same configuration or function as those in Fig. 1 are indicated by the same reference numerals, and duplicated explanations will be omitted.

[0028] The first processing unit 31 averages the deviation D in the process value P acquired before the timing T at which the abnormality sign is detected. The second processing unit 32 averages the deviation D in the process value P acquired after the timing T at which the abnormality sign is detected. The periods for the averaging processes in the first processing unit 31 and the second processing unit 32 are arbitrarily determined for each attribute 37 of the process value.

[0029] In the second calculation unit 12, the amount of fluctuation V is calculated by taking the difference between the deviation D averaged over a predetermined period before the timing T and the deviation D averaged over a predetermined period after the timing T. According to the second embodiment, not only is it possible to eliminate the influence of process values ​​with constantly large deviations, but it is also possible to eliminate the influence of noise and time lags associated therewith, allowing the monitor to recognize abnormally linked process values ​​with high accuracy.

[0030] (Third embodiment) Next, a third embodiment of the present invention will be described with reference to Fig. 5. Fig. 5 is a configuration diagram of a plant monitoring support device 10C (10) according to the third embodiment. The plant monitoring support device 10C of the third embodiment has a configuration in which a standardization unit 35 is further added to the configuration of the first or second embodiment described above. In Fig. 5, parts having the same configuration or function as those in Figs. 1 and 4 are indicated by the same reference numerals, and duplicated explanations will be omitted.

[0031] The normalization unit 35 normalizes the calculated deviation D to obtain a normalized deviation E. The normalized deviation E output from the normalization unit 35 is then accumulated in the second accumulation unit 22 in chronological order for each of the plurality of process values ​​P. Based on this normalized deviation E, the detection unit 17 detects the process value P indicating an abnormality sign, and the selection unit 18 selects the process value P linked to an abnormality (see FIG. 2).

[0032] The normalization unit 35 calculates the normalized deviation E by dividing the deviation D by the variation in the process value P. Here, the standard deviation of the process value P used in the learning unit 27 or the standard deviation of the deviation D is used as the variation in the process value P, which is the denominator of the division. However, the denominator of the division used for normalization is not limited to the variation in the process value P.

[0033] According to the third embodiment, not only can the influence of process values ​​with large steady-state deviations be eliminated, but also the influence of accidental fluctuations in the process values ​​be eliminated, allowing the monitor to recognize abnormally linked process values ​​with high accuracy. Furthermore, it becomes easy to compare the fluctuation amount V even between process values ​​with different attributes 37 (for example, when the physical quantities measured are different, such as pressure and flow rate, or when the physical quantity is the same but the targets are different).

[0034] The steps of the plant monitoring support method and the algorithm of the plant monitoring support program will be described with reference to the flowchart in Fig. 6. First, normal process values ​​P of the monitored object 20 are compressed as input data, and a trained model 28 is constructed so that the restored output data matches the input data (S11).

[0035] Next, the process value P of the total number of monitored objects 20; m = 1 to M m (t n ) is acquired in a time series of n = 1 to N. First, m = 1, n = 1 (S12), and the process value P m (t n ) is obtained (S13). Then, based on the model 28, the process value P m (t n ) is compressed / decompressed and the reconstructed value Rm (t n ) is output (S14).

[0036] Next, the process value P m (t n ) and the regeneration value R m (t n ) deviation D m (t n ) is calculated (S15), a preset threshold value S is obtained (S16), and the two are compared. Then, the deviation D m (t n Unless the process value P exceeds the threshold value S (S17; Yes), m+1 (t n ), the loop from S13 to S17 is run (S18; No, Yes), and the time-series process value P m (t n+1 ) and loops S13 to S18 (S19; No, Yes, END).

[0037] And deviation D m (t n ) exceeds the threshold value S (S17; No), it means that an abnormality sign has been detected, and at that timing T, the deviation D m (t n ) fluctuation V m (t n ) is calculated (S20). Furthermore, the loop of S13 to S17 is repeated in the same way (S18; No, Yes), and other process values ​​P m+1 (t n ) and its fluctuation amount V m+1 (t n ) (S21).

[0038] Or, instead of threshold S, for example, deviation D for the total process quantity m (t n ) fluctuation V m (t n ) is calculated (S20), and the abnormality-linked process amount is selected as the TOP 10, etc. Then, the selected process amount is updated or added to the display on the display unit 29 (S22). Then, the time-series process value P m (t n+1) and loops S13 to S18 (S19; No, Yes, END).

[0039] According to at least one of the embodiments of the plant monitoring support device described above, by selecting an attribute of a process value linked to an abnormality from the amount of variation in deviation calculated based on the timing at which a sign of an abnormality is detected, it becomes possible to efficiently narrow down the process values ​​linked to an abnormal event with high reliability.

[0040] Although several embodiments of the present invention have been described, these embodiments are presented as examples and are not intended to limit the scope of the invention. These embodiments can be implemented in various other forms, and various omissions, substitutions, modifications, and combinations can be made without departing from the spirit of the invention. These embodiments and their modifications are included within the scope and spirit of the invention, as well as the inventions described in the claims and their equivalents.

[0041] The plant monitoring support device described above includes a control device with a highly integrated processor such as a dedicated chip, FPGA (Field Programmable Gate Array), GPU (Graphics Processing Unit), or CPU (Central Processing Unit), a storage device such as ROM (Read Only Memory) or RAM (Random Access Memory), an external storage device such as HDD (Hard Disk Drive) or SSD (Solid State Drive), a display device such as a monitor, an input device such as a mouse or keyboard, and a communication I / F, and can be realized with a hardware configuration using a normal computer. Therefore, the components of the plant monitoring support device can also be realized by a computer processor and can be operated by a plant monitoring support program.

[0042] The plant monitoring support program may be provided by being pre-installed in a ROM or the like. Alternatively, the program may be provided by being stored in an installable or executable file format on a computer-readable storage medium such as a CD-ROM, CD-R, memory card, DVD, or flexible disk (FD).

[0043] The plant monitoring support program according to this embodiment may be stored on a computer connected to a network such as the Internet and provided by downloading it via the network. The plant monitoring support device may also be configured by combining separate modules that independently perform the functions of the components and are interconnected via a network or dedicated lines. [Explanation of symbols]

[0044] 10 (10A, 10B, 10C)...Plant monitoring support device, 11...First calculation unit, 12...Second calculation unit, 15...Acquisition unit, 16...Reproduction unit, 17...Detection unit, 18...Selection unit, 20...Monitoring target, 21...First storage unit, 22...Second storage unit, 25...Sensor, 26...Control device, 27...Learning unit, 28...Model, 29...Display unit, 31...First processing unit, 32...Second processing unit, 35...Normalization unit, 37...Attribute, P,P m (t n )…Process value, R,R m (t n )…Regeneration value, D,D m (t n )…Deviation, S…Threshold, T…Timing, V,V m (t n )...amount of variation.

Claims

1. an acquisition unit that acquires a process value of a monitoring target; a reproduction unit that outputs a reproduced value obtained by compressing / reconstructing the process value acquired in real time based on a model; a first calculation unit that calculates a deviation between the process value and the reproduced value; a detection unit that detects a sign of abnormality in the monitored object based on a set threshold value and the deviation; a second calculation unit that calculates a fluctuation amount of the deviation based on the timing at which the abnormality sign is detected; a selection unit that selects the process value of another attribute that is abnormally linked based on the amount of fluctuation.

2. 2. The plant monitoring support device according to claim 1, a first processing unit that averages the deviation in the process value acquired before the timing at which the abnormality sign is detected; a second processing unit that averages the deviation in the process value acquired after the timing at which the abnormality sign is detected, The fluctuation amount is calculated based on the deviation that has been averaged.

3. 3. The plant monitoring support device according to claim 1, a normalization unit that normalizes the calculated deviation to obtain a normalized deviation; A plant monitoring support device that detects the abnormality sign and selects the abnormality linkage based on the normalized deviation.

4. 3. The plant monitoring support device according to claim 1, The plant monitoring support device includes a learning unit that compresses normal process values ​​as input data and trains the model so that the restored output data matches the input data.

5. 3. The plant monitoring support device according to claim 1, A plant monitoring support device comprising a display unit that displays an attribute of the process value linked to the abnormality.

6. acquiring a process value to be monitored; a step of outputting a reproduced value obtained by compressing / reconstructing the process value acquired in real time based on a model; calculating a deviation between the process value and the reconstructed value; detecting a sign of abnormality in the monitored object based on the set threshold and the deviation; calculating a variation amount of the deviation based on the timing at which the abnormality sign is detected; and selecting the process value of another attribute that is abnormally linked based on the amount of fluctuation.

7. On the computer, acquiring a process value to be monitored; a step of outputting a reproduced value obtained by compressing / reconstructing the process value acquired in real time based on a model; calculating a deviation between the process value and the reconstructed value; detecting a sign of abnormality in the monitored object based on the set threshold and the deviation; calculating a fluctuation amount of the deviation based on the timing at which the abnormality sign is detected; a step of selecting the process value of another attribute that is abnormally linked based on the amount of fluctuation.

Images