Plant management device, control method of plant management device, and program

The plant management device addresses the black box issue of AI-based abnormality detection by providing clear displays and calculations of abnormality and influence, enabling operators to understand and respond to plant abnormalities effectively.

JP2024171192AInactive Publication Date: 2024-12-11THE CHUGOKU ELECTRIC POWER CO INC

Patent Information

Application Number
JP2023088139
Authority / Receiving Office
JP · JP
Patent Type
Applications
Current Assignee / Owner
Filing Date
2023-05-29
Publication Date
2024-12-11
Estimated Expiration
Not applicable · inactive patent

AI Technical Summary

Technical Problem

Existing plant monitoring systems using artificial intelligence for abnormality detection become a 'black box,' making it difficult for operators to understand the rationale behind judgment results, and it is challenging to identify the location and extent of abnormalities within complex plant systems.

Method used

A plant management device that displays abnormality states and influences using influence information storage, calculates index values to quantify abnormality and influence, and provides a systematic neural network to express causal relationships between systems, allowing operators to understand the extent and origin of abnormalities.

Benefits of technology

Enables operators to quickly grasp the state and extent of abnormalities in plant systems, facilitating timely and effective responses to maintain plant stability.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure 2024171192000001_ABST
    Figure 2024171192000001_ABST
Patent Text Reader

Abstract

To more properly grasp a state of an anomaly that has occurred in a plant.SOLUTION: A plant management device comprises: an influence power information storage unit which stores influence power information representing susceptibility from an upstream system for each of systems sharing plant control; a plant data acquisition unit which acquires plant data periodically measured by each system; a second index value calculation unit which calculates a second index value representing the degree of deviation between a first index value indicative of a relation of two types of plant data and a first reference value for each of combinations for selecting two types from plant data measured in the same system per system; a third index value calculation unit which calculates the maximum value of the second index value of each system as a third index value representing the degree of an anomaly in each system; a fourth index value calculation unit which calculates a fourth index value representing the degree of influence of an anomaly which each system receives from an upstream system by using the third index value of each system and the influence power information; and an anomaly information display unit which displays the anomaly state of the plant by listing the third and fourth index values of each system.SELECTED DRAWING: Figure 11
Need to check novelty before this filing date? Find Prior Art

Description

[Technical field]

[0001] The present invention relates to a plant management device, a control method for a plant management device, and a program. [Background technology]

[0002] In plants such as nuclear power plants, various data (plant data) such as temperature, pressure, current, and voltage are measured at various locations within the plant. This plant data changes from moment to moment depending on the operating state of the plant, and is monitored in various locations including the central control room. For example, plant operators check the changing trends of the plant data by looking at trend graphs that display the plant data output from indicators, controllers, and operation monitoring computers in chronological order. By thoroughly monitoring this data, operators can detect malfunctions, abnormalities, and other problems at the plant early on.

[0003] Traditionally, anomaly detection in plants has been carried out by operators monitoring plant data and recognizing anomalies, but in recent years, support systems using AI (Artificial Intelligence) that make full use of large-scale machine learning, such as plant predictive management systems, multivariate time series data analyzers, plant anomaly diagnosis devices, and plant anomaly diagnosis systems, are being put into practical use as systems to support the monitoring of this plant data.

[0004] These support systems detect plant anomalies by performing data processing such as correlation processing and multivariate analysis on large amounts of plant data. This type of data processing using machine learning has excellent characteristics, such as being faster and more real-time than operations performed by operators, being easy to compare with past data, and enabling quantitative judgments.

[0005] As techniques for supporting the monitoring of such plants, techniques such as those disclosed in Patent Documents 1 and 2 are known. [Prior art documents] [Patent documents]

[0006] [Patent Document 1] JP 2014-127276 A [Patent Document 2] JP 2017-062730 A Summary of the Invention [Problem to be solved by the invention]

[0007] Patent Document 1 describes a plant monitoring device that monitors the operating states of a plurality of plants.

[0008] This plant monitoring device includes state quantity acquisition means for acquiring a bundle of state quantities, which is a collection of state quantities for each of a plurality of measurement items, from each of a plurality of plants, unit space creation means for creating a unit space for each of the plurality of plants constituted by a collection of a plurality of the bundles of state quantities, unit space storage means for storing the unit spaces for each of the plurality of plants created by the unit space creation means, Mahalanobis distance calculation means for calculating, when a bundle of state quantities for one of the plurality of plants is acquired by the state quantity acquisition means, a Mahalanobis distance of the bundle of state quantities for the one plant based on the unit space of the one plant stored in the unit space storage means by the unit space creation means, and and a determination means for determining whether or not an operating state of the one plant is normal depending on whether or not the Mahalanobis distance for the one plant calculated by the determination means is within a predetermined threshold value, and the unit space creation means has a similar plant selection means for selecting, when a unit space of the one plant is not stored in the unit space storage means, another plant among the multiple plants that is similar to the one plant and has a unit space stored in the unit space storage means, and a new space creation means for creating a unit space of the one plant using the unit space of the other plant selected by the similar plant selection means, and storing the unit space of the one plant in the unit space storage means.

[0009] Patent document 2 also describes an anomaly diagnosis system that derives the correlation strength between two of a plurality of measurement parameters measured in a nuclear plant as a correlation value, and when a monitoring indication value, which is a correlation value obtained by adding up the derived correlation values, exceeds a preset threshold value, diagnoses abnormal events in a nuclear plant based on the anomaly sign detection result of an anomaly sign monitoring system that detects abnormal signs in the nuclear plant.

[0010] This abnormality diagnosis system includes an abnormality diagnosis control unit that acquires, as the abnormality sign detection result, a parameter transition that is a transition of the correlation value of the measurement parameter that becomes an abnormal value that is determined to be an abnormality occurring in the correlation value during a monitoring period from the undetected time to the abnormality detection time, and a parameter contribution degree that is a ratio of a change in the correlation value of the measurement parameter to a change in the monitoring indication value for detecting an abnormality sign during the monitoring period, where the time when an abnormality sign is detected is defined as a sign detection time and the time before the sign detection time is defined as an undetected time. The abnormality diagnosis system also includes an abnormality diagnosis control unit that acquires, as the abnormality sign detection result, a parameter transition that is a transition of the correlation value of the measurement parameter that becomes an abnormal value that is determined to be an abnormality occurring in the correlation value during a monitoring period from the undetected time to the sign detection time, and a parameter contribution degree that is a ratio of a change in the correlation value of the measurement parameter to a change in the monitoring indication value for detecting an abnormality sign during the monitoring period, abnormal events of the nuclear plant, abnormality sign equipment associated with the abnormal events, and past operation of the nuclear plant. and a database storing the parameter transitions for judgment that are generated based on history and associated with the abnormal event, and the parameter contributions for judgment that are generated based on the past operating history of the nuclear plant and associated with the abnormal event, wherein the abnormality diagnosis control unit performs a consistency determination between the parameter transitions and the parameter contributions included in the abnormality sign detection result obtained from the abnormality sign monitoring system and the parameter transitions and the parameter contributions stored in the database, and identifies the abnormal event and the abnormality sign equipment associated with the parameter transitions and parameter contributions that are determined to match.

[0011] However, when using artificial intelligence such as machine learning in a system that monitors and diagnoses abnormalities in a plant, the process of monitoring and diagnosing abnormalities becomes a black box, making it difficult to provide a rational explanation for the results. Therefore, it is necessary to build a model that makes it easy for operators to understand the results.

[0012] In addition, because the state of a plant changes as various components interact with each other in complex ways, if an abnormality occurs in the plant, it is often difficult to identify the location (origin) of the abnormality or the extent of the abnormality's impact, and there is also concern that it may be difficult to determine whether the impact is local or widespread.

[0013] Therefore, there is a demand for technology that enables operators to better understand abnormal conditions occurring in the plant.

[0014] The present invention has been made in consideration of such problems, and provides a plant management device, a control method for a plant management device, and a program that enable operators to more appropriately understand the state of an abnormality occurring in a plant. [Means for solving the problem]

[0015] A plant management device according to one embodiment of the present invention is a plant management device that displays the state of an abnormality occurring in a plant, and includes an influence information storage unit that stores influence information representing susceptibility to influences from an upstream system that performs more upstream control for each system that shares control in the plant, a plant data acquisition unit that acquires multiple types of plant data periodically measured in each system, and a combination of two types of plant data selected from plant data measured within the same system for each system, the combination of the two types of plant data calculated from measurement values ​​of the two types of plant data measured in a predetermined period. the second index value calculation unit calculating a second index value indicating the degree of deviation between a first index value indicating a relationship between the upstream and downstream data and a first reference value serving as a reference for the relationship; a third index value calculation unit calculating, for each system, a maximum value of the second index values ​​as a third index value indicating the degree of an abnormality occurring in the system; a fourth index value calculation unit calculating, for each system, a fourth index value indicating the degree of influence of the abnormality on each system from an upstream system using the third index value and the influence information of each system; and an abnormality information display unit displaying a list of the third index value and the fourth index value in each system to display the state of the abnormality in the plant.

[0016] In addition, the problems and solutions disclosed in this application will be made clear by the description in the section of the preferred embodiment of the invention and the descriptions in the drawings. Effect of the Invention

[0017] This makes it possible to more accurately grasp the state of any abnormalities occurring in the plant. [Brief description of the drawings]

[0018] [Figure 1] FIG. 1 is a diagram showing an overall configuration. [Diagram 2] FIG. [Diagram 3] FIG. 1 is a diagram illustrating a configuration of a plant management device. [Figure 4] FIG. [Diagram 5] FIG. 13 is a diagram showing an influence information table. [Figure 6] FIG. 13 is a diagram illustrating a plant data management table. [Figure 7] FIG. 13 illustrates a correlation map table. [Figure 8] FIG. 2 is a diagram illustrating a functional configuration of a plant management apparatus. [Figure 9] FIG. 1 is a diagram showing a scatter plot of plant data. [Figure 10] 13 is a diagram for explaining calculation of a third index value and a fourth index value. FIG. [Figure 11] FIG. 1 is an illustration of a lineage neuron heat map. [Figure 12] FIG. 2 is a diagram for explaining a process flow of the plant management device. [Figure 13] FIG. 1 is a diagram showing an overall configuration. [Figure 14] FIG. [Figure 15] FIG. 13 illustrates an event management table. [Figure 16] FIG. 4 is a diagram showing an auxiliary machine management table. [Figure 17] FIG. 4 is a diagram showing the relationship between an operation pattern of an auxiliary machine and a correlation map. [Figure 18] FIG. 13 is a diagram showing an example of an output screen. [Figure 19] FIG. 2 is a diagram showing a process flow of the plant management device. [Figure 20] FIG. 13 is a diagram showing a time series graph display. [Figure 21] FIG. 2 is a diagram showing a process flow of the plant management device. DETAILED DESCRIPTION OF THE PREFERRED EMBODIMENTS

[0019] At least the following matters will become apparent from the description of this specification and the accompanying drawings. Hereinafter, the present invention will be described based on an embodiment with reference to the accompanying drawings. ==First embodiment== 1 shows an example of an overall configuration including a plant management device 100 according to a first embodiment of the present invention, a plant 1000, and an operation monitoring computer 200. In the following description, the plant management device 100 may be referred to as a plant anomaly monitoring system 100, a plant data symptom management system 100, or simply as a system 100.

[0020] The plant 1000 is a facility that produces target products by performing various processes and machining on various materials such as iron, oil, etc. In this embodiment, a nuclear power plant that generates electricity using nuclear fuel as an energy source will be described as an example of the plant 1000.

[0021] In the plant 1000, various controls are performed to obtain target objects from materials, and these controls are shared among management units called systems 1100. In Fig. 1, the plant 1000 is shown to have four systems 1100, system A 1100A to system D 1100D. In the case of a nuclear power plant, for example, the systems 1100 may correspond to a main steam system, a condensate feedwater system, a control rod drive system, a boric acid water injection system, a reactor auxiliary cooling water system, and the like.

[0022] The systems 1100 are interconnected, and in this embodiment, as shown in Fig. 2, the state of control in system A1100A affects the state of control in system B1100B and system C1100C, and the state of control in system C1100C affects the state of control in system D1100D. Therefore, system A1100A is an upstream system with respect to system B1100B and system C1100C, and system C1100C is an upstream system with respect to system D1100D. Conversely, system B1100B and system C110C are downstream systems of system A1100A, and system D1100D is a downstream system of system C1100C. Of course, there are cases where such connections between systems 1100 form a loop.

[0023] The operation monitoring computer 200 is an information processing device such as a computer that collects multiple types of plant data measured in each system 1100 of the plant 1000. The plant data includes, for example, the water level and core temperature of the reactor, the amount of power generated, and the turbine speed. These plant data are distributed, for example, to a central control room (not shown) via a network 500 and monitored by operators.

[0024] The network 500 is any of various information and communication networks such as the Internet, a LAN (Local Area Network), or a telephone network.

[0025] The plant management device 100 is an information processing device such as a computer that acquires the above-mentioned multiple types of plant data from the operation monitoring computer 200 and displays the state of an abnormality that has occurred in the plant 1000. Although details will be described later, the plant management device 100 according to this embodiment displays the state of an abnormality that has occurred in the plant 1000, allowing the operators to more appropriately grasp the state of the abnormality that has occurred in the plant 1000.

[0026] 3 shows a hardware configuration of the plant management device 100. The plant management device 100 is an information processing device such as a computer including a CPU (Central Processing Unit) 110, a memory 120, a communication device 130, a storage device 140, an input device 150, an output device 160, and a recording medium reading device 170.

[0027] The storage device 140 stores various programs, such as the plant management device control program 700 executed by the CPU 110, and data.

[0028] The plant management device control program 700 and various data stored in the storage device 140 are read into the memory 120 and executed or processed by the CPU 110, thereby implementing various functions of the plant management device 100.

[0029] Here, the storage device 140 is a non-volatile storage device such as a hard disk, a solid state drive (SSD), or a flash memory.

[0030] As shown in FIG. 4, the storage device 140 stores a plant management device control program 700, an influence information table 600, a plant data management table 610, a correlation map table 620, an event management table 630, and an auxiliary equipment management table 640.

[0031] Returning to FIG. 3, the recording medium reader 170 reads the plant management device control program 700 and data recorded on a recording medium 800 such as a CD, a DVD, or an SD card, and stores them in the storage device 140 .

[0032] The communication device 130 transmits and receives various data and the plant management device control program 700 to and from the operation monitoring computer 200 and other computers (not shown) via the network 500. For example, the above-mentioned plant management device control program 700 and data can be stored in another computer, and the plant management device 100 can download the plant management device control program 700 and data from this computer. Alternatively, the communication device 130 can communicate with a mobile terminal such as a smartphone or laptop (not shown) carried by an operator of the plant 1000, and transmit and receive various information.

[0033] The input device 150 is a device such as various buttons, switches, a mouse, a keyboard, a microphone, etc. that accepts input of commands and data by a user such as an operator, and functions as an input user interface.

[0034] The output device 160 is, for example, a display device such as a display, a speaker, or the like, and functions as an output user interface. <Influence Information Table> 5 shows an example of the influence information table 600. The influence information table 600 is a table that stores, for each system 1100 that shares and executes control in the plant 1000, influence information that indicates the susceptibility of the system 1100 to an influence from an upstream system 1100 that performs control further upstream.

[0035] The influence information table 600 shown in Figure 5 corresponds to Figure 2, and since system A 1100A is the most upstream system, no influence information corresponding to it is recorded, but for systems B 1100B to D 1100D, influence information from each upstream system is recorded.

[0036] The influence information is composed of a pair of values ​​of w and v.

[0037] w is information that quantifies the susceptibility to influence from other systems 1100, and is defined, for example, by the square ratio (a^2 / A^2) of the supply pipe diameter a in the system 1100 to the rotating equipment outlet pipe diameter A in the upstream system 1100. Note that it is preferable to appropriately review and optimize the value of w by comparing and verifying the actual range of influence of an anomaly that has occurred in the plant 1000 with the range of influence of the anomaly displayed by the plant management device 100. This enables the plant management device 100 to more appropriately display the range of influence of the anomaly.

[0038] v is a value indicating whether or not there is a link between the system 1100 in question and the upstream system 1100, and is 1 if there is a link, and 0 if there is not. Normally, it is linked to the opening and closing of a shut-off valve for the fluid provided between the system 1100 in question and the upstream system 1100. Specifically, if there is a shut-off motor valve between the system 1100 in question and the upstream system 1100, the open / closed state determines whether or not fluid is supplied from the upstream system 1100, so if the motor valve is open, it is 1, and if it is closed, it is 0. In the case of a manual valve, whether it is open or closed is determined at the stage of configuring the system 1100, so if the manual valve is open, 1 is input, and if it is closed, 0 is input in advance. In the case of linkage through an electrical signal, it may be linked to the ON / OFF state of a switch.

[0039] In this manner, in this embodiment, the influence force information is determined for each combination of a system 1100 and one or more upstream systems 1100 that influence the control of the system 1100 in question. <Plant data management table> 6 shows an example of the plant data management table 610. The plant data management table 610 is a table in which a plurality of types of plant data acquired by the plant management device 100 through the operation monitoring computer 200 is stored.

[0040] In the example shown in Fig. 6, the "system number" field lists the system 1100 where the plant data was measured. The "ID" field lists identification information that is uniquely assigned to each type of plant data. The "input point name" field lists the name assigned to the plant data.

[0041] The "Lower Limit of Measurement Range" and "Upper Limit of Measurement Range" columns indicate the upper and lower limits that the measuring instrument for the plant data can measure. The plant management device 100 can normalize the acquired plant data to a range of 0 to 1 by converting the acquired plant data into relative values ​​where the lower limit of the measurement range is 0 and the upper limit of the measurement range is 1.

[0042] The "scanning period" field indicates the measurement period of the plant data.

[0043] The "Alarm Lower Limit" and "Alarm Upper Limit" columns contain the lower and upper limits of a predetermined range that is determined as a normal range for plant data values. For example, when the acquired plant data is smaller than the alarm lower limit or larger than the alarm upper limit, the plant management device 100 outputs an alarm associated with the plant data.

[0044] The "detector number" field contains identification information of the measuring device of the plant data. The "measurement value" field contains the measurement value of the plant data.

[0045] In addition, although not shown in FIG. 6, the plant data management table 610 also lists the type of event occurring, time information (time stamp), and operation patterns of auxiliary equipment such as pumps. <Correlation map table> An example of the correlation map table 620 is shown in Fig. 7. The correlation map table 620 is a table that stores, for each combination of two types of plant data selected from plant data measured in the same system 1100, a first index value (details to be described later) indicating the relationship between the two types of plant data calculated from the measured values ​​of the two types of plant data measured in a predetermined period (e.g., the last 30 minutes or a period during which a specific event occurs). Note that in the correlation map table 620 shown in Fig. 7 and in the following embodiments, plant data may be referred to as a parameter. In the correlation map table 620, a first index value (not shown in Fig. 7) is stored for each combination of two types of plant data selected from nine types of plant data specified by parameters a to i.

[0046] The correlation map table 620 is created for each system 1100. Then, the first index value is calculated by combining the plant data measured in each system 1100.

[0047] Moreover, it is preferable to create the correlation map table 620 by combining plant data related to temperature and by combining plant data related to physical quantities other than temperature (such as current values ​​and pressures).The reason for this is that the plant data related to temperature and the plant data related to physical quantities other than temperature often have significantly different response speeds (time constants) to changes in the control state, so even if an attempt is made to obtain the first index value by combining the former and the latter, it is difficult to obtain the first index value with high accuracy.

[0048] The first index value is a coordinate value when a point determined from each representative value (e.g., average value) of the two types of plant data measured during the above-mentioned predetermined period is plotted on a scatter diagram of the two types of plant data. That is, in this embodiment, the relationship between these plant data is expressed using a pair (coordinate value) of the representative values ​​of the two types of plant data.

[0049] Therefore, when some abnormality occurs in the system 1100, if any of the plant data changes due to the influence of the abnormality, the coordinate value changes. Conversely, by detecting the change in the coordinate value, it is possible to detect that some abnormality has occurred in the system 1100.

[0050] An example of a scatter plot of two types of plant data is shown in Fig. 9. Fig. 9 is a scatter plot of two types of plant data represented by parameters A and B (in Fig. 9, each plant data is normalized to the range of 0 to 1). Each point marked with a circle, square, and triangle indicates a different sampling period.

[0051] For example, the open triangles represent the measured values ​​for the period during which event 1 occurred (first predetermined period) plotted on a scatter plot, and the filled triangles represent the average value (first index value) of the coordinate values ​​of these open triangles. In other words, the filled triangles represent the center of gravity of the open triangles.

[0052] Similarly, the open square marks are the measured values ​​for the period (second predetermined period) during which event 2 occurred plotted on the scatter diagram, and the filled square marks represent the average value (first index value) of the coordinate values ​​of these open square marks. In other words, the filled square marks represent the center of gravity of the open square marks.

[0053] The open circles are the measured values ​​during a period (reference period) when no abnormalities occurred in the plant 1000, plotted on a scatter diagram, and the filled circles represent the average value (first reference value) of the coordinate values ​​of these open circles. In other words, the filled circles represent the center of gravity of the open circles. This first reference value is used as a standard for the relationship between the data of the two plants.

[0054] The Euclidean distance between the solid triangle and the solid circle represents a second index value that indicates the degree of deviation between the first index value in the first predetermined period and the first reference value.

[0055] Similarly, the Euclidean distance between the solid square and the solid circle represents a second index value that indicates the degree of deviation between the first index value in the second predetermined period and the first reference value.

[0056] 9, the second index value (degree of deviation between the first index value and the first reference value) is larger in the second specified period than in the first specified period, so the possibility that some kind of abnormality has occurred in the grid 1100 is higher in the second specified period than in the first specified period. However, both the second index value in the first specified period and the second index value in the second specified period may be within the range of normal fluctuations.

[0057] The correlation map table 620 may target only combinations in which the absolute value of the correlation coefficient of the measured values ​​in the reference period exceeds a predetermined value (e.g., 0.2), in addition to targeting all combinations of two types of plant data selected from plant data measured in the same system 1100. This is because, in the case of a combination of plant data in which the absolute value of the correlation coefficient is small (e.g., 0.2 or less), the change in the first index value is originally large regardless of whether or not there is an abnormality in the plant 1000, making it difficult to detect an abnormality even if it occurs in the plant 1000. In this way, by eliminating combinations that do not contribute to the detection of an abnormality in the plant 1000, the plant management device 100 does not perform unnecessary processing, and it is possible to prevent a decrease in calculation efficiency. <Event management table> 15 shows an example of the event management table 630. The event management table 630 is a table for linking an event that has occurred in the system 1100 to each piece of plant data. That is, each piece of plant data is stored in the plant data management table 610 in association with the event that was occurring at the time of measurement. Of course, when no event has occurred, each piece of plant data is linked to information that no event has occurred.

[0058] According to this embodiment, it becomes possible to create a scatter diagram for each event separately, as shown in Fig. 9. Therefore, it becomes possible to calculate the second index value for each event and verify the association between the abnormality occurring in the plant 1000 and the event.

[0059] As described in the event management table 630 in Fig. 15, each plant data is also linked to time information (time stamp). Therefore, as will be described in detail later, it is possible to analyze the plant data in chronological order, and to analyze abnormalities in more detail. <Auxiliary Equipment Management Table> An example of the auxiliary equipment management table 640 is shown in Fig. 16. The auxiliary equipment management table 640 is a table for managing the operation states (operation patterns) of multiple auxiliary equipment (pumps, etc.) when there is equipment composed of three auxiliary equipment A, B, and C in the system 1100. The example shown in Fig. 16 shows that when there is equipment composed of three auxiliary equipment A, B, and C, either auxiliary equipment A or B can be operated (only auxiliary equipment A can be operated, or only auxiliary equipment B can be operated), but auxiliary equipment C is in a power lock state and is managed so that it cannot be operated. Of course, various operation patterns of the auxiliary equipment are possible other than those shown in Fig. 16.

[0060] When plant data is measured in the system 1100, each piece of plant data is stored in the plant data management table 610 in association with information indicating which auxiliary is in operation (operation pattern of the auxiliary).

[0061] In this manner, it becomes possible to analyze the plant data separately for each auxiliary operation pattern, which makes it possible to verify the correlation between an abnormality occurring in the plant 1000 and the auxiliary operation pattern. <Functional configuration of the plant management device> Next, the functional configuration of the plant management device 100 will be described with reference to the functional configuration diagram shown in FIG.

[0062] As described above, the plant management device 100 realizes various functions as the plant management device 100 by reading the plant management device control program 700 and various data stored in the storage device 140 into the memory 120 and executing or processing them by the CPU 110.

[0063] Specifically, the plant management device 100 realizes the functions of an influence information storage unit 101, a plant data acquisition unit 102, a second index value calculation unit 103, a third index value calculation unit 104, a fourth index value calculation unit 105, an abnormality information display unit 106, and an alarm output unit 107.

[0064] The influence information storage unit 101 stores influence information (w, v) that indicates susceptibility to influence from an upstream system 1100 that performs control further upstream, for each system 1100 that shares and executes control in the plant 1000. In this embodiment, the influence information storage unit 101 is embodied as the influence information table 600 described above.

[0065] The plant data acquiring unit 102 acquires a plurality of types of plant data periodically measured in each system 1100. The plant data acquiring unit 102 acquires the plant data from the operation monitoring computer 200, associates it with the identification information and event information of the system 1100, the operation pattern of the auxiliary equipment, the current time, and the like, and stores it in the plant data management table 610. The plant data acquiring unit 102 may also store in the plant data management table 610 the plant data that has been subjected to a process for removing outliers and normalization, which will be described later.

[0066] The second index value calculation unit 103 calculates, for each system 1100, for each combination of two types of plant data selected from plant data measured within the same system 1100, a second index value indicating the degree of deviation between a first index value indicating the relationship between two types of plant data obtained from the measurement values ​​of the two types of plant data measured within a specified period of time, and a first reference value serving as the standard for the relationship between these two types of plant data.

[0067] As described above, the first index value is the coordinate value (for example, the coordinate value of the center of gravity of each point on the scatter plot measured in a specified period) of a point determined from each representative value of the two types of plant data measured in the specified period, the first reference value is the coordinate value (for example, the coordinate value of the center of gravity of each point on the scatter plot measured in the reference period) of a point determined from each representative value of the two types of plant data measured in a reference period during which no abnormalities have occurred in the plant 1000, and the second index value can be the Euclidean distance between the coordinate value represented as the first index value and the coordinate value represented as the first reference value.

[0068] In this embodiment, the representative value is an average value, but it may be a median value, a maximum value, a minimum value, etc. The first reference value is set individually for each combination of the plant data. The first reference value may be reviewed at an appropriate timing, such as when inspecting the plant 1000.

[0069] In addition, when selecting a combination of two types of plant data, the second index value calculation unit 103 may select a combination of plant data related to temperature and a combination of plant data related to physical quantities other than temperature. This makes it possible to avoid a combination of plant data that differ greatly in tracking speed (time constant) when the state of the plant 1000 changes. This is because a scatter diagram created using plant data with significantly different tracking speeds will have large variations, making it difficult to determine whether a change in the relationship between these plant data is due to an abnormality occurring in the plant 1000 or due to a difference in the time constant. Conversely, when a scatter diagram is created using plant data related to temperature and a combination of plant data related to physical quantities other than temperature, it can be said that the cause of the change in the relationship between the plant data is likely to be an abnormality in the plant 1000.

[0070] The third index value calculation unit 104 calculates, for each system 1100, the maximum value of the second index values ​​as a third index value indicating the degree of the abnormality occurring in that system 1100. That is, in the case of the plant 1000 shown in FIG 1, the third index value calculation unit 104 calculates the third index value for each of the four systems 1100, system A 1100A to system D 1100D.

[0071] Then, the fourth index value calculation unit 105 uses the third index value and influence information of each system 1100 to calculate a fourth index value for each system 1100, which indicates the degree of influence of the abnormality that each system 1100 receives from the upstream system 1100.

[0072] The following description will be given with reference to Fig. 10. Fig. 10 shows a case where the system A 1100A to the system D 1100D are upstream systems of the system E 1100E.

[0073] 10, x1 is the third index value in the line A 1100A, x2 is the third index value in the line B 1100B, x3 is the third index value in the line C 1100C, and x4 is the third index value in the line D 1100D. D is the third index value in the line E 1100E. θ and Y will be described later.

[0074] Furthermore, w1 and v1 are influence force information representing the susceptibility to influence from system A1100A, w2 and v2 are influence force information representing the susceptibility to influence from system B1100B, w3 and v3 are influence force information representing the susceptibility to influence from system C1100C, and w4 and v4 are influence force information representing the susceptibility to influence from system D1100D.

[0075] Then, the fourth index value calculation unit 105 calculates the fourth index value in the branch E1100E by A=Σ(xi×wi×vi).

[0076] In other words, for each combination of a system 1100 and an upstream system 1100 of the system 1100, the fourth index value calculation unit 105 calculates the product of the third index value of the upstream system 1100 and the influence information corresponding to the combination, and calculates the sum of the products as the fourth index value.

[0077] In this manner, it becomes possible to comprehensively reflect the influence of the upstream system 1100 on the system 1100 in the fourth index value.

[0078] Alternatively, the fourth index value calculation unit 105 may calculate the fourth index value in the system E1100E by A=max(xi×wi×vi).

[0079] In other words, for each combination of a system 1100 and an upstream system 1100 of the system 1100, the fourth index value calculation unit 105 calculates the product of the third index value of the upstream system 1100 and the influence information corresponding to the combination, and calculates the maximum value of these products as the fourth index value.

[0080] In this manner, it becomes possible to reflect in the fourth index value the influence of the upstream system 1100 that has the greatest influence on the system 1100 in question, and therefore it becomes possible to find the transmission path of a highly important abnormality.

[0081] Then, the fourth index value calculation unit 105 calculates fourth index values ​​for the other systems 1100 in the same manner.

[0082] The abnormality information display unit 106 displays the abnormality state in the plant 1000 by displaying a list of the third index value and the fourth index value in each system 1100 determined as described above.

[0083] FIG. 11 shows how the abnormality information display unit 106 displays the abnormality state of the plant 1000.

[0084] 11, the third index value of each system 1100 is described as "deviation." Also, the fourth index value of each system is described as "input." In this manner, the operator can get an overview of the abnormality state of each system 1100, and can more quickly grasp the abnormality state in the plant 1000.

[0085] 11 shows an embodiment in which the squares corresponding to the systems 1100 are arranged in a two-dimensional matrix and the third and fourth index values ​​are displayed in each square, but it is also possible to display the display frames corresponding to the systems 1100 in the form of an interconnection diagram as shown in Fig. 2 and then display the third and fourth index values ​​in each display frame. With this embodiment, the connections between the systems 1100 can also be visually displayed, so that even if the operator does not understand the connections between the systems 1100, the operator can easily understand the state of an abnormality that has occurred in the plant 1000.

[0086] Furthermore, as shown in system J and system S in FIG. 11, for a system 1100 whose third index value exceeds a predetermined second reference value (e.g., 0.1), the abnormality information display unit 106 displays a message indicating that an abnormality has occurred in that system 1100 (for example, by coloring the inside of the square in a predetermined color such as yellow, or by displaying a dot pattern, etc.).

[0087] In this manner, the system 1100 in which an abnormality is occurring can be easily distinguished from the other systems 1100, enabling the operator to quickly grasp the abnormal state in the plant 1000.

[0088] Furthermore, as shown in systems J, S, and U in FIG. 11, the abnormality information display unit 106 further displays, for the system 1100 immediately downstream of the system 1100 whose third index value exceeds the second reference value, a message indicating that an abnormality has occurred in the upstream system 1100 (such as by surrounding the square with a thick frame of a predetermined color such as yellow).

[0089] In this manner, the upstream system 1100 in which an abnormality has occurred can be displayed in an easily identifiable manner, enabling operators to quickly grasp the transmission status of the abnormality in the plant 1000.

[0090] 8, when the value of any of the plant data is outside a predetermined range defined as a normal range, the warning output unit 107 outputs a warning associated with the plant data. The warning output unit 107 outputs, for example, to the output device 160 of the plant management device 100, alarm information (warning) indicating that the value of the plant data is outside the normal range.

[0091] When an alarm has been output for any of the plant data, the abnormality information display unit 106 displays an indication that an alarm has been output for the system 1100 in which the plant data is measured, as shown by system FF in Figure 11 (for example, by displaying the inside of the square in a predetermined color such as red or in inverted black and white).

[0092] In this manner, the system 1100 in which an alarm is occurring can be easily distinguished from the other systems 1100, enabling the operator to quickly grasp the abnormal state of the plant 1000.

[0093] Furthermore, when an alarm has been output for any of the plant data, as shown in system N or system P in FIG. 11, the abnormality information display unit 106 further displays, for the downstream system 1100 of the system 1100 for which the alarm has been output, a display indicating that an alarm has been output in the upstream system 1100 (for example, by surrounding the square with a frame of a predetermined color such as red or by displaying the inside of the square with diagonal lines).

[0094] In this manner, the upstream system 1100 in which an alarm has occurred can be displayed in an easily identifiable manner, enabling operators to quickly grasp the status of the transmission of abnormalities in the plant 1000. ==Process flow== Next, the flow of processing by the plant management system 100 according to this embodiment will be described with reference to the flowchart shown in FIG.

[0095] First, the plant management system 100 acquires a plurality of types of plant data measured during a predetermined period (S1000).

[0096] Then, the plant management system 100 removes outliers and normalizes the plant data (S1010).

[0097] Outliers can be removed, for example, by removing measurement values ​​that deviate from ±σ (σ is the standard deviation) or ±2σ. Normalization can be performed by converting the measurement values ​​of the plant data into relative values ​​so that the lower limit of the measurement range of the measuring instrument is 0 and the upper limit is 1.

[0098] Then, the plant management device 100 performs the processes from S1020 to S1050 for each system 1100.

[0099] First, the plant management device 100 creates a scatter diagram as shown in FIG. 9 for each combination of two types of plant data selected from plant data measured in the same system 1100 (S1020).

[0100] Then, for each of these combinations, the plant management device 100 calculates, as the deviation degree, the Euclidean distance (second index value) between the coordinate value (first index value) of a point obtained from each representative value of the two types of plant data and the coordinate value (first reference value) of a point obtained from each representative value of the two types of plant data measured during a reference period in which no abnormalities occurred in the plant 1000 (S1030).

[0101] Then, the plant management device 100 calculates the maximum value of the second index values ​​obtained for each combination as a third index value indicating the degree of the abnormality occurring in the system 1100 (S1040).

[0102] Furthermore, the plant management apparatus 100 calculates a fourth index value that indicates the degree of influence of the abnormality from the upstream system 1100, using the third index value obtained for the upstream system 1100 and the influence information (S1050).

[0103] Then, the plant management device 100 displays a list of the third index value and the fourth index value in each system 1100, thereby displaying the abnormality state in the plant 1000 (S1060).

[0104] In the above manner, the operator can more quickly grasp the state of an abnormality that has occurred in the plant 1000.

[0105] ==Second embodiment== Next, a second embodiment of the present invention will be described. The second embodiment describes the first embodiment more specifically and in detail, and also includes matters that supplement or expand the first embodiment. Hereinafter, the second embodiment will be described with appropriate reference to the drawings described in the first embodiment, while appropriately omitting explanations that overlap with the first embodiment.

[0106] In this embodiment, a plant anomaly monitoring system 100 (plant management device 100) and a plant anomaly monitoring method are provided that enable operators to check for signs of failure in plant data during normal plant operation and easily grasp visually the impact that the signs of failure will have on the entire plant 1000.

[0107] Specifically, the present invention proposes a plant anomaly monitoring system 100 (plant management device 100) and a plant anomaly monitoring method, which are characterized by a mechanism for detecting signs of failure by calculating the data cluster deviation between learning data and operating data registered in advance in each system 1100 of the plant 1000, a feature of preparing a correlation map table 620 based on the operating pattern of the auxiliary equipment, a system neural network with synaptic input / output that selects a data map with a strong correlation and calculates the effect of a deviation between learning data and operating data caused by a failure or the like in the system 1100 or an influence from another system 1100 or the effect of a failure in the system 1100 on other related systems 1100, and a mechanism for evaluating the effect of the failure effect on the plant 1000.

[0108] In a plant 1000 such as a nuclear power plant, during start-up and shutdown operations and normal operation, the plant operation data is periodically checked on indicators and other devices installed in a central control room or on-site control panel to monitor for any abnormalities in the data. The frequency of checking is at regular intervals, and the output of indicators and controllers and parameters output by the operation monitoring computer 200 are checked on trend graphs to check for any trends of change, etc. These are recognized and interpreted by the operators to strengthen data monitoring and conduct on-site patrols to detect abnormalities early.

[0109] Conventionally, the detection of signs of failure was based on the monitoring of parameters by operators and the recognition of abnormalities. In recent years, AI-based support systems using large-scale machine learning, such as plant sign management systems, multivariate time series data analyzers, plant abnormality diagnosis devices and plant abnormality diagnosis systems, have been put into practical use as systems to support the monitoring of such plant data (hereinafter, plant data refers to both event data and process data). These systems detect abnormalities in plants 1000 by measuring the correlation strength of data and the degree of change with sampled data through correlation processing and multivariate analysis of large amounts of plant data.

[0110] This data processing using machine learning has excellent characteristics, such as being more real-time than the process of operators checking the data, interpreting it, and inferring the cause, making it easier to compare with past data, and being able to make quantitative judgments.

[0111] For example, the system is able to detect signs of parameter changes before an operator visually recognizes the change in a parameter and senses an abnormality.

[0112] Therefore, if a sign of a failure can be detected before a system alarm is issued, it will not only contribute to early detection of anomalies but also contribute to stable and safe operation of the plant 1000.

[0113] The following definitions apply to terms used in this system 100.

[0114] In this proposal, "plant data" is divided into three types: analog data, which is input signals from measuring instruments; digital data, such as the ON / OFF status of equipment; and event data, such as calculations within a computer. Each type is explained using terms appropriate for its purpose.

[0115] The process data (hereinafter referred to as "analog data") output from the operation monitoring computer 200 in response to input signals from measuring instruments indicates the output of a quantized analog signal.

[0116] Specifically, these include signals expressed by conductivity, density, dew point, differential pressure, flow rate, liquid level, humidity, neutron flux, concentration, pressure, valve position, temperature, time, vibration, temperature difference, mass, current, potential difference, frequency, power, reactive power, etc.

[0117] Other digital data of the operation monitoring computer 200 (hereinafter referred to as "digital data") refers to the output of binary signalized device ON / OFF signals, power lock status ON / OFF signals, and alarm setting values ​​which are constants.

[0118] Specifically, it is a collection of signals related to alarms and equipment operation of the plant 1000, and is represented by a group of signals related to operations, such as digital signal system (ON, OFF signals) alarm issuance, starting and stopping of pumps and fans, opening and closing of valves, etc.

[0119] Digital data relating to other devices (hereinafter referred to as "event data") refers to the output of the calculation results of the operation monitoring computer 200 and minor warnings and alerts indicated by computer alarms.

[0120] Specifically, this includes plant interlock operation events such as reactor scram (emergency shutdown), turbine trip, generator trip, etc., status alarms indicating that an analog signal has entered an abnormal area, or status notifications such as a surveillance state for a non-required core cooling system, etc.

[0121] It is expected that operator support systems using artificial intelligence will become more widespread in power plant operations in the future. AI-based predictive management systems have the following features: 1.Causal explanation of AI judgment results In recent years, one of the challenges in using artificial intelligence is that the judgment process using machine learning, especially deep learning, has become a black box, making it difficult to explain the judgment results. Therefore, it is necessary to build a model that makes the judgment process easy for operators to understand.

[0122] In addition, the judgment results require interpretation of the data by the operator. Because interpretation depends on the operator's ability, measures to reduce interpretation as much as possible are also necessary.

[0123] When an abnormality occurs in a large-scale plant 1000, it may be difficult to clearly identify the cause and immediately grasp the extent of the impact on related systems. Furthermore, since it is difficult to know the extent of the impact on other systems 1100, there is a concern that it may be difficult to determine whether the failure is limited or will have a serious impact on the plant 1000.

[0124] Therefore, it is necessary to quickly understand not only the judgment results, but also the occurrence of an abnormality, identification of the origin of a fault, and the degree of impact on the entire plant 1000. 2.Generating highly accurate learning data from driving data It is necessary to prepare learning data through preprocessing and to deal with outliers that occur when collecting data on actual equipment. The presence of outliers becomes noise, especially when calculating approximate values ​​using linear regression or multiple regression after collecting correlation data. This affects the quality of the learning data used to distinguish abnormal data. In particular, if there is a lot of noise when calculating the center point (first index value) of the correlation data cluster, as described below, this may affect the judgment. 3.Validity of learning data due to changes in rotating equipment operating patterns System 1100 of plant 1000 often has multiple rotating devices and devices that generate fluids (such as steam ejectors). For example, if there are three similar rotating devices, one or two are usually in operation at all times, and the third is a standby device. Auxiliary devices are switched periodically, and various operation patterns are possible. For this reason, learning data for one operation pattern may not be sufficient.

[0125] For example, if there are three rotating devices as mentioned earlier, and there are patterns of one device operating, two devices operating in pairs, and all three devices operating, there are a total of seven patterns. The process data of system 1100 may change slightly depending on the operating patterns of the rotating devices and their individual characteristics. If learning data is created without properly discriminating between these operating patterns, the range of the data cluster will become large, resulting in a state similar to overlearning. This may affect the accuracy of anomaly detection.

[0126] Conversely, in the case of a system 1100 such as an emergency core cooling system that is normally stopped and operated during periodic testing or when an abnormality occurs, operation data is collected during periodic testing to confirm the function performance. Therefore, continuous data during operation is difficult to collect. Therefore, the sample size of the learning data tends to be small.

[0127] We have described above the characteristics of processing plant data using machine learning, but as shown below, plant data also has characteristics that make it easy to organize as supervised data.

[0128] Plant data (analog data, digital data, event data) output from each measuring instrument and controller in the power plant is input to and output from the operation monitoring computer 200, and the data is output to an operation display such as a display or touch panel, which is a human interface.

[0129] When digital data or event data is output, it is highly likely that the analog data is also affected, so it can be said that the analog data is related to the digital data and event data. In other words, this indicates that the analog data can be labeled by the digital data and event data.

[0130] Generally, in machine learning, there is a supervised learning method that combines analog data, digital data that is a correct label, and event data, but the combination of analog data, digital data, and event data by the operation monitoring computer 200 can also be expressed as a combination of a target variable and an explanatory variable. For this reason, it can be seen that it is easy to use as highly accurate supervised data for comparison and discrimination with abnormal data.

[0131] In this embodiment, a plant anomaly monitoring system 100 and a plant anomaly monitoring method that take these characteristics into consideration are proposed. 1.Causal explanation of AI judgment results (1) Creating a correlation map based on phylogenetic categories One solution to the difficulty of explaining the results of machine learning as a black box is to improve the explainability of the causal relationships of the plant system. As a solution to this problem, the correlation set of parameters (plant data) is made the correlation between parameters for system categorization. This is because the behavior of equipment in the system 1100 is directly affected by the operating state of the rotating equipment in the system 1100, which in turn affects the indications of the measuring instruments, making it easier to explain the causal relationships. This is because the rotating equipment is equipment that circulates fluid within the system 1100, and so there is a tendency for significant differences in parameters to appear between the operating state and the stopped state of the rotating equipment. Note that system categorization here refers to grouping together equipment with the same system number.

[0132] If a correlation with process data of another system 1100 that is known to be correlated by humans through knowledge and experience is selected as in the conventional method (for example, condenser vacuum and seawater temperature), humans can easily understand and explain it, but the correlation is limited to the range that humans can predict, so it is difficult to contribute to the discovery of hidden faults. In addition, the correlation combination of all process data without categorizing the systems increases the calculation cost. In addition, even if the logically incorrect combination that is not related to each other, such as a false correlation, is large, there is a risk of erroneously determining that there is a correlation. However, in this embodiment, improvement is made by dividing the process data by system and detecting them using a brute-force correlation map.

[0133] Furthermore, the correlation map configures a system by utilizing relationships with strong correlations, as will be described later. (2) Representation of causal relationships using a system neural network model When correlation is taken between categories within a system 1100 and an abnormality is identified, in order to determine the degree of influence on other systems 1100, a means for expressing the connections with the other systems 1100 is required.

[0134] In this embodiment, a neural network (multi-layered single-phase perceptron model), which is a type of machine learning model, is adopted to express the relationship between the systems 1100, and the influence between the systems 1100 is expressed as a synaptic input / output value in terms of the distance (deviation) between the center points of the learning data cluster and the operation data cluster of the correlation map, and how a failure in the system 1100 affects the other system 1100 is quantitatively expressed. Similarly, how an influence from the other system 1100 affects the system 1100 is quantitatively expressed. Specifically, the systems 1100 are connected by fluids (pipe connections) or by electrical signals. These are hereinafter referred to as a system neural network. Details will be described separately. 2. Method for generating highly accurate training data for driving data In order to create highly accurate learning data, the process data is divided by a full-scale value (e.g., 0 to 1000 mm) from the software measuring instrument full-scale database, the data is normalized, and it is converted into a value between 0 and 1.

[0135] The data is sampled at regular intervals, for example, every minute, during normal operating conditions when no event data is occurring in the system 1100. The preprocessed data is assumed to be normally distributed, and after accumulating the actual machine data, an approximate formula is found using linear regression or multiple regression, and outlier filtering is performed to eliminate data outside the standard deviation of ±1σ or ±2σ, and the data distribution is organized to become learning data. As a result, the learning data clusters are data clusters organized within the Mahalanobis distance.

[0136] The training data is subjected to clustering processing, and the center points of the clusters are subjected to non-hierarchical clustering using the k-means method.

[0137] The k-means method is usually used for unsupervised data clustering. In this method, data with the same correct answer label is expressed as a single cluster, so there is no need to perform calculations to separate into multiple clusters, and the only process is to calculate the center point of the data cluster classified by the correct answer label. Therefore, the number of clusters is specified as 1. Once clustering is complete, the center point of the cluster is determined.

[0138] Once the center point has been determined, even if additional samples of actual data are added to the training data, the degree of anomaly in the actual data can be quantitatively grasped by calculating how far the additional samples are from the cluster center point of the training data (deviation: Euclidean distance between center points) while keeping the "cluster center point" of the training data fixed.

[0139] In addition, for systems that are basically only started up during periodic tests, such as emergency core cooling systems, data will be measured during operation. 3.Improvement of validity of learning data by changing rotating equipment operation patterns The learning data is based on the operation patterns of the rotating equipment, and in the case of the three pumps mentioned above, there are seven patterns, but pre-processed normalized data for each operation pattern is accumulated, and the data processed by filtering is registered as learning data. Therefore, a total of seven correlation data map tables (correlation map table 620) are prepared according to the operation patterns, and sampling data is stored on the correlation map of the operating unit.

[0140] In addition, in the cells in the correlation data map, the correlation data are represented on the x and y axes, and the appearance time of each data is represented by the number of data samples.

[0141] The arranged data is clustered based on the correct label information as described above, and the cluster centers arranged in two-dimensional space are calculated. The Euclidean distance between the cluster centers of the training data and the sampled cluster centers is measured in advance.

[0142] The plant data predictive management system 100 implements the above improvements, and the system processing will be described in detail below.

[0143] A plant data predictive management system 100 (plant management device 100) according to this embodiment is a system that receives analog data and digital data from an operation monitoring computer 200, monitors predictive data, and performs calculations.

[0144] This system 100 utilizes software data output from an operation monitoring computer 200 which inputs and outputs plant data.

[0145] The hardware configuration of this system 100 has a CPU (central processing unit) 110 and a ROM (read only memory) (corresponding to the storage device 140), an operating system OS is read from the ROM and deployed on the main memory (corresponding to the memory 120) to start up the OS, and under the control of the OS, an application software program (a processing module, corresponding to the plant management device control program 700) is read from the ROM and a GUI (Graphical User Interface) function and various processes are executed. Commands are input using input / output devices such as a display, a key device, a voice input processing device, and a mouse.

[0146] Also, a network device (corresponding to the communication device 130) is provided for inputting data from the operation monitoring computer 200. Although the number is limited, the load is large because the data of the plant 1000 is handled in real time, so the system 100 may adopt multiple CPUs compatible with multithreading, may have a parallel arrangement of multiple GPUs (graphics processing units), or may have an array structure of multiple PC units. The system 100 may also be a cloud computer or a virtual computer.

[0147] This system 100 has the functions of predictive management by processing correlation map data within the system 1100 and predictive impact management by processing a system neural network that expresses the impact of the system 1100, broadly speaking. 1. Data sampling with correlation map data processing (1) A correlation data map (correlation map table 620) is generated for each system 1100 based on the analog data from the operation monitoring computer 200. (2) Two types of correlation data maps are provided: non-temperature and temperature. (3) Based on the digital data from the operation monitoring computer 200, an operation pattern map within the system is generated, which represents the ON / OFF and power lock states of the rotating equipment within the system. (4) Non-temperature system correlation data maps and temperature system correlation data maps are generated for the number of system operation pattern combinations. For example, if there are three rotating machines, seven non-temperature system correlation data map tables and seven temperature system correlation data map tables are prepared to match their operation patterns, for a total of 14 tables. Here, the operation pattern is read from the system operation pattern matrix, and data sampling is performed on one of the seven tables for each.

[0148] The reason for dividing the correlation data map into non-temperature and temperature systems is that non-temperature system parameters have almost no time constant and react instantly, while temperature system parameters change with a thermal time constant, so the temperature parameters do not stabilize until a certain amount of time has passed. Therefore, if the correlation between the non-temperature system and the temperature system is taken, the correlation data will not stabilize due to the influence of the time constant, and the cluster range may become large. (5) Based on the digital data from the operation monitoring computer 200, an alarm setting map (specified by the ANN line shown in FIG. 7) corresponding to each analog data is generated so as to be linked to the non-temperature system and temperature system maps.

[0149] The alarm map is a threshold of the correlation data map and functions as a decision tree that is training data. (6) Data is input into non-temperature and temperature system maps associated with the rotating equipment operation pattern determined by the rotating equipment ON / OFF map. 2. Generate training data for each correlation map When data from the correlation map data processing in 1 above is accumulated, data with strong correlations become a set close to positive or negative linear regression or nonlinear regression approximation. Here, training data is generated by keeping data within the standard deviation of ±σ or ±2σ from the approximation curve of the training data and excluding the rest as outliers. The center point of the training data cluster is also found. (1) Plant data sampled in advance (during a reference period) is accumulated as learning data, and data clusters are generated by performing outlier filtering to retain data with a standard deviation of ±σ or ±2σ and exclude the rest as outliers. Clustering and calculation of the data center point are performed. (2) The correlation coefficient of the learning data clusters arranged in the cells in the correlation map is calculated, and the cells with low correlation are not calculated. (3) The learning data clusters placed in the correlation map cells are stored, and subsequently sampled analog data is registered in the correlation map. Data without occurrence of systematic event data and labeling data when a systematic event occurs are sampled, and clustering and calculation of the data center point are performed. (4) In order to obtain the deviation distance between the learning data center point in the correlation map cell and the driving data cluster center point, the Euclidean distance and Mahalanobis distance between the two points are calculated. (5) Calculate the maximum deviation distance (Euclidean distance) found for each cell. (6) Based on the maximum deviation of the correlation map cells, the maximum value is displayed in a software-based systematic neuron heat map to visually represent the relative degree of deviation of the driving data from the learning data. For example, normality may be represented by green, minor effects and abnormalities by yellow, and warnings by red. In addition, deviations of 0.1 or less are represented by green, deviations over 0.1 are represented by yellow, and warnings are represented by red.

[0150] In addition, when the displayed map is clicked, for cells with high deviation in the correlation map, correlation parameters, correlations, and labeling event names may be displayed, and a scatter diagram may be displayed. 3. Predictive management using neural network processing to represent the impact of system 1100 (1) The maximum value (maximum deviation) is calculated based on the Euclidean distance between the center point of the learning data in the correlation map cell calculated for each system 1100 and the center point of the operation data cluster.

[0151] The degree of deviation is calculated by evaluating the center point of the data cluster when the latest event occurred, or the center point of the data cluster from the latest data sampling start time when no event occurred.

[0152] In other words, the deviation is calculated each time an event occurs, or each time the latest data cluster occurs during a time when no event occurs. (2) An input synapse signal value A from the other system 1100 and an output synapse signal value Y from the own system 1100 are calculated.

[0153] Input synapse signal value A = x1w1v1 + x2w2v2 + x3w3v3 + x4w4v4 (Equation 1) Output synaptic signal value Y = D ≦ 1 ∨(or) θ = 1 (Equation 2) Here, A: input synaptic signal value. The degree of influence from a single or multiple upstream systems 1100.

[0154] x: Deviation signal (0.1 to 1) of the input system (upstream system) 1100. Note that values ​​below 0.1 are treated as 0.

[0155] w: Input synapse weighting value from each system 1100 (the squared ratio (a^2 / A^2) of the other system supply pipe diameter a to the rotating equipment outlet pipe diameter A. For example, 1 when all rotating equipment in the input system 1100 stops, and for example, 0.1 when all rotating equipment in the input system 1100 does not stop). By setting the weighting value w, even if an abnormality occurs in the input system 1100, the degree of impact on the output system 1100 changes due to, for example, differences in pipe diameter.

[0156] v: Valve open / close signal. Specifically, if there is a motorized shutoff valve in the supply system, the open / close state determines whether there is a supply or not, so if the motorized valve is open, a signal of 1 is input, and if it is closed, a signal of 0 is input. Note that in the case of a manual valve, the open / close state is determined at the system configuration stage, so if the manual valve is open, a signal of 1 is input, and if it is closed, a signal of 0 is input in advance. Note that in the case of an electrical signal, the ON / OFF state of a switch can also be used.

[0157] D: Deviation. The deviation is calculated from the distance (absolute deviation value) between the center point of the learning data cluster in the system and the center point of the operation data cluster in the correlation data map, and the maximum deviation in the correlation data map is set to the same value. Note that since the data is normalized, the theoretical maximum deviation distance is √2.

[0158] θ: Alarm occurrence line (ANN line) threshold. If the data of the correlation data cell of the cluster that occurred is below the alarm threshold, it is 0. If it is above the alarm threshold, it is 1. Y: Synapse output signal. Outputs 1 when deviation D or alarm occurs. (3) When the calculated synapse output value Y exceeds 0.1, the influence of the system 1100 in question is substituted into the input synapse signal x input to the other system 1100 via the operating state of the piping or rotating equipment connected to the other system 1100, and it can be expressed that the influence on the other system 1100 is exerted by the integral of the weight w. (4) When the calculated synapse output value Y is 0 to 0.1 or less, it can be expressed that the input system 1100 has no effect on the system 1100 itself, and has no effect on the system 1100 further downstream. (5) When the neuron network is an open system, it can be determined that the influence of the most upstream side of the system affects the downstream side. On the other hand, when it is a closed system, the influence is propagated within each system 1100, but for example, the influence is automatically absorbed by the control system, and the influence of the abnormality is circulated until the parameter change is balanced. (6) The fault origin is when the deviation exceeds 0.1 or hits the alarm line. This is selected as the fault cause origin and displayed in the cell. (7) When the system neuron heat map color shifts to the abnormal side, the plant system or operator operates to converge the parameter changes in the plant 1000. Specifically, when the plant data converges to a normal state due to the operation of the control system, the operation of interlocks, or manual intervention by the operator, it approaches the learning data cluster, and the system neuron heat map abnormality display color and deviation degree of each system 1100 gradually change to green, visually indicating that the abnormality of the entire plant has ended.

[0159] Here, we will explain the system neuron heat map frame display. In the system neuron heat map display, the discrepancy alone exceeds 0.1 and is displayed in yellow (in Figure 11, it is displayed as a dot).

[0160] If the deviation D is within the Mahalanobis distance of the center point from the training data, the deviation is 0 and the lineage neuron heat map is green. The downstream lineage frame is not displayed.

[0161] If the discrepancy D is less than 0.1, the lineage neuron heatmap is green. No downstream lineage frame is displayed.

[0162] If the discrepancy D alone exceeds 0.1, the lineage neuron heat map will be displayed in yellow. The heat map will be displayed as the starting point. The downstream side frame will be displayed in a yellow frame (bold frame in Figure 11).

[0163] When an alarm occurs in a system, the system neuron heat map is displayed in red (black and white inverted in Figure 11). The system neuron heat map is displayed as the starting point. The downstream system frame is displayed in a red frame (diagonal line in Figure 11).

[0164] For systems that are subject to frame display, the starting point will not be displayed.

[0165] This makes it possible to visually grasp the extent to which the abnormality in the starting system 1100 has spread, and to confirm whether the abnormality in that system 1100 has been contained within that system 1100 or has spread to other systems 1100.

[0166] In addition, the following describes the expected behavior when a measuring instrument in the system 1100 malfunctions and goes out of the measurement range (underscale, overscale). If the deviation in the system 1100 becomes 1 due to a single failure of these instruments, the downstream synapse output also outputs the product value of the deviation, weight value, and valve state (open, set to 1 here) to the downstream system 1100. In the case of a single failure of an instrument without an interlock, since there is no change in the process value of the downstream system 1100, it is determined that there is no change in the deviation value number of the downstream system 1100. In that case, the operator can compare the input synapse value with the deviation value and confirm that there is no influence from the upstream system 1100, and can determine that there is a single failure of the instrument.

[0167] In addition, in the case of a measuring instrument with an interlock, the device operation will occur due to the interlock, and there is a possibility that some effect will be caused on the state of the system 1100.

[0168] The synapse input value A is a disturbance value given to the downstream system 1100, and the synapse output value Y can be expressed as the response of the system 1100 to the disturbance. However, since the disturbance may be an actual change in the process value of the actual system 1100 or an apparent change due to an instrument failure, it is difficult to model the output response by multiplying the sum of the input and the initial value by a transfer function. Therefore, quantitative changes and the behavior of the entire plant can be visually grasped by changing the related system neuron heat map and expanding the frame display.

[0169] Next, a configuration diagram showing the overall configuration of the present system 100 will be described with reference to FIG.

[0170] The configuration of the system 100 includes the following functions for each line 1100.

[0171] As a means for inputting data from the plant measuring instruments to the operation monitoring computer 200, the measuring instrument data, which is an analog electric signal, is AD converted and then input to the operation monitoring computer 200.

[0172] The operation monitoring computer 200 outputs an event data input section (digital data), a power on / off data input section, a device ON / OFF data input section (digital data), and alarm data (digital data).

[0173] The system A learning data creation unit (corresponding to the plant data acquisition unit 102 and the second index value calculation unit 103) is provided with a measuring instrument full-scale database (corresponding to the plant data management table 610), a data pre-processing (normalization) unit, a data pre-processing (outlier processing) unit, a learning data cluster correlation coefficient calculation processing unit, a learning data clustering process (cluster center point calculation), and a machine learning data creation unit (supervised data).

[0174] The data pre-processing (normalization) unit outputs system parameter data (non-temperature system) and system parameter data (temperature system).

[0175] The neuron unit of system A is provided with a rotating equipment ON / OFF map (FIG. 16), a system statistical data correlation map table, a system temperature correlation map table, and an alarm data correlation map table (corresponding to the correlation map table 620).

[0176] The alarm data correlation map table is provided to overlay an alarm line on the system statistical data correlation map table and the system temperature correlation map table, which have the same data array.

[0177] The rotating equipment ON / OFF map is provided to allow the selection of correlation maps created for each auxiliary operation pattern in the system statistical data correlation map table and the system temperature correlation map table.

[0178] The system state calculation unit (corresponding to the influence information storage unit 101, the third index value calculation unit 104, and the fourth index value calculation unit 105) calculates the center point calculated by the operation data clustering process (cluster center point calculation) input in the system statistical data correlation map table and the system temperature correlation map table selected in accordance with the auxiliary operation state, and the center point obtained by the learning data clustering process (cluster center point calculation), and determines the output value Y of the other system output synapse by calculating the Euclidean distance with the maximum center point deviation D in the cell of the map and the Mahalanobis distance to confirm whether or not the center point of the operation data cluster exists within the learning data cluster range.

[0179] When an event occurs, the driving data is sampled as labeling data, and each time the center point of the data cluster of the driving data for that event is calculated using the k-means method, the Euclidean distance between the center point and the learning data cluster is calculated, and the maximum value is output after comparing with other cells.

[0180] Sampling is performed each time an event occurs, the center point of the data cluster is calculated, and when the next event occurs, the calculation of the center point is repeated in the same manner, thereby constantly calculating the Euclidean distance from the training data.

[0181] The system neuron heat map display unit (corresponding to the anomaly information display unit 106 and the alarm output unit 107) displays in yellow if the central point deviation in the correlation map exceeds 0.1, and displays in red if it hits the alarm line. If the deviation is 0.1 or less, it displays in green.

[0182] Similarly, the input synapse receives the product of the maximum deviation x of the correlation map data, the weight w, and the valve opening / closing v as input from the other system 1100, and the influence is input to the system A neuron.

[0183] The output synapse substitutes the maximum deviation Y of the correlation map data for x as the output from system A, outputs the product of the weight w and the valve opening / closing v, and outputs the degree of influence to neurons of other systems.

[0184] Learning data is accumulated when data is accumulated during normal plant operation when no events have occurred in the system. At that time, outliers are detected from the accumulated data cluster set and removed. The outlier filter detects data that is not within a 1σ or 2σ range as outliers and removes them. For data that has undergone outlier processing, the center point of the data cluster is calculated using the k-means method, and the deviation value of the subsequent data cluster is calculated.

[0185] As described above, the learning data is easily influenced by the performance specific to the rotating equipment because the parameters in the system change due to the influence of the fluid in the system 1100. This is because the tendency of the parameters may change between the operating state when the rotating equipment is new and the time when aging occurs. For example, in the case of a pump, a decrease in head and a decrease in flow rate may be considered.

[0186] When considering that changes over time in rotating equipment are a sign of failure, it is convenient to collect learning data from when the equipment is new in order to capture signs of subsequent parameter changes.

[0187] In addition, when rotating equipment is replaced, parameters may naturally change, so the previous learning data will be discarded and the learning data will be reconstructed based on the operating data of the newly replaced equipment, making it possible to always update to the latest state. Specifically, a mechanism will be provided for exchanging learning data on the system side.

[0188] When rotating equipment deteriorates over time or when plant equipment is inspected for regular inspections, the correlation data may change as the initial performance deviates. If the learning data is old, the deviation of the correlation data cluster may become large in some cases, and the data may no longer be appropriate. Therefore, after startup at each regular inspection, when normal operation begins, the learning data may be updated and the correlation map data may be measured based on the new learning data.

[0189] The sampling interval for the learning data is about one minute, but in order to accumulate data and ensure accuracy, the data may be enriched by sampling every second, for example.

[0190] The learning data is accumulated for a certain period of time after the plant is started, but may be accumulated for a period of time until all patterns of auxiliary machine switching have been completed. Furthermore, past data may be temporarily input, and the learning data may be updated after the learning data has been accumulated for a certain period of time.

[0191] Next, the system neural network (connection diagram) shown in FIG. 14 will be described.

[0192] The connections of the systems 1100 are illustrated assuming a power plant system, and the network of each system 1100 is illustrated as an example. The power plant system network is an open system, and some systems have a reactor system, a condenser system, a feed-condensate system, a heater-drain system, etc., connected in a closed cycle.

[0193] With the exception of the fuel system (oil, coal, nuclear fuel, etc.), the main disturbance factors for a power plant system are the atmospheric system, which is the final heat sink for the power plant, the seawater system, and the external power supply system.

[0194] For this reason, seawater temperature, outside air temperature, atmospheric pressure, external power supply voltage, and frequency are listed on the left side as the most upstream disturbance factors, and the system 1100 that is directly affected by the disturbance factors is the second-stage system 1100. From the third stage onwards, systems 1100 connected from the upstream systems 1100 are connected, and finally a power plant system network is formed in which all systems 1100 are connected via utility systems 1100 such as water, steam, air, and on-site power supplies.

[0195] The systems 1100 are connected by pipes and power sources. The properties of these fluids supplied from each system 1100 depend on the performance of the rotating equipment, the operating / stopped state, and the operating pattern. The properties (temperature, pressure, flow rate, etc.) of the fluid distributed to other systems 1100 based on the valve state of each system 1100 change depending on the operating state of the supplying system 1100, and the effect is determined by the diameter and pressure of the pipes.

[0196] Therefore, it is advisable to determine the input weight w of the synapse from the piping and instrumentation diagram. To determine the degree of influence on the other system 1100, for example, the diameter of the main pipe of this system 1100 can be compared with the diameter of the supply pipe to the other system 1100, and the weight value can be determined based on what percentage of the main pipe it occupies. If a system 1100 supplies all of its flow to the other system 1100, the discharge flow rate of the rotating equipment of that system 1100 is almost equal to the system flow rate.

[0197] However, when the diameter of the supply pipe is greater than the diameter of the main pipe, a shutoff valve or a pressure reducing valve is provided to prevent an excessive flow rate, and the actual impact on the other system 1100 is controlled by the system configuration, such as the valve state. For this reason, a valve opening / closing input is provided as v in the synapse input to distinguish whether or not there is an impact on the subsequent stage. Also, when an automatic adjustment valve is used, the impact 1100 of the other system is influenced by the opening of the adjustment valve. For this reason, the opening of the automatic adjustment valve, 0 to 100%, may be normalized to the range of 0 to 1 and input as v in the synapse input to represent the impact on the subsequent stage.

[0198] For example, even if the system neuron heat map of the own system 1100 is displayed in yellow, it is conceivable that the downstream system 1100 may remain displayed in green because no fluid is being supplied or the impact is minor.

[0199] Next, the correlation data map and alarm lines will be described with reference to FIG.

[0200] The correlation data map is divided into two systems, a non-temperature system and a temperature system in the system 1100, and correlations are calculated in a round-robin manner for the parameters in each system. Each correlation in the correlation map is called a cell, and a learning data cluster created based on past operating data is input into each cell.

[0201] The correlation data is collected from cells having a correlation coefficient other than 0 to ±0.2, excluding cells that have been transformed into a matrix and cells with the same parameters arranged on a diagonal line. The learning data input in advance into the cells is basically data in which no events have occurred in the system 1100 and is composed of plain data clusters that have been processed for outliers, and if the data is strongly correlated, the data clusters of the learning data and the operation data are easily separated, making it easy to calculate the cluster center Euclidean distance.

[0202] The alarm line (ANN line) also functions as a decision tree for the data, and data that exceeds the alarm line is immediately determined to be abnormal. In addition, when the normalized data due to abnormal output from a measuring instrument is below 0 or above 1, it is also determined to be outside the range and is also determined to be abnormal by the decision tree.

[0203] Next, synaptic inputs and outputs of the system neural network will be described with reference to FIG.

[0204] This model neural network is a composite model that combines a connected single-layer perceptron model and a system neuron's degree of self-firing (synaptic output) represented by the degree of deviation of the center points of data clusters in a correlation map.

[0205] This makes it possible to express not only synaptic input from the other system 1100, but also the impact on the plant 1000 by synaptic output within the system 1100. For example, even if synaptic input is made, if there is a slight change or no impact in the correlation deviation degree within the system neuron, it can be determined that there is a failure in the upstream system 1100 alone, and if synaptic input and synaptic output occur simultaneously, it indicates that the failure of the other system 1100 has propagated to the system 1100 itself and has propagated to the downstream system 1100.

[0206] In this proposal, the connections between neurons are expressed by the above-mentioned (Equation 1), and the deviation D is expressed by the Euclidean distance between the data cluster centers of the correlation map cells. Also, when a certain alarm line is determined by the decision tree, it is output.

[0207] The weight value w of the system synapse is fixed because it is determined by the piping design of the system 1100. If changes occur due to modifications to the system 1100, settings for the synapse path and the weight value w may be added separately to reflect the changes.

[0208] The synapse output Y is set as the synapse output with the deviation value or alarm value set to 1. Basically, it is possible that the degree of influence of the failure in the system 1100 that is the origin propagates to the downstream systems 1100 becomes weaker.

[0209] Although the deviation D is set to 0.1 as the threshold for no synapse output, the system designer may change the value as a hyperparameter. If the deviation D is always output at 0.1 or more, the accuracy of the learning data may be poor, or there may be many outliers due to a lack of sampling data when an event occurs. For this reason, it is also possible to set and tune the deviation output threshold for each system.

[0210] Next, the Euclidean distance between the learning data cluster and the event occurrence data cluster in the correlation map cell and the cluster center point will be described with reference to FIG.

[0211] The data clusters indicated by circles (white circles) in the cells are learning data, and the data clusters indicated by triangles (white triangles) and squares (white squares) are driving data with event data labeled.

[0212] As explained above, the center points (black circles, black triangles, black squares) of each data cluster are calculated using the k-means method, and if the Euclidean distance exceeds 1, it is output to the system output synapse.

[0213] During normal operation, no warnings are issued in principle, and data is distributed inside the warning line. When the labeled operating data overlaps with the learning data, it is normal, but when it is outside the learning data cluster and distributed inside the warning line, there is a possibility of a minor abnormality. Therefore, the Euclidean distance between the cluster center points is used to quantitatively indicate how far the data deviates from the learning data, which is ideal data.

[0214] In addition, data from the alarm map shown in FIG. 7 is reflected in the correlation map cells to form a decision tree.

[0215] Note that an event may occur multiple times. Each event is labeled with the event and time data (timestamp), so the degree of discrepancy is calculated using the data cluster at the time of the most recent event occurrence and reflected in the systematic neuron heat map display.

[0216] If there is a driving data cluster within the learning data cluster, even if the distance between the center points is taken, if the Euclidean distance is the same as when the learning data cluster and a driving data cluster that is completely separated from it are placed, it will be difficult to identify anomalies. Since the learning data has a 1σ or 2σ data distribution that excludes outliers, the Euclidean distance and the Mahalanobis distance are calculated simultaneously.

[0217] Therefore, if the center point of the driving data cluster is located within the Mahalanobis distance of the learning data, it is determined that there is no abnormality and is displayed in green, for example, as 0. If it is greater than the Mahalanobis distance, it is switched to Euclidean distance and displayed in green or yellow according to the degree of deviation.

[0218] Next, the data sampling trigger based on the occurrence or non-occurrence of an event in the correlation data map table will be described with reference to FIG.

[0219] Figure 15 is a matrix for sampling data in which event data has been labeled onto the driving data described in Figure 9. Basically, the data is divided into labeled data and unlabeled data, and it is possible to determine what factor caused each data.

[0220] Moreover, data sampling is performed when the event is ON, and ends when the event is OFF, and data sampling of non-labeled data is started. Data sampling may be performed in one-minute units, for example.

[0221] In addition, by attaching time data (time stamp) (yyyy / mm / dd / mm / ss) as additional labeling to both labeled and unlabeled data, data clusters can be separated even if data occurs under the same data conditions. It also makes it possible to track sampling time series changes in data clusters, which can be useful for detailed investigations later. Even if different intra-system events occur at the same time, the data can be easily separated by sampling each event.

[0222] Next, a matrix diagram of an auxiliary operation pattern will be described with reference to FIG.

[0223] In Fig. 16, the operation pattern of the auxiliary equipment is one of three auxiliary equipment operating, but other patterns such as two of three units operating or all three units operating are also possible. Also, since the power to the auxiliary equipment may be turned off during inspection, etc., in such cases the operation pattern is not established and the correlation map data table is not selected when the auxiliary equipment cannot be started. Equipment ON / OFF map input is performed according to the operating state.

[0224] This map determines the selection and data entry of the correlation data map table.

[0225] FIG. 17 shows how a correlation map table is generated for each auxiliary operation pattern.

[0226] As mentioned above, if there are three auxiliary devices, there are seven possible auxiliary device operation patterns, and therefore seven patterns of correlation map tables are prepared.

[0227] Next, the lineage neuron heat map will be described with reference to FIG.

[0228] The system neuron heat map is displayed on the system display, and is displayed in a list for each system starting from system A. For correlation map cells in system 1100, an alarm occurs when the cell is 1 and turns red (black and white inverted in Fig. 11). If it exceeds 0.11, it turns yellow (dot display in Fig. 11), and if it is 0.1 or less, it turns green (normal display in Fig. 11).

[0229] Also displays the deviation (maximum deviation) value. Displays synapse input A and the degree of influence.

[0230] The frame displays (bold frame and diagonal line displays in Figure 11) indicate the related systems (downstream systems) of system 1100 where the starting abnormality occurred, and if there is an impact, they are displayed in red (diagonal line display) or yellow (bold frame display), respectively, and if there is no impact on the systems, they remain displayed in green.

[0231] The lineage neuron heatmap displays the dissociation and synaptic inputs, indicating the influence of lineage 1100.

[0232] As shown on the right side of the system neuron heat map display, if the starting point is displayed for external factors such as seawater temperature, atmospheric temperature, atmospheric pressure, system voltage, and system frequency that cannot be controlled by the power plant alone, it is easy to show that it is an external factor. If the starting point is due to the system, it shows that a failure has occurred inside the power plant, which will affect other systems 1100. For this reason, a yellow display frame (bold frame display) and a red display frame (diagonal line display) are displayed for the related systems 1100, and the influence of the entire power plant system can be seen from the arrangement of the frames and the color change of the system neuron heat map.

[0233] Regarding the color display of external factors such as seawater temperature, for example, if the past temperature change data distribution is within 2σ, it may be green, if it is within 2σ to 3σ, yellow, and if it is outlier data, red. However, since seawater temperature has a direct effect on the output of the power plant, the green range may be the design temperature range where the rated output of the power plant can be maintained.

[0234] In this way, external factors may be set to green, yellow, or red, assuming that they follow a normal distribution based on past data.

[0235] The lineage neuron heat map display may return to green when the causal event is resolved. In that case, a memory database may be provided to track past events and reference the database.

[0236] When the system neuron heat map window of the fault origin occurrence is clicked, an autoseries graph is displayed as shown in Fig. 20, but it may also be displayed as a system tree (Fig. 2 or Fig. 14) that displays the connections of the systems 1100. This allows the correlation of the systems 1100 to be visually grasped by showing the connections of the systems 1100 in a tree form.

[0237] In this proposal, the system neuron heat map display of the deviation of the system 1100 is only the maximum value, but in order to evaluate the deviation component in the correlation parameter cell in the system 1100, it is also possible to calculate and display parameters such as the minimum value, the median value, the average value, and the standard deviation in addition to the maximum value by clicking on the system neuron heat map window. If the numerical value of each calculated value is high, it indicates that there is a large deviation between the learning data and the operating data. In this way, it is also possible to display the change in the operating state of the system numerically.

[0238] Although the synaptic input / output values ​​are not displayed on the lineage neuron heat map, the calculation results may be displayed by clicking on the lineage neuron heat map window in the same manner as above.

[0239] FIG. 18 shows an explanatory diagram of the lineage neuron window.

[0240] The lineage neuron heatmap shows only minimal information, but you can click on the heatmap window to view more detailed information.

[0241] Here, the system name, details of discrepancy, events, maximum correlation map cell, average discrepancy, median discrepancy, details of input synapses (total, input synapse value from other systems), and output synapses are displayed. In addition, a command button may be provided to display a time series graph of the correlation map maximum discrepancy in the system neuron heat map shown in Figure 20.

[0242] Next, a time series graph display of maximum discrepancy of a correlation map in a lineage neuron heat map will be described with reference to FIG.

[0243] Regarding the system neuron heat map display, if it is necessary to know the time series change of the deviation or the calculated value with the synapse total value, for example, by clicking on the system neuron heat map window, a deviation change graph with the deviation on the vertical axis and time on the horizontal axis may be displayed as shown in FIG. 20. The squares in the graph indicate that event data has occurred, and the event data with the maximum deviation is displayed. When these event data occur, it is suggested that some kind of failure has occurred. In addition, after the event data is cleared, normal operation is resumed. By looking at the change in the maximum deviation at that time, it is possible to determine whether the abnormality is continuing or has ended.

[0244] In addition, by displaying the parameter of the correlation map cell that is causing the maximum deviation (for example, the correlation between the XX pump outlet flow rate and the YY pump outlet pressure), it is possible to display which parameter is affected, which can assist in the investigation of the cause.

[0245] In addition, while FIG. 20 shows a time series data graph of the maximum deviation degree, synaptic input data may also be incorporated into the time series data graph to verify the time series effect.

[0246] Next, a flow of the system neuron heat map color display and the system neural network synapse output will be described with reference to FIG.

[0247] First, the plant management system 100 refers to the auxiliary equipment management table 640 and checks the rotating equipment operating state (S2000).

[0248] The plant management device 100 determines the selection of the correlation data map table and the data input according to the rotating equipment operating state (S2010).

[0249] Next, the plant management device 100 uses the selected correlation data map table to calculate the deviation (second index value) from the Euclidean distance between the center point of the learning data cluster (first reference value) and the center point of the operation data cluster (first index value) (S2020).Then, the plant management device 100 obtains the maximum deviation (third index value) in the entire system 1100 including the non-thermometer and temperature system (S2030).

[0250] The plant management device 100 determines whether the correlated sampling data exceeds the alarm line or a predetermined upper or lower limit value, and whether the maximum deviation is 1 or more (S2040), and if Yes, displays the heat map of the system 1100 in question in red (black and white inverted display in FIG. 11) and displays the heat map of the downstream system 1100 of the system 1100 in a red frame (diagonal line display in FIG. 11) (S2050). The plant management device 100 also displays the "starting point" on the heat map of its own system 1100 (S2100).

[0251] The plant management device 100 also determines whether the sum of the synaptic input of the other system 1100 and the maximum deviation of the own system 1100 is 1 or more (S2060), and if it is 1 or more, displays the heat map of the system 1100 in yellow (displayed as a dot in FIG. 11) and displays the heat map of the downstream system 1100 of the system 1100 in yellow (displayed with a thick frame in FIG. 11) (S2070). The plant management device 100 also displays the "starting point" on the heat map of the own system 1100 (S2100).

[0252] In addition, the plant management device 100 determines whether the sum of the maximum deviation degrees of the own system 1100 is greater than or equal to 0.1 and less than 1 (S2080), and if the answer is Yes, displays the heat map of the system 1100 in yellow (displayed as a dot in Figure 11) (S2090).

[0253] If any of S2040, S2060, and S2080 is No, the plant management device 100 displays the heat map of the system 1100 in green (normal display in FIG. 11) (S2120).

[0254] The plant management device 100 performs the above process sequentially for each system 1100 (S2110), thereby displaying red, green, and yellow in the system neuron heat map, and displaying the respective starting points and frames.

[0255] Next, with reference to FIG. 12, a description will be given of a flowchart for calculating the degree of discrepancy between data clusters representing the presence or absence of event data, displaying a systematic neuron heat map and graph output, and calculating synaptic input / output.

[0256] When the plant management device 100 detects that an event is ON (S3000), it refers to the event management table 630 to identify the correlation map table 620 for the event, and samples the plant data (operation data) in association with the event and stores it in the plant data management table 610 (S3010).

[0257] Then, the plant management system 100 calculates the center point of this plant data cluster (S3020), and calculates the Euclidean distance (degree of deviation) from the center point of the learning data cluster (S3030).

[0258] The plant management device 100 obtains the maximum value of the deviation degree in the system 1100 (S3100), stores it in the storage database 140, and performs an analysis using application software for operational data cluster analysis (S3090).

[0259] In addition, the plant management device 100 uses the synapse input values ​​(x, w, v) obtained from the upstream system 1100 (S3120) to calculate the synapse output value Y of its own system 1100 (S3130), stores it in the storage database 140, and performs analysis using application software for operating data cluster analysis (S3090).

[0260] The plant management device 100 displays the deviation and synapse input value in a heat map using colors based on the results of the calculations in S3100 and S3130 (S3110, S3140, S3150, S3160, S3180).

[0261] Furthermore, the plant management system 100 appropriately displays a graph as shown in FIG. 20 (S3170).

[0262] Next, when the plant management device 100 detects that the event is OFF (S3040), it refers to the event management table 630 to identify the correlation map table 620 for when there is no event, and samples the plant data (operation data) in association with the absence of an event and stores the sampled data in the plant data management table 610 (S3050).

[0263] Then, the plant management system 100 calculates the center point of this plant data cluster (S3060), and calculates the Euclidean distance (degree of deviation) from the center point of a learning data cluster without an event (S3070).

[0264] Thereafter, when the next event occurs (S3080), the plant management system 100 starts the process of S3010.

[0265] In the above manner, the operator can quickly grasp the state of any abnormality that has occurred in the plant 1000.

[0266] As a supplement to the present system 100, since the operation data cluster is labeled and time-stamped with events or digital data, the deviation can be calculated in the same manner by comparing a data cluster at the time of the occurrence of a past event with the same current event data cluster in a chronological order. Therefore, an abnormality may be discovered depending on the presence or absence of deviation of the data cluster when the data cluster is performed according to normal procedures such as regular testing.

[0267] Also, because it is time-stamped, the trajectory of the sampled data can be tracked, as well as compared to historical data trajectories.

[0268] These processes may be performed by outputting data from the storage database when an operator determines an abnormality, and facilitating comparison with past data using separate evaluation application software (S3090 in FIG. 21). Also, the application system may automatically detect deviations from learning data in the transition of the center point of the time-series operation data cluster of the same past event, and use this as a sign of failure management.

[0269] In addition, depending on the plant 1000, the number of system parameter data (non-temperature system) and system parameter data (temperature system) may be small, with a minimum of one each. In this case, a correlation map cell may be formed by combining the temperature system and the non-temperature system.

[0270] The above-described embodiment is for the purpose of facilitating understanding of the present invention, and is not intended to limit the present invention. The present invention may be modified or improved without departing from the spirit of the present invention, and equivalents thereof are also included in the present invention.

[0271] For example, the system 100 can be applied to a plant 1000 that is made up of multiple related systems 1100, such as a nuclear or thermal power plant, a chemical plant, or the like.

[0272] The system 100 can also be applied to cyber security of the plant 1000. Even if a false signal is inserted into a single control measuring device or an unauthorized control behavior occurs due to a signature-based attack such as unauthorized access from an external network or installation of malware by an unauthorized peripheral device, it is possible to detect an abnormality in the relevant location and confirm the impact on the plant.

[0273] In addition, this system 100 is a plant anomaly monitoring system 100 and a plant anomaly monitoring method assumed to be used in a nuclear power plant, but when applied to another plant 1000, it is necessary to configure it in accordance with that system 1100. However, even if a person skilled in the art can easily imagine and make improvements, they can be considered to be the same as long as the gist of the present invention is applied. [Explanation of symbols]

[0274] 100 Plant management device 101 Influence information storage unit 102 Plant Data Acquisition Department 103 Second index value calculation unit 104 Third index value calculation unit 105 Fourth index value calculation unit 106 Abnormality Information Display Unit 107 Alarm output unit 110 CPU 120 Memory 130 Communication Equipment 140 Storage device 150 Input Device 160 Output Device 170 Recording medium reader 200 Operation monitoring computer 500 Network 600 Influence Information Table 610 Plant Data Management Table 620 Correlation Map Table 630 Event Management Table 640 Auxiliary Equipment Management Table 700 Plant Management Equipment Control Program 800 Recording media 1000 plants 1100 line

Claims

1. A plant management device that displays the state of an abnormality occurring in a plant, an influence information storage unit that stores influence information representing susceptibility of each system that shares control in the plant to an influence from an upstream system that performs control at a higher upstream; A plant data acquisition unit that acquires a plurality of types of plant data periodically measured in each system; a second index value calculation unit that calculates, for each combination of two types of plant data selected from plant data measured within the same system, a first index value indicating a relationship between the two types of plant data obtained from measurement values ​​of the two types of plant data measured within a predetermined period of time and a second index value indicating a degree of deviation between the first index value and a first reference value serving as a reference for the relationship; a third index value calculation unit that calculates, for each system, a maximum value of the second index values ​​as a third index value that indicates the degree of an abnormality occurring in the system; a fourth index value calculation unit that calculates, for each system, a fourth index value that indicates a degree of influence of an abnormality on each system from an upstream system, using the third index value and the influence information of each system; an abnormality information display unit that displays a list of the third index value and the fourth index value in each system to display a state of an abnormality in the plant; A plant management device comprising:

2. The plant management device according to claim 1, the first index value is a coordinate value obtained by plotting a point determined from each representative value of the two types of plant data on a scatter plot of the two types of plant data measured during the predetermined period, the first reference value is a coordinate value when a point determined from each representative value of the two types of plant data measured during a reference period is plotted on the scatter diagram, the second index value is a Euclidean distance between the coordinate value represented as the first index value and the coordinate value represented as the first reference value; Plant management equipment.

3. The plant management device according to claim 1, When selecting a combination of the two types of plant data, the second index value calculation unit selects a combination of plant data related to temperature and a combination of plant data related to a physical quantity other than temperature. Plant management equipment.

4. The plant management device according to claim 1, The abnormality information display unit displays, for a system in which the third index value exceeds the second reference value, a message indicating that an abnormality has occurred in the system. Plant management equipment.

5. The plant management device according to claim 4, The abnormality information display unit further displays, for a downstream system of the system in which the third index value exceeds the second reference value, a message indicating that an abnormality has occurred in an upstream system. Plant management equipment.

6. The plant management device according to claim 1, an alarm output unit that outputs an alarm associated with any of the plant data when the value of the plant data deviates from a predetermined range defined as a normal range; Further equipped with When an alarm is output for any of the plant data, the abnormality information display unit displays a message indicating that an alarm is output for the system in which the plant data is measured. Plant management equipment.

7. The plant management device according to claim 6, The abnormality information display unit further displays, for a downstream system of the system to which the alarm is output, a message indicating that an alarm has been output in an upstream system. Plant management equipment.

8. The plant management device according to claim 1, the influence information is determined for each combination of the system and one or more upstream systems that affect control of the system, the fourth index value calculation unit calculates, for each combination of the system and an upstream system of the system, a product of a third index value of the upstream system and influence information corresponding to the combination, and calculates a sum of the products as the fourth index value. Plant management equipment.

9. A method for controlling a plant management device that displays a state of an abnormality occurring in a plant, comprising: The plant management device, storing influence-force information representing susceptibility to influence from an upstream system performing control at a higher level for each system that shares control in the plant; Acquire multiple types of plant data periodically measured in each system, For each system, for each combination of two types of plant data selected from plant data measured within the same system, a first index value indicating a relationship between the two types of plant data obtained from measurement values ​​of the two types of plant data measured within a predetermined period is calculated, and a second index value indicating a degree of deviation between the first index value and a first reference value serving as a reference for the relationship; For each system, the maximum value of the second index value is calculated as a third index value representing the degree of abnormality occurring in the system; calculating, for each system, a fourth index value that indicates the degree of influence of an abnormality on each system from an upstream system using the third index value and the influence information of each system; displaying a list of the third index value and the fourth index value in each system to display an abnormality state in the plant. A method for controlling a plant management device.

10. A program for displaying the status of an abnormality occurring in a plant, On the computer, a step of storing influence information representing susceptibility of each system that performs control in a shared manner in the plant to being influenced by an upstream system that performs control at a higher upstream; A procedure for acquiring multiple types of plant data periodically measured in each system; a step of calculating, for each combination of two types of plant data selected from plant data measured within the same system, a first index value indicating a relationship between the two types of plant data obtained from measurement values ​​of the two types of plant data measured within a predetermined period of time, and a second index value indicating a degree of deviation between the first index value and a first reference value serving as a reference for the relationship; calculating, for each system, a maximum value of the second index values ​​as a third index value representing the degree of an abnormality occurring in the system; calculating, for each system, a fourth index value that indicates a degree of influence of an abnormality on each system from an upstream system, using the third index value and the influence information of each system; a step of displaying a list of the third index value and the fourth index value in each system to display an abnormality state in the plant; A program for executing the above.

Citation Information

Patent Citations

  • Grounding terminal block

    JP2014127276A

  • Abnormality diagnosis system

    JP2017062730A

Cited By

  • Video monitoring and AI linked intelligent alarm verification system

    CN121686712A

  • A smart alarm verification system that integrates video surveillance and AI

    CN121686712B

  • Process-data processing method and computer device

    WO2026152739A1