JP2022165744A - Analyzing system, learning device, anomaly sign detection system, anomaly sign detection method, and program - Google Patents

Abstract

To provide an analyzing system that is able to improve the detection accuracy of an anomaly sign.SOLUTION: An analyzing system for detecting an anomaly sign of a plant includes: a signal classifying unit 52 that groups signals from sensors into signal groups by using signal definition information and classifies the signal groups by using information indicating a relation between signals; a signal-set setting unit 53 that sets a set of signals, in which two signals belonging to the signal groups are set as one set, by using a classification result and a rule for generating a set of signals; a correlation coefficient calculation unit 54 that, for each set of signals, calculates time-series data of a correlation coefficient of the two signals for a reference period; a model generation unit 55 that generates a learning model including the time-series data of the correlation coefficient as reference data; a correlation coefficient calculation unit 63 that, for each set of signals, calculates time-series data of the correlation coefficient of the two signals for a monitor target period; and an anomaly sign detection unit 64 that detects an anomaly sign by using the time-series data of the correlation coefficient and the learning model.SELECTED DRAWING: Figure 2

Description

The present disclosure relates to an analysis system, a learning device, an anomaly sign detection system, an anomaly sign detection method, and a program.

In power plants, chemical plants, and other plants, time-series data acquired by various sensors is used for monitoring signs of anomalies, detecting anomalies, and the like. Time-series data may also be used to monitor equipment or facilities outside of plants. For example, Patent Literature 1 discloses a plant monitoring diagnostic device that generates a learning model using normal time-series data acquired by a sensor that detects physical quantities of each part of a facility to be monitored, and detects signs of abnormality. is disclosed.

Japanese Patent Application Laid-Open No. 2020-190910

However, even if the time-series data acquired by the sensor is actually normal, it can change in various ways, and even if there is a sign of abnormality, if the divergence from normal data is large, it is judged to be normal. there is a possibility. On the other hand, if an abnormal sign is overlooked, there is a possibility that damage will occur, and if an abnormal sign is detected excessively, the burden on the monitor will increase. Therefore, further improvement in detection accuracy is desired.

The present disclosure has been made in view of the above, and an object thereof is to obtain an analysis system capable of improving detection accuracy of signs of abnormality.

In order to solve the above-described problems and achieve the object, the analysis system according to the present disclosure uses a plurality of signals respectively acquired by a plurality of sensors provided in a plant to be monitored to detect abnormal signs of the plant. grouping a plurality of signals into signal groups using predetermined signal definition information indicating the location of the detection target of the signal and the physical quantity of the detection target, and dividing the signal group into the signal group a signal classifier for classifying using information indicating the relationship between belonging signals. The analysis system further classifies two signals among the plurality of signals belonging to the signal group into one set using the result of classification by the signal classifier and a rule for generating a predetermined set of signals for each classification. A signal set setting unit that sets a signal set to be used as a signal set, and for each signal set set by the signal set setting unit, time-series data of a correlation coefficient of two signals that constitute the signal set in a reference period is calculated. a first correlation coefficient calculator; and a model generator that generates a learning model that includes the time-series data of the correlation coefficient calculated by the first correlation coefficient calculator as reference data for detection of an abnormal sign. The analysis system further calculates, for each signal pair set by the signal pair setting unit, time-series data of the correlation coefficients of the two signals that form the signal pair among the plurality of signals in the period to be monitored. 2 correlation coefficient calculation unit; an abnormality sign detection unit that detects an abnormality sign of the plant using the time-series data of the correlation coefficient calculated by the second correlation coefficient calculation unit and the learning model; .

Advantageous Effects of Invention According to the present disclosure, it is possible to improve the detection accuracy of signs of abnormality.

1 is a diagram showing a configuration example of an abnormality sign detection system according to an embodiment; FIG. FIG. 4 is a diagram showing an example functional configuration of a learning device and an anomaly sign detection device according to an embodiment; 1 is a diagram showing a configuration example of a computer system that implements a learning device according to an embodiment; FIG. 4 is a flow chart showing an example of a learning processing procedure in the learning device according to the embodiment; FIG. 4 is a diagram showing an example of sensor data according to the embodiment; FIG. 4 is a diagram showing an example of sensor data according to the embodiment; A diagram showing an example of classification information according to the embodiment 3 is a flowchart showing an example of a signal classification procedure according to an embodiment; FIG. 10 is a diagram showing an example of a signal group set using the first pattern of classification B according to the embodiment; FIG. 11 is a diagram showing an example of a signal group set using the second pattern of classification B according to the embodiment; FIG. 4 is a diagram showing an example of a signal set set using a pattern of classification C according to the embodiment; FIG. 10 is a diagram showing an example of a signal group set using the first pattern of classification D according to the embodiment; FIG. 11 is a diagram showing an example of a signal group set using the second pattern of classification D according to the embodiment; Flowchart showing an example of an abnormality sign detection processing procedure in the abnormality sign detection device according to the embodiment FIG. 4 is a diagram showing an example of signal attribute information according to the embodiment;

An analysis system, a learning device, an anomaly sign detection system, an anomaly sign detection method, and a program according to embodiments will be described below in detail with reference to the drawings.

FIG. 1 is a diagram illustrating a configuration example of an abnormality sign detection system according to an embodiment; The abnormal sign detection system 1 of the present embodiment monitors plants such as power plants, chemical plants, and other plants, and detects abnormal signs of the monitored plants. An example in which the monitoring target of the abnormality sign detection system 1 is a power plant will be described below, but the monitoring target of the abnormality sign detection system 1 is not limited to the power plant.

As shown in FIG. 1, the sign-of-abnormality detection system 1 includes sensors 2-1 to 2-n that detect the states of monitored objects, a database device 3, and an analysis system 4. FIG. n is an integer of 2 or more. In the example shown in FIG. 1 , the display data providing device 7 and the display device 8 may be included in the abnormal sign detection system 1 .

The sensors 2-1 to 2-n detect physical quantities such as temperature, flow rate, or pressure, and transmit values indicating the detected physical quantities, that is, engineering values, together with corresponding times as sensor data to the database device 3 via the communication network. do. Further, the sensors 2-1 to 2-n include those that detect a value indicating whether or not the device to be monitored is operating and transmit it to the database device 3 as sensor data along with the time via the communication network. It may be The communication network may be a wireless or wired LAN (Local Area Network), a WAN (Wide Area Network) including the Internet, or a mixture thereof. Hereinafter, the sensors 2-1 to 2-n are also referred to as the sensor 2 when they are indicated without distinguishing between them.

The database device 3 stores the time-series data received from the sensors 2-1 to 2-n as sensor data. Each sensor data is associated with a signal name, and the database device 3 stores the signal name or information indicating the signal name in association with the sensor data.

The analysis system 4 acquires sensor data, which is time-series data, from the database device 3, generates a learning model using the sensor data during normal times, and uses the learning model to detect signs of abnormality in the sensor data at the monitoring target date and time. It detects and transmits the detection result to the database device 3 . In the following description, the analysis system 4 of the present embodiment determines that an abnormality sign has been detected when it detects that it is not normal. It can also be applied when it is determined that there is an abnormality when it is detected that there is no such thing.

The analysis system 4 of the present embodiment uses the distance between normal time-series data and time-series data to be monitored to determine the similarity of two signals, for example, like a DTW (Dynamic Time Warping) method. Then, the similarity is used to detect signs of abnormality. The method of detecting an abnormal sign in the analysis system 4 is not limited to the example using DTW, and the time-series data in the normal state is used as learning data, and the deviation between the time-series data and the time-series data of the monitoring target date and time is used. Any method may be used as long as it determines an abnormality sign.

The display data providing device 7 acquires each information from the database device 3 in response to a request from the display device 8 , generates display information for displaying the various information, and transmits the generated display information to the display device 8 . The display data providing device 7 has, for example, a function as a Web server, and acquires display information from the display data providing device 7 by designating a URL (Uniform Resource Locator) corresponding to the display device 8, thereby obtaining sensor data. , detection results by the analysis system 4, and the like can be displayed. The display data providing device 7 may directly acquire the detection result from the analysis system 4 and transmit it to the display device 8 .

The display device 8 may be, for example, a computer capable of receiving operations from an administrator of the power plant to be monitored, and may be a portable terminal. The display device 8 may have a function of receiving input from the administrator and transmitting the received input to the display data providing device 7 or the analysis system 4 .

Although the database device 3 and the display data providing device 7 are provided separately in the example shown in FIG. Also, the database device 3, the display data providing device 7, and the analysis system 4 may be integrated into one device. Moreover, the display data providing device 7 and the display device 8 may be omitted by providing the analysis system 4 with a display unit and realizing the function as the display device 8 .

The analysis system 4 includes a learning device 5 that performs learning using sensor data in a normal state, and an abnormal sign detection device 6 that detects an abnormal sign of the sensor data at the monitoring target date and time using the learning result of the learning device 5. . In the example shown in FIG. 1, the learning device 5 and the sign-of-abnormality detection device 6 are provided separately, but the learning device 5 and the sign-of-abnormality detection device 6 are integrated to be realized as one analysis device. may

FIG. 2 is a diagram showing an example of the functional configuration of the learning device 5 and the sign-of-abnormality detection device 6 of this embodiment. As shown in FIG. 2, the learning device 5 includes a data acquisition unit 51, a signal classification unit 52, a signal set setting unit 53, a correlation coefficient calculation unit 54, a model generation unit 55, a signal definition storage unit 56, and a pattern storage unit 57. , a learning model storage unit 58 and a learning information transmission unit 59 .

The data acquisition unit 51 acquires sensor data in a specified period after the sensor data, which is time-series data, from the database device 3 . The designation of the period to be acquired by the data acquisition unit 51 may be input using input means of the learning device 5 (not shown), or may be received from another device to designate the period. The data acquisition unit 51 outputs the acquired sensor data to the correlation coefficient calculation unit 54 . Since the learning device 5 of the present embodiment performs learning using normal sensor data, the designated period is the period during which the monitoring target was normal.

The signal classification unit 52 classifies the sensor data acquired by the data acquisition unit 51 using the signal definition information and the classification information stored in the signal definition storage unit 56, and notifies the signal group setting unit 53 of the classification result. do. The signal definition information is information indicating what, in which device, and what is detected by each signal. In the present embodiment, it is assumed that the signal name of each signal is named based on information such as what is detected at which location, and the signal definition information indicates which device the signal corresponds to for each signal. device correspondence information to indicate, naming convention information indicating the naming convention for signal names, and the signal names themselves. Classification information is information indicating rules for classifying signals. Details of the signal definition information and the classification information will be described later.

The signal group setting unit 53 generates a signal group using the classification result notified from the signal classifying unit 52 and the pattern information stored in the pattern storage unit 57, and compares the signal group information indicating the signal group. The relational coefficient calculation unit 54 is notified. The pattern information is information indicating rules for generating signal sets for each classification. Details of the pattern information will be described later. A signal set is a set consisting of two signals. Note that the processing of the signal classification unit 52 and the signal group setting unit 53 may be performed each time data is acquired by the data acquisition unit 51, but once the processing is performed, addition, change, deletion, etc. of the signal may be performed. may be performed when is performed.

The correlation coefficient calculation unit 54 calculates the correlation coefficient of the sensor data input from the data acquisition unit 51 for each signal group indicated by the signal group information notified from the signal group setting unit 53, thereby determining the correlation number of time-series data is generated and output to the model generation unit 55 . The correlation coefficient calculation unit 54 also notifies the model generation unit 55 of the signal set information notified from the signal set setting unit 53 .

Using the time-series data input from the correlation coefficient calculation unit 54, the model generation unit 55 calculates, for each signal set, a threshold value for determining normality regarding the divergence from normal data, and the like, along with the time-series data. It is stored in the learning model storage unit 58 as a learning model. The model generation unit 55 also stores the signal set information notified from the correlation coefficient calculation unit 54 in the learning model storage unit 58 .

The learning information transmission unit 59 transmits the signal set information and the learning model stored in the learning model storage unit 58 to the abnormality sign detection device 6 as learning information. Note that the learning information transmission unit 59 transmits the learning model when the learning model is updated. The learning information transmitting unit 59 may transmit the signal group information each time the learning model is updated, or may transmit the signal group information when the signal group information is updated.

The abnormality sign detection device 6 includes a data acquisition unit 61 , a learning information reception unit 62 , a correlation coefficient calculation unit 63 , an abnormality sign detection unit 64 , a learning model storage unit 65 and a detection result transmission unit 66 .

The data acquisition unit 61 acquires the sensor data of the monitoring target date and time from the database device 3 and outputs the sensor data to the correlation coefficient calculation unit 63 . The learning information receiving unit 62 receives the learning information described above from the learning device 5, stores the learning model in the learning model storage unit 65, and notifies the correlation coefficient calculation unit 63 of the signal set information.

The correlation coefficient calculation unit 63 calculates the correlation coefficient of the sensor data input from the data acquisition unit 61 for each signal group indicated by the signal group information notified from the learning information reception unit 62, thereby obtaining the correlation number of time-series data is generated and output to the abnormality sign detection unit 64 .

The abnormal sign detection unit 64 uses the learning model stored in the learning model storage unit 65 and the time series data of the correlation coefficient input from the correlation coefficient calculation unit 63 to detect an abnormal sign, The detection result is notified to the detection result transmission unit 66 .

The detection result transmission unit 66 transmits the detection result notified from the abnormality sign detection unit 64 to the database device 3 . When receiving the detection result from the detection result transmitting unit 66, the database device 3 stores the detection result.

In the present embodiment, the learning device 5 automatically generates a signal set of signals assumed to have a correlation, and calculates the correlation coefficient of the time-series data of the two signals corresponding to the signal set as the time-series data. learn as As a result, it is possible to improve the detection accuracy of signs of abnormality compared to the case where learning is performed using the time-series data of a single signal while saving the administrator's trouble. For example, assume that there is an exchange of substances between tank A and tank B. If you measure the temperature of tank A and the temperature of tank B and use each time-series data independently to detect signs of abnormality, even if a leak occurs in the tank, the time-series data of the single temperature will not change the value. There are cases where the number is small and judged to be normal. Even in such a case, the time-series data of the correlation coefficient shows a greater change due to the occurrence of a tank leak than the single time-series data, so the correlation coefficient can be used to detect signs of abnormality. In particular, when the signal names corresponding to each sensor data are named according to a defined rule, it is only necessary to analyze and classify the signal names and determine the signal set using the pattern information for each class. Therefore, it is possible to reduce the time and effort required for preparation in advance. Furthermore, when an administrator designates a correlation function, if all conceivable combinations of signal sets are designated, the number of signal sets becomes enormous in a system with a large number of signals. For example, if all combinations of these signals are calculated, _m C ₂ (C indicates a combination) signal sets are generated, where m is the number of signal sets. Thousands of signals are used in power plants, and the number of combinations is enormous. In the present embodiment, for each classification, the case of generating all combined signal sets, the case of generating signal sets with some combinations, the other cases, etc. are determined, and the correlation coefficient is calculated. The number of signal sets to be calculated can be suppressed compared to the case of calculating signal sets for all combinations.

In addition, when the learning device 5 and the abnormal sign detection device 6 are used as one device, the learning information transmission unit 59, the learning information reception unit 62 and the learning model storage unit 65 may not be provided, and the signal group setting is performed. The unit 53 also notifies the correlation coefficient calculation unit 63 of the signal group information, and the abnormality sign detection unit 64 detects an abnormality sign using the learning model stored in the learning model storage unit 58 . Further, when the learning device 5 and the abnormal sign detection device 6 are provided separately, the learning model storage unit 58 is provided in the database device 3 or another database device, and the abnormal sign detection device 6 is provided in the database device 3 or another database device. The learning model storage unit 65 may not be provided by acquiring the learning model from the . In this case, the signal set information may also be stored in the database device 3 or another database device.

Next, the hardware configuration of each device will be described. The learning device 5 and the sign-of-abnormality detection device 6 of the present embodiment are implemented by executing a program in which the processing in the learning device 5 and the signs-of-abnormality detection device 6 are described, respectively, on the computer system. 5 and an abnormality sign detection device 6. FIG. 3 is a diagram showing a configuration example of a computer system that implements the learning device 5 of this embodiment. Similarly, the sign-of-abnormality detection device 6 is also realized by the computer system shown in FIG. 3, for example. As shown in FIG. 3, this computer system comprises a control section 101, an input section 102, a storage section 103, a display section 104, a communication section 105, and an output section 106, which are connected via a system bus 107. there is

In FIG. 3, the control unit 101 is a processor such as a CPU (Central Processing Unit), and executes a program describing processing in the learning device 5 and the abnormal sign detection device 6 of the present embodiment. The input unit 102 is composed of, for example, a keyboard and a mouse, and is used by the user of the computer system to input various information. The storage unit 103 includes various memories such as RAM (Random Access Memory) and ROM (Read Only Memory) and storage devices such as a hard disk, and stores programs to be executed by the control unit 101 and necessary data obtained in the course of processing. store data, etc. The storage unit 103 is also used as a temporary storage area for programs. The display unit 104 includes a display, LCD (liquid crystal display panel), etc., and displays various screens to the user of the computer system. A communication unit 105 is a receiver and a transmitter that perform communication processing. The output unit 106 is a printer or the like. Note that FIG. 3 is an example, and the configuration of the computer system is not limited to the example in FIG.

Here, an operation example of the computer system until the program of the present embodiment becomes executable will be described. In the computer system having the configuration described above, for example, a program is stored in storage unit 103 from a CD-ROM or DVD-ROM set in a CD (Compact Disc)-ROM drive or DVD (Digital Versatile Disc)-ROM drive (not shown). installed on. Then, when the program is executed, the program read from storage unit 103 is stored in the main storage area of storage unit 103 . In this state, control unit 101 executes processing as learning device 5 or abnormality sign detection device 6 according to the program stored in the main storage area of storage unit 103 .

In the above description, a program describing the processing in the learning device 5 and the abnormality sign detection device 6 is provided using a CD-ROM or DVD-ROM as a recording medium. For example, a program provided via a transmission medium such as the Internet via the communication unit 105 may be used depending on the configuration, the capacity of the program to be provided, and the like.

Signal classifier 52, signal set setter 53, correlation coefficient calculator 54, and model generator 55 of learning device 5 shown in FIG. 3 is realized by being executed by the control unit 101 shown in FIG. The signal definition storage unit 56, the pattern storage unit 57, and the learning model storage unit 58 shown in FIG. 2 are part of the storage unit 103 shown in FIG. Data acquisition unit 51 and learning information transmission unit 59 shown in FIG. 2 are implemented by communication unit 105 and control unit 101 shown in FIG. The abnormality sign detection device 6 may be realized by a plurality of computer systems. Also, the learning device 5 may be realized by a cloud system.

The correlation coefficient calculation unit 63 and the abnormality sign detection unit 64 of the abnormality sign detection device 6 shown in FIG. 2 are executed by the control unit 101 shown in FIG. It is realized by being executed. The learning model storage unit 65 shown in FIG. 2 is a part of the storage unit 103 shown in FIG. Data acquisition unit 61 and detection result transmission unit 66 shown in FIG. 2 are implemented by communication unit 105 and control unit 101 shown in FIG. Note that the learning device 5 may be realized by a plurality of computer systems. Moreover, the abnormality sign detection device 6 may be realized by a cloud system.

Similarly to the learning device 5 and the abnormality sign detection device 6, the database device 3 is also realized by the computer system illustrated in FIG. The operation example until the computer system becomes ready to execute the program for operating as the learning device 5 and the sign of abnormality detection device 6 is the same as that of the learning device 5 and the sign of abnormality detection device 6 .

For example, the program of the present embodiment causes the computer system to use predetermined signal definition information indicating the location of the detection target of the signal and the physical quantity of the detection target for the plurality of signals acquired by the plurality of sensors 2. a step of classifying the signal groups into signal groups using information indicating the relationship between the signals belonging to the signal groups; A step of setting a signal set in which two signals among a plurality of signals belonging to a signal group are set as one set using a generating rule; A step of calculating first time-series data, which is time-series data of the correlation coefficient of the two constituent signals, and a step of generating a learning model that includes the calculated first time-series data as reference data in detection of an abnormal sign. and calculating, for each set signal set, second time-series data that is time-series data of the correlation coefficients of the two signals constituting the signal set among the plurality of signals in the period to be monitored; and a step of detecting an abnormality symptom of the plant using the second time-series data and the learning model.

Similarly, the database device 3, the display data providing device 7 and the display device 8 shown in FIG. 1 are realized by the computer system illustrated in FIG. These devices may also be realized by a plurality of computer systems.

Next, the operation of this embodiment will be described. FIG. 4 is a flow chart showing an example of a learning process procedure in the learning device 5 of this embodiment. The learning device 5 first classifies signals (step S1). Specifically, the signal classification unit 52 classifies the learning target signal using the signal definition information and the classification information stored in the signal definition storage unit 56, and notifies the signal set setting unit 53 of the classification result. . Here, a signal is sensor data acquired by the sensor 2 . Hereinafter, sensor data with the same signal name will be referred to as the same signal even if the date and time of acquisition are different. When one sensor 2 detects two or more physical quantities, a signal name is assigned to each physical quantity, and the two or more physical quantities are different signals.

5 and 6 are diagrams showing examples of sensor data according to the present embodiment. In the example shown in FIG. 5, the sensor data is a signal containing the date and time and a numerical value indicating the detected physical quantity. In the example shown in FIG. 6, the sensor data is a signal that includes the date and time and a value indicating whether it is ON or OFF. In FIG. 6, characters ON and OFF are shown, but these may be converted to binary values such as 0 and 1 in the learning process.

As illustrated in FIGS. 5 and 6, sensor data is time-series data. The signal classification unit 52 classifies the signals, which are sensor data, using the signal definition information. In the present embodiment, as described above, the signal definition information includes, for each signal, device correspondence information indicating which device the signal corresponds to, and naming rule information including naming rules for signal names. The devices are devices provided in the power plant, such as turbines, drain pumps, pumps, and generators. The device correspondence information includes, for example, information indicating correspondence such that a signal entitled "Blade Path Temperature #1" corresponds to the turbine and a signal entitled "Drain Temperature #A1" corresponds to the drain pump. The device correspondence information may include information indicating a list of signal names corresponding to the device for each device, or may include information indicating the device corresponding to the signal name for each signal name. good.

In this embodiment, the signal name includes a first character string that indicates the location to be detected. The location to be detected indicates a location to be detected in units of internal devices, such as blade paths, drains, and XX valves, which are devices within the device that constitutes the above-described device, for example. Information indicating the locations of blade paths, drains, XX valves, etc. is also called internal equipment information. The signal name further includes a second character string, which is a character string indicating a physical quantity to be detected, such as temperature, degree of opening, flow rate, operating state (operating or not). It should be noted that the operating state can also be considered as one of the physical quantities indicating the state of the object to be detected. The signal name may further include location identification information including information indicating the detailed location of the detection target. May contain information. The naming rule information of this embodiment includes information indicating these rules.

For example, the signal name "blade path temperature #1" contains "blade path" as the first string, "temperature" as the second string, and "1" as the location identifier. The signal name "drain temperature #A1" includes "drain" as the first character string, "temperature" as the second character string, "1" as the location identification information, and "A "including. In this example, three systems, A, B, and C, are provided to detect the drain temperature at the same location, and "drain temperature #A1" indicates the A system. and Here, an example in which "#" is included before the location identification information and the system identification information is shown, but "#" may not be included.

In the case of the naming rule as described above, information indicating that the signal name is composed in the order of character string 1, character string 2, location identification information, and system identification information is held as naming rule information. As noted above, location identification information and system identification information may not be included in the signal name. For example, by including a list of character strings that can be used as character string 2 in the naming rule information, or by including a list of character strings that can be used as character string 1 and character string 2 in the naming rule information, the naming rule The information can be used to extract string1 and string2. Similarly, by including in the naming rule information that the system identification information is an alphabet such as A, B, and C following the character string 2, the system identification information can be extracted from the signal name. Similarly, by including in the naming rule information that the location identification information is a numerical value arranged after the character string 2, the location identification information can be extracted from the signal name.

The nomenclature described above is an example, and if the nomenclature used in the power plant differs from the above nomenclature, the nomenclature information may be similarly generated according to the nomenclature used. Also, in the above example, the location identification information, the information indicating the physical quantity or the state are character strings, but they may be represented by alphanumeric characters. Also, the system identification information and the location identification information are not limited to the above-described alphabetic characters or numbers, and may be indicated by other information. Also, the order of character string 1, character string 2, location identification information, and system identification information constituting the signal name is not limited to this order.

The signal classification unit 52 uses the signal name and the signal definition information to find the corresponding device and internal device information for each signal to be learned, and combines the obtained result with the classification information stored in the signal definition storage unit 56. to classify each signal. FIG. 7 is a diagram showing an example of classification information according to this embodiment. As shown in FIG. 7, the classification information of the present embodiment is information indicating whether or not there is a related signal and to which of classifications A, B, C and D it corresponds. Related signals are, for example, serial-related signals and parallel-related signals, and are described as serial and parallel in FIG. Signals in a series relationship are signals whose signal names do not include location identification information, and which have the same character string 1 and character string 2 but different system identification information, and the signal names include location identification information. These signals have the same character string 1, character string 2 and location identification information but different system identification information. Signals in a parallel relationship are signals that have the same character string 1 and character string 2 but different location identification information for signals that do not contain system identification information in their signal names, and signals that do not contain system identification information in their signal names. The signals having character string 1, character string 2, and location identification information are the same, but the location identification information is different.

The classification information shown in FIG. 7 classifies signals with neither series-related signals nor parallel-related signals into Class A, and signals with parallel-related signals but no series-related signals into Class A. Signals that are classified into B, those that have series-related signals but no parallel-related signals are classified into Class C, and those that have parallel-related signals and also series-related signals are classified into Class D. It indicates to classify. Note that the classification method indicated by the classification information shown in FIG. 7 is merely an example, and may be appropriately set according to the status of the power plant, the manager's request, and the like.

The signal classification unit 52 classifies, for example, a signal with a signal name of "blade path temperature #2", a signal with a signal name of "blade path temperature #3", etc., in addition to a signal with a signal name of "blade path temperature #1". If the signals to be learned include signals that differ only in identification information but are identical in other respects, these signals are classified into classification B. FIG. In addition, the signal classification unit 52 classifies signals having different system identification information, such as signal names "YY flow rate A," "YY flow rate B," and "YY flow rate C," as signals to be learned. If so, classify these signals in category C. In addition, the signal classification unit 52 classifies a signal with the signal name “drain temperature #A1”, a signal with the signal name “drain temperature #B1”, a signal with the signal name “drain temperature #A2”, and a signal with the signal name “drain temperature #B2”. Signals to be learned include signals that include both system identification information and location identification information in their signal names, and that at least one of the system identification information and location identification information is different and the others are the same, such as signals. If so, the signal classifier 52 classifies these signals into class D. A detailed processing procedure of the processing in step S1 will be described later.

Returning to the description of FIG. After step S1, the learning device 5 sets signal pairs according to the classification results (step S2). Specifically, the signal group setting unit 53 uses the classification result and the pattern information stored in the pattern storage unit 57 to create a signal group in which two of the plurality of signals belonging to the signal group constitute one group. set. The pattern information is a rule for generating a predetermined signal set, and is a rule for how to generate a signal set for each of Classification A, Classification B, Classification C, and Classification D. FIG. Therefore, still, signals belonging to class A do not generate a signal set. Here, it is assumed that information indicating that no signal set is generated, such as classification A, is part of the information indicating rules for how to generate signal sets. For category A, the pattern information stores information indicating that no signal set is generated. Note that when no signal set is generated, a correlation coefficient, which will be described later, is not calculated, so the signal, that is, the time-series data of the sensor data itself is used for learning, which will be described later.

Here, using the numerical value that is the location identification information of each signal and the letters of the alphabet that is the system identification information of each signal, the signals are arranged in ascending numerical order and then in alphabetical order, and adjacent signals in the order of arrangement are regarded as one set. Generate a signal set. For example, since the location identification information is a numerical value such as 1 or 2, the signals are arranged in ascending order of numerical value, and adjacent signals in the arranged order are regarded as one set of signals. Also, the order of the signals is cyclically defined, and the order of the signals is cyclically defined such that the numerical value next to the numerical value with the largest location identification information is regarded as the numerical value with the smallest location identification information.

For example, assume that the number of signals classified into category B and having the same character string 1 and character string 2 in the signal name is p. In this case, the signal with the location identification information of 1 and the signal with the location identification information of 2 are regarded as one signal set, and the signal with the location identification information of 3 and the signal with the location identification information of 4 are regarded as one signal set. , a rule for generating signal sets is determined so that all signals are used once in any signal set. Alternatively, a signal with location identification information of 1 and a signal with location identification information of 2 form one signal set, and a signal with location identification information of 2 and a signal with location identification information of 3 form one signal set. A rule for generating signal pairs is defined so that every signal is used twice in one of the signal pairs. In this case, if the location identification information is from 1 to p, the signal with the location identification information of p is followed by the signal with the location identification information of 1. A signal set consisting of is also set. Here, as a pattern corresponding to the classification B, a first pattern that generates a signal group so that all signals are used once in any signal group, and a first pattern that generates a signal group so that all signals are used once in any signal group Suppose that pattern information defines a second pattern that generates a set of signals as used in . As to which of the first pattern and the second pattern is to be used, the signal group setting unit 53 may display the corresponding signal on a display unit (not shown) to prompt the administrator to make a selection, thereby accepting the selection. However, it may be set in advance by the administrator for each of character string 1 and character string 2.

For classification C, the rules are such that the signals are arranged in alphabetical order of system identification information, namely A, B, C, D, . Determined. That is, signals having the same character string 1 and character string 2 in the signal name among the signals classified into the category C are arranged in alphabetical order of the system identification information, and the signals with the system identification information A and the signals with the system identification information B are arranged. A signal with system identification information A and a signal with system identification information C are regarded as one signal set, and a signal with system identification information B and a signal with system identification information C are regarded as one signal set. signal set.

Class D is a combination of Class B and Class C. Therefore, here, the second pattern of category B described above is the first pattern of category D, the pattern of category C is the second pattern of category D, and in category D, the first pattern of category D and the second pattern of category D are used. A rule is defined such that both sets of signals with patterns are generated. As the first pattern of the category D, the first pattern of the category B described above may be used. Also, in the same manner as in the category B, in the category D also, the administrator may set which of the first pattern of the category B and the second pattern of the category B is used.

It is conceivable that the location identification information is not continuous due to the deletion of the sensor 2 or the like. By generating a signal set with each other, a signal set can be generated in the same manner as in the above example. For example, if there are a signal with location identification information 1 and a signal with location identification information 3, but there is no signal with location identification information 2, a signal with location identification information 1 and a signal with location identification information 3 do not exist. A signal set is generated with the signals. Similarly, even if the system identification information is not continuous, it is sufficient to arrange the system identification information in alphabetical order and generate a signal set from adjacent signals in the order of arrangement.

After step S2, the learning device 5 calculates the correlation coefficient of each signal pair (step S3). Specifically, for each signal set set by the signal set setting unit 53, the correlation coefficient calculation unit 54, which is the first correlation coefficient calculation unit, calculates the difference between the two signals that make up the signal set in the reference period. First time-series data, which is time-series data of the correlation coefficient, is calculated and output to the model generation unit 55 . Specifically, for example, the correlation coefficient calculation unit 54 acquires sensor data of two signals that form a signal set in a period designated as a learning target from the database device 3 via the data acquisition unit 51 . Then, the correlation coefficient calculation unit 54 calculates a correlation coefficient using sensor data for a certain period of time among the acquired sensor data, and calculates the correlation coefficient by shifting the certain period by the increment of the time-series data. By repeating this process, time-series data of correlation coefficients is generated. For example, the correlation coefficient calculator 54 calculates the correlation coefficient using data for 60 minutes, and shifts the 60-minute interval every minute to generate the time-series data of the correlation coefficient.

Next, the learning device 5 generates a learning model using the time-series data of correlation coefficients (step S4), and terminates the process. Specifically, the model generation unit 55 generates a learning model that includes the time-series data of the correlation coefficients calculated by the correlation coefficient calculation unit 54 as reference data for detection of signs of abnormality. That is, the time-series data of the correlation coefficient input from the correlation coefficient calculation unit 54 is stored in the learning model storage unit 58 as reference data, that is, teacher data used for determination of signs of abnormality. In addition, the model generation unit 55 obtains variations using the time-series data of the correlation coefficients stored in the learning model storage unit 58, and uses the variations to determine whether or not abnormal symptoms are normal. A threshold is determined and stored in the learning model storage unit 58 as a learning model. In this embodiment, the learning model includes the teacher data and thresholds described above.

The learning model generated as described above is transmitted to the abnormal sign detection device 6 by the learning information transmission unit 59 . In addition, the learning information transmission unit 59 also transmits the signal group information indicating the signal group set in step S2 to the abnormality sign detection device 6 as described above. In addition, when the learning device 5 and the sign of abnormality detection device 6 are integrated, or when the learning model storage unit 58 is configured to be accessible from the sign of abnormality detection device 6, the learning information transmission unit 59 There is no need to send the model to the anomaly symptom detection device 6 .

Next, the method of classifying signals in step S1 described above will be described. FIG. 8 is a flowchart showing an example of a signal classification procedure according to this embodiment. As shown in FIG. 8, the signal classification unit 52 divides signal names (step S11). Specifically, for each signal, the signal classification unit 52 uses the naming rule information stored in the signal definition storage unit 56 to classify the signal name into a first character string, a second character string, location identification information, and a system identification. Divide into information.

The signal classification unit 52 extracts signal names having the same first character string and second character string for each device (step S12). Specifically, the signal classification unit 52 uses the device correspondence information stored in the signal definition storage unit 56 to classify the signal names so that the signals corresponding to the same device belong to the same group and the signal names correspond to different devices. Group by device so that each is in a separate group. Then, the signal classification unit 52 extracts signals having the same first character string and second character string within the group. The signal classification unit 52 further groups the extracted signal names into signal groups according to the first character string and the second character string. For example, when there are a plurality of signal names corresponding to the same device with a first character string of X1 and a second character string of Y1, the signal classification unit 52 classifies these signal names into signal group #1. be a group of Then, when there are a plurality of signal names with the first character string X2 and the second character string Y1, the signal group setting unit 53 groups these signal names into the signal group #2.

Next, the signal classification unit 52 extracts signal names having the same system identification information from among the signal names extracted in step S12 (step S13). Specifically, for each signal group, the signal classification unit 52 extracts signal names having the same system identification information among signal names belonging to the same signal group.

Next, the signal classification unit 52 extracts signal names having the same location identification information from among the signal names extracted in step S12 (step S14). Specifically, for each signal group, the signal classification unit 52 extracts signal names having the same location identification information among the signal names belonging to the same signal group.

Next, the signal classifier 52 classifies the signal using the extraction result (step S15), and ends the process. Specifically, in step S15, the signal classification unit 52 classifies signals using the extraction result of step S13, the extraction result of step S14, and the classification information stored in the signal definition storage unit 56. FIG. For example, for each signal group, when a plurality of signal names are extracted in step S13 and a plurality of signal names are not extracted in step S14, the signal classification unit 52 classifies signals corresponding to signal names belonging to the signal group. are in a parallel relationship, they are classified into classification B based on the classification information. As described above, the signal name may not contain the system identification information, but it is assumed that the signal names that do not contain the system identification information have the same system identification information. Similarly, it is assumed that signal names that do not contain location identification information have the same location identification information.

Further, if a plurality of signal names are not extracted in step S13 and a plurality of signal names are extracted in step S14 for each signal group, the signal classification unit 52 determines that the signal corresponding to the signal name belonging to the signal group is Since there is a serial relationship, signals corresponding to signal names belonging to the signal group are classified into category C. Further, when a plurality of signal names are extracted in step S13 and a plurality of signal names are extracted in step S14 for each signal group, the signal classification unit 52 determines that the signals corresponding to the signal names belonging to the signal group are in a parallel relationship. Since there are also signals in a series relationship, signals corresponding to signal names belonging to the signal group are classified into category D. Further, for each signal group, if a plurality of signal names are not extracted in step S13 and a plurality of signal names are not extracted in step S14, the signal classification unit 52 classifies signal names belonging to the signal group. Classify the signal into category A. The signal classification unit 52 notifies the signal group setting unit 53 of the signal name belonging to the signal group and the information indicating the classification corresponding to the signal group as the classification result.

In this manner, the signal classification unit 52 of the present embodiment groups a plurality of signals into signal groups using predetermined signal definition information indicating the location of the detection target of the signal and the physical quantity of the detection target. , the signal group is classified using classification information, which is information indicating the relationship between the signals belonging to the signal group. Specifically, in the example described above, the signal classification unit 52 extracts information indicating the corresponding device, internal device, detailed location, and physical quantity from the signal name, and performs grouping using the extracted information.

The processing procedure shown in FIG. 8 is a classification procedure based on the signal name naming rule assumed here, and the signal classification method is not limited to the example shown in FIG. It may be determined as appropriate according to naming rules.

Next, an example of the signal set set in step S2 of this embodiment will be described. FIG. 9 is a diagram showing an example of a set of signals set using the first pattern of classification B according to the present embodiment. In the example shown in FIG. 9, the blade path temperature is measured at 12 points, and the signal names of the corresponding signals are "blade path temperature #1" and "blade path temperature #2." , . . . “blade path temperature #12”. In this case, these 12 signals correspond to the same device, have the same first character string and second character string in the signal name, have the same system identification information, and have the same location identification information. Since there is no identical one, it is classified in Category B. In the example shown in FIG. 9, since it is set to use the first pattern among the two patterns of classification B for these signals, two signals with adjacent signal names when arranged in order of location identification information Signal sets are generated and each signal belongs to any one signal set. As shown in FIG. 9, in the order of the signal names, the first two signal pairs shown in (1) are generated, and then the third and fourth signal name signals (2) are generated. A signal group is automatically generated by generating a signal group by skipping one of the signals that make up the signal group and sequentially shifting it downward, such as generating the signal group shown in . be able to. In addition, in the example shown in FIG. 9, the order of arrangement is described at the top.

FIG. 10 is a diagram showing an example of a set of signals set using the second pattern of classification B according to the present embodiment. The example shown in FIG. 10 shows 12 blade path temperature signals as in the example shown in FIG. , a signal set is generated by two signals with adjacent signal names when arranged in order of location identification information, and each signal belongs to two signal sets. As shown in FIG. 10, in the order of the signal names, the first two signal pairs shown in (1) are generated, and then the second and third signal name signals (2) are generated. A signal group is automatically set by generating a signal group without sequentially shifting down the signals that are arranged earlier among the signals that make up the signal group, such as generating the signal group shown in . can do. It should be noted that in the example shown in FIG. 10, the order of arrangement is listed at the top.

FIG. 11 is a diagram showing an example of a set of signals set using the pattern of classification C according to the present embodiment. In the example shown in FIG. 11, three sensors 2 for measuring the YY flow rate are provided: A system, B system, and C system. The signal names of these three signals are "YY flow rate A", "YY flow rate B", and "YY flow rate C", respectively. In this case, these three signals correspond to the same device, have the same first character string and second character string in the signal name, do not have a plurality of the same system identification information, and have the same location identification information. It is classified into category C because there is Since the pattern of signal pairs of category C is a pattern that generates signal pairs in all combinations, as shown in FIG. A signal set of "flow rate C", "YY flow rate C" and "YY flow rate A" is set.

FIG. 12 is a diagram showing an example of a set of signals set using the first pattern of classification D according to the present embodiment. In the example shown in FIG. 12, the sensor 2 for measuring the drain temperature has three detailed measurement locations, and three sensors 2 of A, B and C systems are provided for each location. The signal names of these nine signals are "drain temperature #A1", "drain temperature #B1", "drain temperature #C1", "drain temperature #A2", "drain temperature #B2", and "drain temperature #". C2", "drain temperature #A3", "drain temperature #B3", and "drain temperature #C3". In this case, these nine signals correspond to the same device, have the same first character string and second character string in the signal name, have a plurality of the same system identification information, and have the same location identification information. It is classified into Class D because there is As for the classification D, there are the first pattern and the second pattern as described above, and FIG. 12 shows the signal set set using the first pattern.

FIG. 13 is a diagram showing an example of signal sets set using the second pattern of classification D according to the present embodiment. The example shown in FIG. 13 shows an example of a set of signals set using the second pattern of classification D with respect to nine signals similar to the example shown in FIG. In classification D, a signal set corresponding to both the first pattern shown in FIG. 12 and the second pattern shown in FIG. 13 may be set, or one of the patterns may be selected and set by the administrator. may be

Next, the abnormality sign detection processing of this embodiment will be described. FIG. 14 is a flowchart showing an example of an abnormality sign detection processing procedure in the abnormality sign detection device 6 of the present embodiment. As shown in FIG. 14, the sign-of-abnormality detection device 6 acquires time-series data to be monitored (step S21). Specifically, the data acquisition unit 61 acquires the sensor data of the monitoring target date and time from the database device 3 and outputs the sensor data to the correlation coefficient calculation unit 63 .

Next, the abnormality sign detection device 6 calculates a correlation coefficient using the set signal set (step S22). Specifically, the correlation coefficient calculation unit 63, which is the second correlation coefficient calculation unit, configures the signal group among the plurality of signals in the monitoring target period for each signal group set by the signal group setting unit 53. Second time-series data, which is time-series data of the correlation coefficients of the two signals, is calculated. That is, the correlation coefficient calculation unit 63 refers to the signal set information notified from the learning information reception unit 62, and for each signal set, one set of corresponding sensor data from the sensor data input from the data acquisition unit 61. is used to calculate the time-series data of the correlation coefficient in the same manner as the correlation coefficient calculation unit 54 and output to the abnormality sign detection unit 64 .

Next, the sign-of-abnormality detection device 6 uses the calculated time-series data of the correlation coefficient and the learning model to perform the sign-of-abnormality detection process (step S23), and terminates the process. Specifically, for example, the abnormality sign detection unit 64 detects the difference between the time-series data of the correlation coefficient input from the correlation coefficient calculation unit 63 and the teacher data stored as the learning model in the learning model storage unit 65. is calculated, and if the deviation is less than or equal to the threshold value stored as the learning model in the learning model storage unit 65, it is determined that the sensor data to be monitored is normal, and if the deviation exceeds the threshold value, an abnormality symptom is detected. Determine that there is.

The abnormal sign detection unit 64 outputs the result of the abnormal sign detection processing in step S23, that is, the determination result of whether or not the abnormal sign is detected to the detection result transmission unit 66. Send detection results to Then, as described above, the detection result is displayed on the display device 8 via the display data providing device 7 . In addition, the power plant manager checks the detection results, and if the detection results indicate that there is an abnormality sign, check the monitoring target and confirm that it is actually normal rather than an abnormality sign. Then, the corresponding sensor data may be input to the learning device 5 so as to be learned by the learning device 5 as normal data. By feeding back the detection results in this way, the accuracy of detection of signs of abnormality can be further improved.

In the example described above, the analysis system 4 classifies the signals using the signal names and sets the signal groups using the classification results. Not limited to this, the analysis system 4 stores signal attribute information indicating what and where each signal is detected in the signal definition storage unit 56, and uses the signal attribute information to automatically generate a signal set. may That is, in the present embodiment, classification may be performed using signal definition information indicating what and where each signal is detected. In the example where the information is the signal definition information and the device correspondence information, the signal name and the naming rule information described above are used, the device correspondence information, the signal name and the naming rule information are the signal definition information. The signal attribute information may further include signal-related information indicating the presence/absence of serial-related signals and the presence/absence of parallel-related signals.

FIG. 15 is a diagram showing an example of signal attribute information according to this embodiment. The signal attribute information shown in FIG. 15 includes, for each signal, equipment corresponding to the signal (equipment constituting the plant), signal name, location (internal equipment), detailed location, physical quantity detected by the sensor 2 (in the figure physical quantity), the presence or absence of signals in a series relationship (indicated as serial in the figure), and information indicating the presence or absence of signals in a parallel relationship (indicated as parallel in the figure). Thus, signal attribute information, which is an example of signal definition information, is information in which a device, an internal device, and a physical quantity are associated with a signal. The location (internal device) indicates the location in units of internal devices, like the second character string of the signal name described above. The detailed location is the location where the sensor 2 in the internal device is installed, similar to the location indicated by the location identification information described above. Signals in a serial relationship are signals in a multiple system relationship such as the A system and the B system, which have the same device, location (internal device), detailed location, and detection target, as in the above example. Signals in a parallel relationship are signals having the same device, location (internal device) and detection target, but different detailed locations, as in the above example.

When using the signal attribute information shown in FIG. 15, the signal attribute information is stored in the signal definition storage unit 56, and the signal classification unit 52 uses the signal attribute information and the classification information stored in the signal definition storage unit 56. , to classify the signal. In the example shown in FIG. 15, the signal attribute information includes the presence/absence of signals in series relationship and the presence/absence of signals in parallel relationship. can be classified. For example, the signal classification unit 52 assigns signals having the same device, location (internal device), and physical quantity to be detected to the same signal group. In some cases, this signal group is classified as B. For classification B, the signal set setting unit 53 sets all signals belonging to a signal group so that each signal belonging to the signal group is included in a predetermined number of signal sets, as in the above example. Set up pairs. Further, the signal classification unit 52 classifies signals constituting the signal group as C if there are signals with the same detailed location and there are no signals with different detailed locations. For classification C, the signal set setting unit 53 sets the signal set of combinations of all two signals belonging to the signal group, as in the above example. Further, among signals constituting a signal group, a signal having the same detailed location and a signal having a different detailed location is classified as D. For each signal group, the signal classification unit 52 classifies the signal names of the signals belonging to the signal group and the information indicating the classification corresponding to the signal group as a classification result, as in the case of classification using the signal name. 52.

When the signal attribute information shown in FIG. 15 is used, the signal classification unit 52 arranges the signal names of signals belonging to the same signal group in arbitrary order. By treating this order of arrangement in the same manner as the order of arrangement based on the location identification information described above, a signal set can be similarly generated.

In the example shown in FIG. 15, the presence or absence of a signal in a series relationship and the presence or absence of a signal in a parallel relationship are determined so that it is easy to grasp which category from category A to category D the signal belongs to directly from the signal attribute information. Although the information to indicate is provided, this information does not have to be included in the signal attribute information. The presence or absence of a signal in a series relationship and the presence or absence of a signal in a parallel relationship are determined by the device, location (internal device), detailed location and detection corresponding to the signal, as in the example of classifying using the signal name described above. It can be obtained from the target. For example, as in the case of classifying using the signal name, the signal classifying unit 52 uses the signal attribute information to determine whether there are multiple signals having the same device, location (internal device), detailed location, and detection target, and , device, location (internal device), and detection target, and signals that do not have a plurality of signals with the same detailed location are classified as C. In addition, the signal classification unit 52 determines that a plurality of signals having the same device, location (internal device), detection target, and detailed location exist and that the device, device, location (internal device), detailed location, and detection target are the same. If there is no signal whose detailed location is different among the signals, it is classified as B. In addition, the signal classification unit 52 classifies a plurality of signals having the same device, location (internal device), detection target, and detailed location, and a signal having the same device, device, location (internal device), detailed location, and detection target. If there is a signal with a different detailed location, it is classified as D.

As described above, in the present embodiment, signals are grouped into signal groups presumed to be related to each other using device correspondence information and signal attribute information indicating what and where each signal is detected. Then, a signal group is automatically generated using pattern information indicating a method of generating a signal group predetermined for each relationship between signals in the signal group, and a correlation coefficient is calculated for each signal group. Then, the time-series data of the correlation function is used to perform learning for detection of signs of abnormality. Therefore, it is possible to improve the detection accuracy of the signs of abnormality. In addition, it is possible to reduce the labor of work as compared with the case of manually setting each signal group. Furthermore, the number of signal pairs to be set can be suppressed compared to the case of setting signal pairs as a combination of all signals, and processing can be performed efficiently.

The configurations shown in the above embodiments are only examples, and can be combined with other known techniques, or can be combined with other embodiments, without departing from the scope of the invention. It is also possible to omit or change part of the configuration.

1 abnormality sign detection system, 2-1 to 2-n sensors, 3 database device, 4 analysis system, 5 learning device, 6 abnormality sign detection device, 7 display data providing device, 8 display device, 51, 61 data acquisition unit, 52 signal classification unit, 53 signal set setting unit, 54, 63 correlation coefficient calculation unit, 55 model generation unit, 56 signal definition storage unit, 57 pattern storage unit, 58, 65 learning model storage unit, 59 learning information transmission unit, 62 learning information receiving unit, 64 abnormality symptom detecting unit, 66 detection result transmitting unit.

Claims (11)

An analysis system that detects an abnormality symptom of a plant using a plurality of signals respectively acquired by a plurality of sensors provided in a plant to be monitored,
The plurality of signals are grouped into signal groups using predetermined signal definition information indicating the location of the detection target of the signal and the physical quantity of the detection target, and the signal group is divided between the signals belonging to the signal group. a signal classification unit that classifies using information indicating relationships;
A signal set in which two of the plurality of signals belonging to the signal group are set as one set by using the result classified by the signal classification unit and a rule for generating a signal set predetermined for each classification. a signal group setting unit for setting
a first correlation coefficient calculation unit that calculates, for each of the signal sets set by the signal set setting unit, time-series data of correlation coefficients of the two signals that constitute the signal set in a reference period;
a model generation unit that generates a learning model that includes the time-series data of the correlation coefficient calculated by the first correlation coefficient calculation unit as reference data in detecting the abnormal sign;
A second phase for calculating, for each of the signal sets set by the signal set setting unit, time-series data of correlation coefficients of the two signals forming the signal set among the plurality of signals in the period to be monitored. a relational coefficient calculator;
an abnormality sign detection unit that detects an abnormality sign of the plant using the time-series data of the correlation coefficient calculated by the second correlation coefficient calculation unit and the learning model;
An analysis system comprising: 2. The signal definition information is information in which a device constituting the plant, an internal device constituting the device, and a physical quantity detected by the sensor are associated with the signal. The analysis system described in . 3. The analysis system according to claim 2, wherein the signal definition information further includes information indicating a detailed location where the sensor is installed in the internal device for each signal. The signal classification unit causes the signals having the same device, the internal device, the detailed location, and the physical quantity to belong to the same signal group,
With respect to the signal group composed of the signals having the same device, internal device, detailed location, and physical quantity, the signal group setting unit combines all two signals belonging to the signal group. 4. The analysis system of claim 3, wherein the set of signals of . The signal classification unit causes the signals having the same device, the internal device, and the same physical quantity and having different detailed locations to belong to the same signal group;
With respect to the signal group composed of the signals having the same physical quantity and different detailed locations, the signal set setting unit determines that all of the signals belonging to the signal group are the 4. The analysis system according to claim 3, wherein said signal sets are set such that each said signal belonging to a signal group is included in a predetermined number of said signal sets. The signal name of the signal includes information indicating the equipment that constitutes the plant, the internal equipment that constitutes the equipment, the detailed location where the sensor is installed, and the physical quantity detected by the sensor. , the signal definition information includes the signal name;
The signal classification unit extracts information indicating the device, the internal device, the detailed location, and the physical quantity from the signal name, and groups the signals into the signal groups using the extracted information. The analysis system according to any one of claims 3 to 5. the signal name of the signal includes information indicating the system corresponding to the sensor in the multiple system;
The signal classifying unit causes the signals having the same device, the internal device, the detailed location and the physical quantity and having different information indicating the system to belong to the same signal group,
With respect to the signal group composed of the signals having the same device, internal device, detailed location and physical quantity and different information indicating the system, the signal group setting unit determines whether the signal group belongs to the signal group. 7. The analysis system according to claim 6, wherein said signal sets are set such that each said signal belonging to said signal group is included in a respectively defined number of said signal sets. A learning device that performs learning for detecting signs of abnormality in a plant using a plurality of signals respectively acquired by a plurality of sensors provided in a plant to be monitored,
The plurality of signals are grouped into signal groups using predetermined signal definition information indicating the location of the detection target of the signal and the physical quantity of the detection target, and the signal group is divided between the signals belonging to the signal group. a signal classification unit that classifies using information indicating relationships;
A signal set in which two of the plurality of signals belonging to the signal group are set as one set by using the result classified by the signal classification unit and a rule for generating a signal set predetermined for each classification. a signal group setting unit for setting
a first correlation coefficient calculation unit that calculates, for each of the signal sets set by the signal set setting unit, time-series data of correlation coefficients of the two signals that constitute the signal set in a reference period;
a model generation unit that generates a learning model that includes the time-series data of the correlation coefficient calculated by the first correlation coefficient calculation unit as reference data in detecting the abnormal sign;
A learning device comprising: a plurality of sensors provided in the plant to be monitored;
The analysis system according to any one of claims 1 to 7, wherein a plurality of signals respectively acquired by the plurality of sensors are used to detect signs of abnormality in the plant;
An anomaly sign detection system comprising: An abnormality sign detection method in an analysis system for detecting an abnormality sign of a plant using a plurality of signals respectively acquired by a plurality of sensors provided in a plant to be monitored,
The plurality of signals are grouped into signal groups using predetermined signal definition information indicating the location of the detection target of the signal and the physical quantity of the detection target, and the signal group is divided between the signals belonging to the signal group. a step of performing a classification process of classifying using information indicating relationships;
A step of setting a signal set in which two of the plurality of signals belonging to the signal group constitute one set, using the result of the classification process and a rule for generating a signal set predetermined for each classification. When,
a step of calculating first time-series data, which is time-series data of correlation coefficients of two signals constituting the signal set in a reference period, for each of the set signal sets;
generating a learning model that includes the calculated first time-series data as reference data in detecting the abnormal sign;
a step of calculating second time-series data, which is time-series data of correlation coefficients of two signals constituting the signal set among the plurality of signals in the monitoring target period for each set signal set;
a step of detecting an abnormality sign of the plant using the second time-series data and the learning model;
An anomaly sign detection method, comprising: A computer system that uses a plurality of signals respectively acquired by a plurality of sensors provided in a plant to be monitored to detect signs of abnormality in the plant,
The plurality of signals are grouped into signal groups using predetermined signal definition information indicating the location of the detection target of the signal and the physical quantity of the detection target, and the signal group is divided between the signals belonging to the signal group. a step of performing a classification process of classifying using information indicating relationships;
A step of setting a signal set in which two of the plurality of signals belonging to the signal group constitute one set, using the result of the classification process and a rule for generating a signal set predetermined for each classification. When,
a step of calculating first time-series data, which is time-series data of correlation coefficients of two signals constituting the signal set in a reference period, for each of the set signal sets;
generating a learning model that includes the calculated first time-series data as reference data in detecting the abnormal sign;
a step of calculating second time-series data, which is time-series data of correlation coefficients of two signals constituting the signal set among the plurality of signals in the period to be monitored, for each set signal set;
a step of detecting an abnormality sign of the plant using the second time-series data and the learning model;
A program characterized by causing the execution of

Images