Learning device, abnormal sign detection device, abnormal sign detection system, learning method and program

Abstract

To obtain a learning device capable of shortening the time required from initial inference to abnormality sign detection processing using similar waveform analysis even in a situation where abnormalities occur less frequently.SOLUTION: It is a learning device 7 that generates trained data used to detect abnormality signs, and includes: a learning data acquisition unit 51 that acquires regular time series data as learning data; and a learning unit 52 that determines the similarity of each of a plurality of learning waveforms based on the degree of similarity between the learning waveforms, which are waveforms generated from the learning data, and generates the plurality of leaning data and reference count which is the number of times each of the plurality of learning waveforms was determined to be a similar waveform as learned data.SELECTED DRAWING: Figure 2

Description

[Technical Field]

[0001] The present disclosure relates to a learning device, an abnormal sign detection device, an abnormal sign detection system, a learning method, and a program that perform learning for abnormal sign detection. [Background technology]

[0002] A technology has been proposed for detecting abnormal signs using similar waveform analysis, which analyzes the similarity between waveforms in time-series data. In anomaly detection using similar waveform analysis, the more waveforms there are to be compared, the longer it takes to detect abnormal signs.

[0003] For example, Patent Document 1 below discloses a technology for detecting abnormal vibrations such as earthquakes by detecting vibration waveforms using a vibration sensor that converts vibrations into electrical signals and comparing the detected signal waveform to be inferred with a model signal waveform pattern of a representative seismic wave. In the technology disclosed in Patent Document 1, when the signal waveform to be inferred is compared with the model signal waveform pattern of a seismic wave and determined to be similar to the model signal waveform pattern, the number of times the model signal waveform pattern was determined to be similar is recorded, and the model signal waveform pattern and the signal waveform to be inferred are compared in descending order of the number of times recorded. This makes it possible to shorten the time it takes to detect an abnormality such as an earthquake when it occurs. [Prior art documents] [Patent documents]

[0004] [Patent Document 1] Japanese Patent Application Publication No. 03-291536 Summary of the Invention [Problem to be solved by the invention]

[0005] However, the technology disclosed in Patent Document 1 cannot shorten the time required for similar waveform analysis in a normal state where no abnormality occurs. Therefore, there is a problem that the time required for similar waveform analysis cannot be shortened in a situation where the frequency of abnormalities occurring is low. Furthermore, because the technology uses the number of times that the signal waveform to be inferred and the model signal waveform pattern are determined to be similar in the comparison result, the time required to detect an abnormality cannot be shortened unless the abnormality occurs multiple times.

[0006] The present disclosure has been made in consideration of the above, and aims to provide a learning device that can reduce the time required for abnormal sign detection processing using similar waveform analysis from the initial inference, even in situations where the frequency of abnormalities occurring is low. [Means for solving the problem]

[0007] In order to solve the above-described problems and achieve the object, a learning device according to the present disclosure is a learning device that generates learned data used to detect abnormal signs, and includes a learning data acquisition unit that acquires normal time-series data as learning data, and a similarity calculator that calculates a similarity between learning waveforms that are waveforms generated from the learning data. a threshold value used to detect an abnormality sign is calculated from the plurality of deviation degrees calculated for each of the plurality of learning waveforms; A plurality of learning waveforms; A threshold value, The present invention is characterized by comprising a learning unit that generates, as learned data, the number of references, which is the number of times each of the plurality of learning waveforms has been determined to be a similar waveform. [Effects of the Invention]

[0008] According to the present disclosure, it is possible to reduce the time required for the abnormality sign detection process using similar waveform analysis from the initial inference, even in a situation where the frequency of abnormalities occurring is low. [Brief explanation of the drawings]

[0009] [Figure 1] FIG. 1 is a diagram showing a configuration example of an abnormality sign detection system according to a first embodiment; [Figure 2] FIG. 2 shows the functional configuration of the abnormality symptom detection device shown in FIG. 1. [Figure 3] FIG. 3 is an explanatory diagram of the processing performed by the learning waveform generation unit shown in FIG. 2. [Figure 4] FIG. 3 is a diagram showing an example of information output by the similar waveform determination unit shown in FIG. 2; [Figure 5] FIG. 3 is a diagram showing an example of the number of references generated by the learned data generation unit shown in FIG. 2. [Figure 6] An explanatory diagram of the process in which the abnormal sign determination unit shown in Figure 2 generates an inferred waveform from inferred data. [Figure 7] FIG. 3 is a diagram showing an example of an evaluation screen generated by the learned data evaluation unit shown in FIG. 2. [Figure 8] FIG. 8 is a diagram showing a first example of an operation on the evaluation screen shown in FIG. 7; [Figure 9] FIG. 8 is a diagram showing a second example of an operation on the evaluation screen shown in FIG. 7; [Figure 10] 3 is a flowchart illustrating the operation of the abnormal sign detection device shown in FIG. 2 in the learning phase. [Figure 11] 3 is a flowchart illustrating the operation of the abnormal sign detection device shown in FIG. 2 in the inference phase. [Figure 12] FIG. 1 is a diagram showing an example of the configuration of a computer system that realizes an abnormality sign detection device according to a first embodiment. [Figure 13] FIG. 1 is a diagram illustrating a configuration example of an abnormality sign detection system according to a modification of the first embodiment. DETAILED DESCRIPTION OF THE INVENTION

[0010] Hereinafter, a learning device, an abnormal sign detection device, an abnormal sign detection system, a learning method, and a program according to embodiments of the present disclosure will be described in detail with reference to the accompanying drawings. However, the present invention is not limited to these embodiments.

[0011] Embodiment 1 1 is a diagram illustrating an example configuration of an abnormal sign detection system 1 according to a first embodiment of the present disclosure. The abnormal sign detection system 1 includes a power generation facility 2-1 and a power transmission and distribution facility 2-2 that are monitored, a data collection device 3 that collects data from the monitored facilities, a data storage device 4 that stores the data collected by the data collection device 3, an abnormal sign detection device 5 that detects an abnormal sign of the monitored facilities from the collected data, and a display terminal 6 that displays information related to the abnormal sign detection system 1. The abnormal sign detection system 1 has a function of detecting an abnormal sign of the monitored facilities by performing similar waveform analysis based on data indicating the state of the monitored facilities.

[0012] The data collection device 3 collects data from each of the power generation equipment 2-1 and the power transmission and distribution equipment 2-2 to be monitored, and stores the collected data as time-series data in the data storage device 4. The data collected by the data collection device 3 is data indicating the state of the monitored equipment, such as data acquired by sensors and operation data.

[0013] The data storage device 4 stores the time series data collected by the data collection device 3. The data storage device 4 can provide the stored time series data to the abnormal sign detection device 5 in response to a request from the abnormal sign detection device 5.

[0014] The abnormality sign detection device 5 can detect abnormalities in the power generation facility 2-1 and the power transmission and distribution facility 2-2 that are the monitoring targets, based on the time-series data stored in the data accumulation device 4. The detailed configuration of the abnormality sign detection device 5 will be described later.

[0015] The display terminal 6 has a function of providing information about the abnormal sign detection system 1 to a user of the abnormal sign detection system 1 by displaying a display screen. The display terminal 6 can also output a display screen generated by the abnormal sign detection device 5. For example, the display terminal 6 can display the detection results of the abnormal sign detection device 5. The display terminal 6 can also display an evaluation screen for learning data, which will be described later.

[0016] Fig. 2 is a diagram showing the functional configuration of the abnormal sign detection device 5 shown in Fig. 1. The abnormal sign detection device 5 has a learning data acquisition unit 51, a learning unit 52, a learned data storage unit 53, an inference data acquisition unit 54, an abnormal sign determination unit 55, a detection result storage unit 56, and a learned data evaluation unit 57.

[0017] The learned data acquisition unit 51, the learning unit 52, and the learned data storage unit 53 are also referred to as a learning device 7 that generates learned data used to detect abnormal signs. In the first embodiment, the abnormal sign detection device 5 has the functions of the learning device 7.

[0018] The learning data acquisition unit 51 acquires normal time series data as learning data from the data storage device 4. Here, the normal time series data is defined as time series data obtained when the power generation facility 2-1 and the power transmission and distribution facility 2-2 to be monitored are operating normally. The learning data acquisition unit 51 outputs the acquired learning data to the learning unit 52.

[0019] The learning unit 52 generates learned data used to detect abnormal signs based on the learning data acquired by the learning data acquisition unit 51, and outputs the generated learned data to the learned data storage unit 53. Specifically, the learning unit 52 includes a learning waveform generation unit 521, a similar waveform determination unit 522, a threshold calculation unit 523, and a learned data generation unit 524.

[0020] The learning waveform generation unit 521 generates a learning waveform 11, which is a waveform corresponding to data for each unit interval extracted from the learning data acquired by the learning data acquisition unit 51. FIG. 3 is an explanatory diagram of the processing performed by the learning waveform generation unit 521 shown in FIG. 2. The learning data input to the learning waveform generation unit 521 is, for example, time-series data including timestamps and values ​​associated with the timestamps, as shown in FIG. 3. The learning waveform generation unit 521 determines a waveform corresponding to interval data 10-1, obtained by extracting learning data for a unit time from a certain point in time, as learning waveform 11-1, and determines a waveform corresponding to interval data 10-2, obtained by extracting learning data for a unit time whose start point is shifted by a predetermined time or number of data points from interval data 10-1, as learning waveform 11-2. Here, the learning waveform 11-2 is a waveform corresponding to interval data 10-2, which is extracted from the learning data with its start point shifted by one minute from that of learning waveform 11-1. Similarly, the learning waveform generating unit 521 can set the waveform corresponding to the section data 10-3 extracted from the learning data with the start point shifted by one minute from the learning waveform 11-2 as the learning waveform 11-3.

[0021] The learning waveform generation unit 521 extracts section data from the learning data in the same manner thereafter to generate learning waveforms 11. Hereinafter, when referring to a specific waveform, a hyphen and a number are added after the symbol to distinguish the waveform, and when not referring to a specific waveform, the waveform may be referred to as learning waveform 11. For the sake of explanation, the number used to distinguish each learning waveform 11 here is referred to as learning waveform number i.

[0022] Returning to the explanation of FIG. 2, the similar waveform determination unit 522 determines whether each of the multiple learning waveforms 11 is similar based on the similarity between the multiple learning waveforms 11 generated by the learning waveform generation unit 521. Here, the distance between the waveforms is used as the similarity, with the shorter the distance, the higher the similarity. The distance may be any distance that can indicate the similarity, such as Euclidean distance or DTW (Dynamic Time Warping) distance. Specifically, the similar waveform determination unit 522 calculates the distance between the multiple learning waveforms 11 generated by the learning waveform generation unit 521 in a round-robin manner. The similar waveform determination unit 522 determines the learning waveform 11 with the smallest calculated distance, i.e., the largest similarity, as a similar waveform to that learning waveform 11. Furthermore, for each of the multiple learning waveforms 11, the similar waveform determination unit 522 determines the distance from the learning waveform determined to be similar, i.e., the smallest distance, as the degree of deviation. The similar waveform determination unit 522 outputs to the threshold calculation unit 523 the multiple learning waveforms 11, the learning waveform number i that identifies the determined similar waveform for each of the multiple learning waveforms 11, and the degree of deviation that is the distance from the similar waveform.

[0023] 4 is a diagram showing an example of information output by the similar waveform determination unit 522 shown in FIG. 2. The similar waveform determination unit 522 outputs information in which, for each learned waveform number i, the degree of deviation is associated with a similar waveform number, which is similar waveform information that identifies a similar waveform. Here, the similar waveform number can be the learned waveform number i of the learned waveform 11 determined to be a similar waveform.

[0024] Returning to the explanation of FIG. 2, the threshold calculation unit 523 calculates a threshold used to detect abnormal signs based on the similarity between the learning waveforms 11. Specifically, the threshold calculation unit 523 can calculate a threshold from the degree of outlier calculated for each of the multiple learning waveforms 11. For example, the threshold calculation unit 523 can set the threshold to 3σ of the degree of outlier, that is, three times the standard deviation. The threshold calculation unit 523 outputs the calculated threshold and the information output by the similar waveform determination unit 522 to the learned data generation unit 524.

[0025] The learned data generation unit 524 generates learned data including the multiple learned waveforms 11 output by the learned waveform generation unit 521, a learned waveform number i which is similar waveform information indicating each of the multiple learned waveforms 11 determined to be a similar waveform by the similar waveform determination unit 522, a threshold calculated by the threshold calculation unit 523, and a reference count which is the number of times each of the multiple learned waveforms 11 is determined to be a similar waveform. The learned data generation unit 524 counts the number of times each learned waveform 11 is determined to be a similar waveform based on the similar waveform information output by the similar waveform determination unit 522. Figure 5 is a diagram showing an example of the reference count generated by the learned data generation unit 524 shown in Figure 2. The reference count is associated with the learned waveform number i. The learned data generation unit 524 stores the generated learned data in the learned data storage unit 53.

[0026] The learned data storage unit 53 stores the learned data generated by the learned data generation unit 524. The learned data storage unit 53 can store the learned waveform 11, similar waveform information, a threshold value, and the number of references as learned data.

[0027] The functions of the learning device 7 described above are executed in a learning phase. The learning phase is executed before the inference phase described below, and it is assumed that learned data has already been generated by the time the inference phase is executed.

[0028] The inference data acquisition unit 54 acquires inference data, which is time-series data of the inference target, and outputs the acquired inference data to the abnormal sign determination unit 55.

[0029] The abnormal sign determination unit 55 determines whether or not there are abnormal signs in the inference data by performing similar waveform analysis using the inference waveform 21, which is a waveform generated from the inference data acquired by the inference data acquisition unit 54, and the learning waveform 11 included in the learned data.

[0030] FIG. 6 is an explanatory diagram of the process by which the abnormal sign determination unit 55 shown in FIG. 2 generates the inferred waveform 21 from the inferred data. The inferred data input to the abnormal sign determination unit 55 is, for example, time-series data including timestamps and values ​​associated with the timestamps, as shown in FIG. 6. The abnormal sign determination unit 55 generates the inferred waveform 21 by extracting waveforms for each unit interval from the inferred data. The abnormal sign determination unit 55 determines the waveform corresponding to the section data 20-1, which is inferred data for a unit interval extracted from a certain point in time, as the inferred waveform 21-1, and determines the waveform corresponding to the section data 20-2, which is inferred data for a unit interval extracted by shifting the start point of the section data 20-1 by a predetermined time or number of data points. Here, the inferred waveform 21-2 is the waveform corresponding to the section data 20-2, which is extracted from the inferred data with the start point shifted by one minute from the inferred waveform 21-1. Similarly, the abnormal sign determination unit 55 can determine as the inferred waveform 21-3 the waveform corresponding to the section data 20-3 extracted from the inferred data with the start point shifted by one minute from the inferred waveform 21-2.

[0031] The abnormal sign determination unit 55 extracts section data 20 from the inference data in the same manner thereafter to generate inferred waveforms 21. Hereinafter, when referring to a specific waveform, a hyphen and a number are added after the symbol to distinguish the waveform, and when not referring to a specific waveform, the waveform may be referred to as an inferred waveform 21. For the sake of explanation, the number used to distinguish each inferred waveform 21 here will be referred to as an inferred waveform number j.

[0032] Returning to the explanation of Figure 2, the abnormal sign determination unit 55 compares the inferred waveform 21 generated by the above method with the learned waveform 11 in descending order of the number of times of reference among the multiple learned waveforms 11 included in the learned data, and determines whether or not there is an abnormal sign in the estimated data based on the comparison result. The abnormal sign determination unit 55 stores the determination result in the detection result storage unit 56 as the detection result.

[0033] The detection result storage unit 56 stores the detection result by the abnormality sign determination unit 55.

[0034] The learned data evaluation unit 57 generates an evaluation screen for evaluating the learned data stored in the learned data storage unit 53, and displays the generated evaluation screen on a display screen of the display terminal 6 or the like.

[0035] FIG. 7 is a diagram illustrating an example of an evaluation screen generated by the learned data evaluation unit 57 shown in FIG. 2. The evaluation screen may include a similar waveform list display area, a reference count ranking display area, and a learning data display area. The similar waveform list display area includes a "waveform No." corresponding to the learning waveform number i, the "degree of deviation" of each learning waveform 11, and a "similar waveform No." corresponding to the learning waveform number i of the learning waveform 11 determined to be a similar waveform for each learning waveform 11. The reference count ranking display area includes a "waveform No." corresponding to the learning waveform number i sorted in descending order of the number of references, and the "number of references" of the learning waveform 11 indicated by the "waveform No." Note that, although the table in the reference count ranking display area in FIG. 7 is displayed in descending order of the number of references, it can also be sorted, for example, in descending order of the number of references, or in ascending or descending order of the "waveform No." The learning data display area displays the waveforms of the learning data by representing the learning data as a graph.

[0036] 8 is a diagram showing a first example of an operation on the evaluation screen shown in FIG. 7. When a user performs an operation to select any row of the table displayed in the similar waveform list display area of ​​the evaluation screen shown in FIG. 7, the learning waveforms 11 corresponding to the "waveform No." and "similar waveform No." included in the selected row are displayed in the learning data display area. For example, in the example of FIG. 8, a portion of the waveform of the learning data corresponding to the learning waveform 11-3 with a "waveform No." of "3" and a portion corresponding to the learning waveform 11-13 with a "similar waveform No." of "13" are displayed using hatching to distinguish them from the other portions.

[0037] 9 is a diagram showing a second example of an operation on the evaluation screen shown in FIG. 7. When any row of the table displayed in the reference count ranking display area of ​​the evaluation screen shown in FIG. 7 is selected, the learning waveform 11 corresponding to the "waveform No." included in the selected row and the learning waveform 11 that is the reference source of the learning waveform 11 corresponding to the "waveform No." are displayed in the learning data display area. For example, in the example of FIG. 9, a portion of the waveform of the learning data corresponding to learning waveform 11-13 with "waveform No." "13" and portions corresponding to learning waveforms 11-2 to 11-5 that are the reference sources of learning waveform 11-13, i.e., that are determined to be similar waveforms to learning waveform 11-13, are displayed using hatching to distinguish them from other portions.

[0038] The learned data evaluation unit 57 can also process the learned data stored in the learned data storage unit 53 based on an operation on the generated evaluation screen. For example, if a user looks at the evaluation screen and finds a learned waveform 11 that has been referenced an extremely large number of times, the user can delete other learned waveforms 11 that reference that learned waveform 11 from the learned data, thereby reducing the number of learned waveforms 11 without affecting the accuracy of detecting abnormal signs. When an operation to delete a learned waveform 11 is performed on the evaluation screen, the learned data evaluation unit 57 deletes the learned waveform 11 that is the target of the operation from the learned data storage unit 53.

[0039] Next, the operation of the abnormal sign detection device 5 will be described. Fig. 10 is a flowchart for explaining the operation of the learning phase of the abnormal sign detection device 5 shown in Fig. 2. The learning data acquisition unit 51 acquires, as learning data, time-series data when the power generation facility 2-1 and the power transmission and distribution facility 2-2 to be monitored are operating normally (step S101). The learning waveform generation unit 521 of the learning unit 52 generates a learning waveform 11 from the learning data (step S102). Here, the number of learning waveforms 11 generated by the learning waveform generation unit 521 is defined as N, and the multiple learning waveforms 11 generated by the learning waveform generation unit 521 are referred to as learning waveforms 11-1 to 11-N, respectively.

[0040] The similar waveform determination unit 522 performs the following steps S103 and S104 for each of the N learning waveforms 11-1 to 11-N, where learning waveform number i is 1 to N.

[0041] The similar waveform determination unit 522 calculates the distance between the learning waveform 11-i of learning waveform number i and the learning waveforms 11 other than learning waveform number i in a round-robin manner (step S103). The similar waveform determination unit 522 determines the smallest value of the distances calculated for the learning waveform 11-i in step S103 as the degree of deviation, determines the learning waveform 11 with the smallest distance as the similar waveform to the learning waveform 11-i of learning waveform number i, and records the degree of deviation and the learning waveform number of the similar waveform (step S104).

[0042] After the processes of steps S103 and S104 are completed for each of the learning waveforms 11-1 to 11-N, the threshold calculation unit 523 calculates a threshold from the degree of deviation of all of the learning waveforms 11-1 to 11-N (step S105).

[0043] Next, the learned data generating unit 524 counts the number of times all the learned waveforms 11-1 to 11-N are referred to, and ranks the learned waveforms 11-1 to 11-N in descending order of the number of times they are referred to (step S106).

[0044] The learned data generation unit 524 stores the learned waveforms 11-1 to 11-N generated in step S102, similar waveform information indicating similar waveforms of each learned waveform 11-1 to 11-N determined in step S104, the threshold value calculated in step S105, and the number of times referenced counted in step S106 as learned data in the learned data storage unit 53 (step S107).

[0045] 11 is a flowchart for explaining the operation of the inference phase of the abnormal sign detection device 5 shown in FIG. The inference data acquisition unit 54 acquires time-series data of the inference target as inference data (step S201). The abnormal sign determination unit 55 generates an inference waveform 21 from the inference data (step S202). Here, the number of inference waveforms 21 generated by the abnormal sign determination unit 55 is set to M, and the multiple inference waveforms 21 generated by the abnormal sign determination unit 55 are referred to as inference waveforms 21-1 to 21-M, respectively.

[0046] The abnormal sign determination unit 55 performs the following steps S203 to S210 for each of the inferred waveforms 21-1 to 21-M whose inferred waveform numbers j are 1 to M.

[0047] First, the abnormal sign determination unit 55 sets the rank number n to 1 (step S203). Note that here, the smaller the rank number n, the more times it has been referred to. The abnormal sign determination unit 55 acquires the nth ranked learning waveform 11 from the learned data (step S204). Note that here, the nth ranked learning waveform 11 is referred to as learning waveform 11-n. The abnormal sign determination unit 55 calculates the distance between the inferred waveform 21-j and the learning waveform 11-n (step S205).

[0048] The abnormal sign determination unit 55 determines whether the distance calculated in step S205 is equal to or less than a threshold (step S206). If the distance is equal to or less than the threshold (step S206: Yes), the abnormal sign determination unit 55 determines that the target inferred waveform 21 is normal (step S207). If the distance is not equal to or less than the threshold (step S206: No), the abnormal sign determination unit 55 determines whether n is equal to or greater than the number of learned waveforms N (step S208). If n is equal to or greater than the number of learned waveforms N (step S208: Yes), the abnormal sign determination unit 55 determines that the target inferred waveform has an abnormal sign (step S209). If n is not equal to or greater than the number of learned waveforms N (step S208: No), the abnormal sign determination unit 55 increments n (step S210) and repeats the process from step S204.

[0049] When the processes from step S203 to step S210 are completed for each of the inferred waveforms 21-1 to 21-M, the abnormal sign determination unit 55 stores the determination results of all the inferred waveforms 21-1 to 21-M as detection results in the detection result storage unit 56 (step S211).

[0050] Next, a hardware configuration of the abnormal sign detection device 5 of this embodiment will be described. In the abnormal sign detection device 5 of this embodiment, a computer program in which processing in the abnormal sign detection device 5 is described is executed on a computer system, causing the computer system to function as the abnormal sign detection device 5. FIG. 12 is a diagram showing an example configuration of a computer system that realizes the abnormal sign detection device 5 of the first embodiment. As shown in FIG. 12, this computer system includes a control unit 101, an input unit 102, a storage unit 103, a display unit 104, a communication unit 105, and an output unit 106, which are connected via a system bus 107.

[0051] In FIG. 12 , the control unit 101 is a processor such as a CPU (Central Processing Unit) and executes a program describing the processing performed by the abnormal sign detection device 5 of this embodiment. Note that a portion of the control unit 101 may be implemented using dedicated hardware such as a GPU (Graphics Processing Unit) or an FPGA (Field-Programmable Gate Array). The input unit 102 is configured, for example, by a keyboard, a mouse, etc., and is used by a user of the computer system to input various pieces of information. The storage unit 103 includes various types of memory, such as RAM (Random Access Memory) and ROM (Read Only Memory), and a storage device, such as a hard disk, and stores programs to be executed by the control unit 101, necessary data obtained during processing, etc. The storage unit 103 is also used as a temporary storage area for programs. The display unit 104 is configured by a display, an LCD (Liquid Crystal Display Panel), etc., and displays various screens to the user of the computer system. The communication unit 105 is a receiver and transmitter that performs communication processing. The output unit 106 is a printer, a speaker, etc. Note that Fig. 12 is just an example, and the configuration of the computer system is not limited to the example of Fig. 12. For example, in this embodiment, the abnormal sign detection system 1 has the display terminal 6, and therefore the abnormal sign detection device 5 does not need to have the display unit 104.

[0052] Here, an example of the operation of the computer system until the program of this embodiment is ready to be executed will be described. In the computer system having the above configuration, for example, a computer program is installed in storage unit 103 from a CD-ROM or DVD-ROM inserted in a CD (Compact Disc)-ROM drive or DVD (Digital Versatile Disc)-ROM drive (not shown). 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 abnormal sign detection device 5 of this embodiment in accordance with the program stored in storage unit 103.

[0053] In the above explanation, a program describing the processing in the abnormal sign detection device 5 is provided on a CD-ROM or DVD-ROM as a recording medium, but this is not limiting, and depending on the configuration of the computer system, the capacity of the program to be provided, etc., it is also possible to use a program provided over a transmission medium such as the Internet via the communication unit 105.

[0054] The program of this embodiment causes, for example, a computer system that generates learned data used to detect abnormal signs to execute the following steps: acquiring normal time series data as learned data; determining similar waveforms of each of a plurality of learned waveforms 11 based on the similarity between the learned waveforms 11, which are waveforms generated from the learned data; and generating, as learned data, the plurality of learned waveforms 11 and the number of times each of the plurality of learned waveforms 11 has been determined to be a similar waveform, which is the number of times each of the plurality of learned waveforms 11 has been referenced.

[0055] The learning unit 52, the abnormal sign determination unit 55, and the learned data evaluation unit 57 shown in FIG. 2 are realized by the control unit 101 shown in FIG. 12 executing a computer program stored in the storage unit 103 shown in FIG. 12. The storage unit 103 shown in FIG. 12 is also used to realize the learning data acquisition unit 51, the learning unit 52, the inference data acquisition unit 54, the abnormal sign determination unit 55, and the learned data evaluation unit 57 shown in FIG. 2. The learned data storage unit 53 and the detection result storage unit 56 shown in FIG. 2 are part of the storage unit 103 shown in FIG. 12. The learning data acquisition unit 51 and the inference data acquisition unit 54 shown in FIG. 2 are realized by the communication unit 105 and the control unit 101 shown in FIG. 12. The abnormal sign detection device 5 may also be realized by multiple computer systems. For example, the abnormal sign detection device 5 may be realized by a cloud computer system.

[0056] Moreover, each of the data collection device 3, data storage device 4, and display terminal 6 shown in FIG. 1 can be realized by the computer system shown in FIG.

[0057] 2 is just an example, and the division of functions in each device is not limited to the above example as long as the abnormal sign detection system 1 can perform the above-mentioned operations. For example, in the example shown in Fig. 2, the learning device 7 is provided in the abnormal sign detection device 5, but as shown in the following modified example, the learning device 7 may be provided in a device different from the abnormal sign detection device 5.

[0058] <Modification> FIG. 13 is a diagram illustrating an example configuration of an abnormal sign detection system 8 according to a modified example of the first embodiment. The abnormal sign detection system 8 includes a learning device 7 and an abnormal sign detection device 5-1. The function of the learning device 7 is similar to that of the example shown in FIG. 2, and therefore a description thereof will be omitted here. The abnormal sign detection device 5-1 includes an inference data acquisition unit 54 and an abnormal sign determination unit 55. The functions of the inference data acquisition unit 54 and the abnormal sign determination unit 55 are similar to those of the example shown in FIG. 2, and therefore a description thereof will be omitted here.

[0059] As shown in FIG. 13, the learning device 7 may be a device separate from the abnormal sign detection device 5-1. The separation of the functions of the learning device 7 and the abnormal sign detection device 5-1 shown here is merely an example, and the present embodiment is not limited to this example. For example, in the examples of FIGS. 2 and 13, the learned data storage unit 53 is provided in the learning device 7, but the learned data storage unit 53 may be provided in a device different from the learning device 7. Furthermore, although the learned data evaluation unit 57 is omitted in FIG. 13, the abnormal sign detection system 8 may include the learned data evaluation unit 57 in the learning device 7, or the function of the learned data evaluation unit 57 may be implemented in a device different from the learning device 7. Furthermore, each function of the learning unit 52 may be implemented in a different device.

[0060] As described above, this embodiment provides a learning device 7 that generates learned data used to detect abnormal signs. The learning device 7 includes a learning data acquisition unit 51 that acquires normal time-series data as learning data, and a learning unit 52 that determines whether a plurality of learning waveforms are similar to each other based on the similarity between the learning waveforms, which are waveforms generated from the learning data, and generates learned data including the plurality of learning waveforms and a reference count, which is the number of times each of the plurality of learning waveforms is determined to be a similar waveform. This enables abnormal signs to be detected based on the reference count, which is the number of times each of the plurality of learning waveforms 11 generated from normal time-series data is determined to be a similar waveform. Specifically, the abnormal sign detection device 5 performs similar waveform analysis in descending order of the reference count, thereby reducing the time required to determine that a waveform is similar to a normal learning waveform. Therefore, in abnormal sign detection systems 1 and 8 that determine a waveform as normal when a similar waveform is found through waveform analysis similar to a normal learning waveform, the time required for abnormal sign detection processing using similar waveform analysis from the initial inference can be reduced even in situations where abnormalities occur infrequently.

[0061] The learning unit 52 can also generate, as learned data, similar waveform information indicating waveforms similar to each of the multiple learned waveforms 11. By saving not only the number of references generated from the similar waveform information but also similar waveform information such as that shown in Fig. 4 as learned data, it becomes possible to refer to which waveforms have been determined to be similar waveforms after the learned data has been generated, and therefore it becomes possible to use the similar waveform information when evaluating and updating the learned data.

[0062] The learned data can also be evaluated and updated using an evaluation screen generated by the learned data evaluation unit 57. The learned data evaluation unit 57 generates an evaluation screen for evaluating the learned data. This evaluation screen can include the number of times each of the multiple learned waveforms 11 has been referenced. The evaluation screen can also display the multiple learned waveforms 11 in order of the number of times they have been referenced. The evaluation screen can also include similar waveform information that indicates similar waveforms of each of the multiple learned waveforms. The learned data evaluation unit 57 can also have a function for processing the learned data based on an operation on the generated evaluation screen.

[0063] According to this embodiment, it is possible to provide an abnormal sign detection system 1 in which the learning device 7 is provided in an abnormal sign detection device 5 that detects abnormal signs using learned data, or it is possible to provide an abnormal sign detection system 8 in which the learning device 7 that generates the learned data and the abnormal sign detection device 5-1 that detects abnormal signs using the learned data are different devices.

[0064] The abnormal sign detection device 5, which has the functions of the learning device 7, includes a learning data acquisition unit 51 that acquires normal time series data as learning data, a learning unit 52 that determines similar waveforms of each of a plurality of learning waveforms 11 based on the similarity between the learning waveforms 11, which are waveforms generated from the learning data, and generates learned data including the plurality of learning waveforms 11 and the number of times each of the plurality of learning waveforms 11 has been determined to be a similar waveform, an inference data acquisition unit 54 that acquires inference data, which is the time series data to be inferred, and an abnormal sign determination unit 55 that determines whether or not there is an abnormal sign in the inference data by performing similar waveform analysis using an inference waveform 21, which is a waveform generated from the inference data, and the learning waveform 11, and the abnormal sign determination unit 55 compares the inference waveform 21 with the learning waveform 11 in descending order of the number of times it has been referenced among the learning waveforms 11 included in the learned data.

[0065] The abnormal sign detection system 8 also includes a learning device 7 that generates learned data used to detect abnormal signs, and an abnormal sign detection device 5-1 that detects abnormal signs using the learned data. The learning device 7 includes a learning data acquisition unit 51 that acquires normal time series data as learning data, and a learning unit 52 that determines similar waveforms of each of a plurality of learning waveforms 11 based on the similarity between the learning waveforms 11, which are waveforms generated from the learning data, and generates, as learned data, the plurality of learning waveforms 11 and the number of times each of the plurality of learning waveforms 11 has been determined to be a similar waveform. The abnormal sign detection device 5-1 includes an inference data acquisition unit 54 that acquires inference data, which is time series data to be inferred, and an abnormal sign determination unit 55 that performs similar waveform analysis using the inference waveform, which is a waveform generated from the inference data, and the learning waveform, to determine whether the inference data contains an abnormal sign. The abnormal sign determination unit 55 compares the inference waveform 21 with the learning waveform 11 in descending order of the number of times it has been referenced among the learning waveforms 11 included in the learned data.

[0066] In addition, the learning unit 52 of the learning device 7 further generates a threshold value used to detect abnormal signs as learned data based on the similarity between the learning waveforms 11, and the abnormal sign determination unit 55 of the abnormal sign detection device 5, 5-1 terminates the similar waveform analysis when it finds a waveform similar to the inferred waveform 21 among the learning waveforms 11 using the similarity between the inferred waveform 21 and the learning waveform 11 and the threshold value.

[0067] The configurations shown in the above embodiments are examples of the contents of the present disclosure, and may be combined with other known technologies, and parts of the configurations may be omitted or modified within the scope of the gist of the present disclosure.

[0068] For example, in the above embodiment, the monitored objects are the power generation facility 2-1 and the power transmission and distribution facility 2-2. However, the monitored objects may be at least one of the power generation facility 2-1 and the power transmission and distribution facility 2-2, other facilities in the power grid, facilities in various plants, other devices, or natural features such as rivers. The monitored objects are not limited as long as the data collection device 3 can collect time-series data from the monitored objects. The time-series data may be, for example, data acquired by a sensor, an operation history that allows operation input to the monitored facilities, or operational data of the monitored facilities. For example, if the monitored object is a river, the time-series data may be the water level of the river acquired by a sensor.

[0069] Various aspects of the present disclosure are summarized below as appendices.

[0070] (Appendix 1) A learning device that generates learned data used to detect abnormal signs, a learning data acquisition unit that acquires normal time series data as learning data; a learning unit that determines whether a plurality of learning waveforms are similar to each other based on the similarity between the learning waveforms, which are waveforms generated from the learning data, and generates the plurality of learning waveforms and a reference count, which is the number of times each of the plurality of learning waveforms is determined to be a similar waveform, as the learned data; A learning device comprising: (Appendix 2) The learning device according to claim 1, wherein the learning unit further generates similar waveform information indicating the similar waveforms of each of the plurality of learning waveforms as the learned data. (Appendix 3) The learning device described in Appendix 1 or 2, characterized in that the learning unit further generates a threshold value used to detect the abnormal sign as the learned data based on the similarity between the learning waveforms. (Appendix 4) The learning unit For each of the plurality of learning waveforms, the similarity between the learning waveforms is calculated in a round-robin manner, the learning waveform with the largest similarity is determined to be the similar waveform, and the similarity between the learning waveform and the similar waveform is determined to be the deviation degree of the learning waveform; 4. The learning device according to claim 3, wherein the threshold value is calculated from a plurality of degrees of deviation calculated for each of a plurality of the learning waveforms. (Appendix 5) a learned data evaluation unit that generates an evaluation screen for evaluating the learned data; 5. The learning device according to claim 1, further comprising: (Appendix 6) 6. The learning device according to claim 5, wherein the evaluation screen includes the number of times each of the plurality of learning waveforms has been referenced. (Appendix 7) 7. The learning device according to claim 6, wherein the evaluation screen displays the plurality of learning waveforms in order of the number of times they have been referenced. (Appendix 8) 8. The learning device according to claim 5, wherein the evaluation screen includes similar waveform information indicating the similar waveforms of each of the plurality of learning waveforms. (Appendix 9) The learning device described in any one of appendices 5 to 8, characterized in that the learned data evaluation unit processes the learned data based on operations on the evaluation screen. (Appendix 10) 10. The learning device according to any one of claims 1 to 9, wherein the learning data is time-series data indicating the state of at least one of a power generation facility and a power transmission and distribution facility. (Appendix 11) An abnormality sign detection device that detects abnormality signs using learned data, a learning data acquisition unit that acquires normal time series data as learning data; a learning unit that determines whether a plurality of learning waveforms are similar to each other based on the similarity between the learning waveforms, which are waveforms generated from the learning data, and generates the plurality of learning waveforms and a reference count, which is the number of times each of the plurality of learning waveforms is determined to be a similar waveform, as the learned data; an inference data acquisition unit that acquires inference data, which is time-series data to be inferred; an abnormality sign determination unit that determines whether or not there is an abnormality sign in the inference data by performing a similar waveform analysis using an inference waveform, which is a waveform generated from the inference data, and the learning waveform; Equipped with The abnormal sign detection device is characterized in that the abnormal sign judgment unit compares the inferred waveform with the learning waveform in descending order of the number of times the learning waveform has been referenced among the learning waveforms included in the learned data. (Appendix 12) a learning device that generates learned data used to detect abnormal signs; an abnormality sign detection device that detects an abnormality sign using the learned data; Equipped with The learning device a learning data acquisition unit that acquires normal time series data as learning data; a learning unit that determines whether a plurality of learning waveforms are similar to each other based on the similarity between the learning waveforms, which are waveforms generated from the learning data, and generates the plurality of learning waveforms and a reference count, which is the number of times each of the plurality of learning waveforms is determined to be a similar waveform, as the learned data; and The abnormality symptom detection device an inference data acquisition unit that acquires inference data, which is time-series data to be inferred; an abnormality sign determination unit that determines whether or not the inference data has an abnormality sign by performing a similar waveform analysis using an inference waveform, which is a waveform generated from the inference data, and the learning waveform; Equipped with The abnormal sign detection system is characterized in that the abnormal sign judgment unit compares the inferred waveform with the learning waveform in descending order of the number of times the learning waveform has been referenced among the learning waveforms included in the learned data. (Appendix 13) the learning unit further generates, as the learned data, a threshold value used for detecting the abnormal sign, based on the similarity between the learning waveforms; The abnormal sign detection system described in Appendix 12, characterized in that the abnormal sign determination unit terminates the similar waveform analysis when a waveform similar to the inferred waveform is found among the learning waveforms using the similarity between the inferred waveform and the learning waveform and the threshold value. (Appendix 14) The abnormality symptom detection device a learned data evaluation unit that generates an evaluation screen for evaluating the learned data; and a display terminal that displays the evaluation screen; 14. The abnormal sign detection system according to claim 12 or 13, further comprising: (Appendix 15) A learning method for a learning device that generates learned data used to detect abnormal signs, comprising: A step of acquiring normal time series data as learning data; determining similar waveforms of each of the plurality of training waveforms based on the similarity between the training waveforms, which are waveforms generated from the training data; generating, as the learned data, a plurality of the learning waveforms and a reference count, which is the number of times each of the plurality of learning waveforms is determined to be the similar waveform; A learning method comprising: (Appendix 16) A computer system that generates trained data used to detect abnormal signs, A step of acquiring normal time series data as learning data; determining similar waveforms of each of the plurality of training waveforms based on the similarity between the training waveforms, which are waveforms generated from the training data; generating, as the learned data, a plurality of the learning waveforms and a reference count, which is the number of times each of the plurality of learning waveforms is determined to be the similar waveform; A program characterized by executing the following. [Explanation of symbols]

[0071] 1,8 Abnormal sign detection system, 2-1 Power generation equipment, 2-2 Power transmission and distribution equipment, 3 Data collection device, 4 Data storage device, 5,5-1 Abnormal sign detection device, 6 Display terminal, 7 Learning device, 10-1 to 10-3, 20, 20-1 to 20-3 Section data, 11, 11-1 to 11-N Learning waveform, 21, 21-1 to 21-M Inference waveform, 51 Learning data acquisition unit, 52 Learning unit, 53 Learned data storage unit, 54 Inference data acquisition unit, 55 Abnormal sign judgment unit, 56 Detection result storage unit, 57 Learned data evaluation unit, 101 Control unit, 102 Input unit, 103 Storage unit, 104 Display unit, 105 Communication unit, 106 Output unit, 107 System bus, 521 Learning waveform generation unit, 522 Similar waveform judgment unit, 523 Threshold calculation unit, 524 Learned data generation unit.

Claims

1. A learning device that generates learned data used to detect abnormal signs, a learning data acquisition unit that acquires normal time series data as learning data; a learning unit that calculates similarities between learning waveforms that are waveforms generated from the learning data, determines the learning waveform with the greatest similarity as a similar waveform for each of the learning waveforms, sets the similarity to the similar waveform as an outlier of the learning waveform, calculates a threshold value used to detect the abnormal sign from the outliers calculated for each of the plurality of learning waveforms, and generates the plurality of learning waveforms, the threshold value, and a reference count that is the number of times each of the plurality of learning waveforms is determined to be a similar waveform as the learned data; A learning device comprising:

2. 2. The learning device according to claim 1, wherein the learning unit further generates similar waveform information indicating the similar waveforms of each of the plurality of learning waveforms as the learned data.

3. a learned data evaluation unit that generates an evaluation screen for evaluating the learned data; The learning device according to claim 1 , further comprising:

4. 4. The learning device according to claim 3, wherein the evaluation screen includes the number of references for each of the plurality of learning waveforms.

5. 4. The learning device according to claim 3, wherein the evaluation screen displays the plurality of learning waveforms in order of the number of times they have been referred to.

6. 4. The learning device according to claim 3, wherein the evaluation screen includes similar waveform information indicating the similar waveforms of each of the plurality of learning waveforms.

7. The learning device according to claim 3 , wherein the learned data evaluation unit processes the learned data based on an operation on the evaluation screen.

8. The learning device according to claim 1 , wherein the learning data is time-series data indicating the state of at least one of a power generation facility and a power transmission and distribution facility.

9. An abnormal sign detection device that detects abnormal signs using learned data, a learning data acquisition unit that acquires normal time series data as learning data; a learning unit that calculates similarities between learning waveforms that are waveforms generated from the learning data, determines the learning waveform with the greatest similarity as a similar waveform for each of the learning waveforms, sets the similarity to the similar waveform as an outlier of the learning waveform, calculates a threshold value used to detect the abnormal sign from the outliers calculated for each of the plurality of learning waveforms, and generates the plurality of learning waveforms, the threshold value, and a reference count that is the number of times each of the plurality of learning waveforms is determined to be a similar waveform as the learned data; an inference data acquisition unit that acquires inference data, which is time-series data to be inferred; an abnormality sign determination unit that determines whether or not there is an abnormality sign in the inference data by performing a similar waveform analysis using an inference waveform, which is a waveform generated from the inference data, the learning waveform, and the threshold value; Equipped with The abnormal sign detection device is characterized in that the abnormal sign judgment unit compares the inferred waveform with the learning waveform in descending order of the number of times the learning waveform has been referenced among the learning waveforms included in the learned data.

10. a learning device that generates learned data used to detect abnormal signs; an abnormality sign detection device that detects an abnormality sign using the learned data; Equipped with The learning device a learning data acquisition unit that acquires normal time series data as learning data; a learning unit that calculates similarities between learning waveforms that are waveforms generated from the learning data, determines the learning waveform with the greatest similarity as a similar waveform for each of the learning waveforms, sets the similarity to the similar waveform as an outlier of the learning waveform, calculates a threshold value used to detect the abnormal sign from the outliers calculated for each of the plurality of learning waveforms, and generates the plurality of learning waveforms, the threshold value, and a reference count that is the number of times each of the plurality of learning waveforms is determined to be a similar waveform as the learned data; and The abnormality symptom detection device an inference data acquisition unit that acquires inference data, which is time-series data to be inferred; an abnormality sign determination unit that determines whether or not the inference data has an abnormality sign by performing a similar waveform analysis using an inference waveform, which is a waveform generated from the inference data, the learning waveform, and the threshold value; Equipped with The abnormal sign detection system is characterized in that the abnormal sign determination unit compares the inferred waveform with the learning waveform in descending order of the number of times the learning waveform has been referenced among the learning waveforms included in the learned data.

11. The abnormal sign detection system described in Claim 10, characterized in that the abnormal sign judgment unit terminates the similar waveform analysis when a waveform similar to the inferred waveform is found among the learning waveforms using the similarity between the inferred waveform and the learning waveform and the threshold value.

12. The abnormality symptom detection device a learned data evaluation unit that generates an evaluation screen for evaluating the learned data; and a display terminal that displays the evaluation screen; The abnormal sign detection system according to claim 10 or 11, further comprising:

13. A learning method for a learning device that generates learned data used to detect abnormal signs, comprising: A step of acquiring normal time series data as learning data; calculating a similarity between training waveforms which are waveforms generated from the training data; determining, for each of the learning waveforms, the learning waveform having the greatest degree of similarity as a similar waveform; a step of setting the degree of similarity with the similar waveform as an outlier of the learning waveform, and calculating a threshold value used for detecting the abnormal sign from the plurality of outliers calculated for each of the plurality of learning waveforms; generating, as the learned data, the plurality of learning waveforms, the threshold value, and a reference count, which is the number of times each of the plurality of learning waveforms is determined to be the similar waveform; A learning method comprising:

14. A computer system that generates trained data used to detect abnormal signs, A step of acquiring normal time series data as learning data; calculating a similarity between training waveforms which are waveforms generated from the training data; determining, for each of the learning waveforms, the learning waveform having the greatest degree of similarity as a similar waveform; a step of setting the degree of similarity with the similar waveform as an outlier of the learning waveform, and calculating a threshold value used for detecting the abnormal sign from the plurality of outliers calculated for each of the plurality of learning waveforms; generating, as the learned data, the plurality of learning waveforms, the threshold value, and a reference count, which is the number of times each of the plurality of learning waveforms is determined to be the similar waveform; A program characterized by executing the following.

Images