JP5150590B2 - Abnormality diagnosis apparatus and abnormality diagnosis method - Google Patents

Description

  The present invention relates to an abnormality diagnosis apparatus that detects an abnormality of a plant.

  The plant abnormality diagnosis device detects the occurrence of an abnormality or an accident based on measurement data from the plant when an abnormal transient or accident occurs in the plant.

  Patent Document 1 discloses a diagnostic apparatus using an adaptive resonance theory (ART) which is one of clustering techniques. ART is a method of classifying input data into a plurality of categories (= clusters). Here, the category is a group of data having similarities.

  In an abnormality diagnosis technique using ART, first, normal operation data is input to the ART to create a normal category. This operation is called learning. Next, the measured value of the plant is input to the ART. If the measured value has the same characteristics as the driving data used for learning, the measured value is classified into the category generated during learning. If the characteristics are different, classify them into a new category (new category) that is different from when learning. Finally, the rate of occurrence of new categories is calculated, and when this exceeds a threshold, it is determined as abnormal.

JP 2005-165375 A

  In the clustering technique, there is a parameter that determines the size of a cluster (the size of a category in ART). This parameter is called a resolution parameter. Generally, when classifying certain data into clusters, if the resolution is set coarsely, the number of clusters decreases, and if the resolution is set finely, the number of clusters increases.

  When clustering is used for anomaly diagnosis, if the resolution is rough, even if an anomaly occurs, the trend of the data cannot be detected. Conversely, if the resolution is fine, even if the resolution is fine, a minute trend change is regarded as abnormal. There is a possibility to judge.

  An object of the present invention is to provide an abnormality diagnosis device that can appropriately adjust a resolution parameter so as to meet the purpose of abnormality diagnosis and detect an abnormality with high accuracy.

  In order to achieve the above object, the abnormality diagnosis apparatus of the present invention has the following configuration. That is, a reference signal determining means for creating a reference signal based on a measurement signal at the time of normal plant, a classification means for classifying the reference signal into a category according to data similarity, and a current measurement signal being the classification means In the abnormality diagnosis apparatus provided with a diagnosis means for diagnosing an abnormality when the quantity that does not belong to the category that has occurred exceeds a certain value, the reference signal determination means uses two or more data as a measurement signal when the plant is normal A learning unit that has a function of dividing into groups, wherein the classifying unit classifies into a category using one of the data groups, and a test unit that classifies into a category using one of the remaining data groups not used in the learning unit; And a resolution parameter adjusting unit that adjusts a resolution parameter used for classification into categories using classification results of the learning unit and the test unit.

  By using the abnormality diagnosis apparatus of the present invention, the resolution parameter can be optimally adjusted, so that abnormality diagnosis can be performed with high accuracy. In addition, the labor required for adjusting the resolution parameter can be reduced.

It is a block diagram which shows the 1st Example of the abnormality diagnosis apparatus of this invention. It is a figure explaining one Example of the method classified into a category, and one Example of a classification result in a diagnostic means, a learning part, and a test part. It is a flowchart figure which shows the procedure after acquiring the measurement signal of a plant until it displays abnormal information on an image display apparatus. It is a figure explaining the Example of a plant, and the operation example when the abnormality diagnosis apparatus of this invention is applied to a plant. It is a figure explaining the example of failure of abnormality diagnosis. It is a figure explaining the operation mode and aged deterioration of a gas turbine plant. It is a figure explaining the operation example of a reference signal determination means. It is a figure explaining operation | movement of a classification | category means. It is a figure explaining resolution parameter adjustment progress. It is a resolution parameter confirmation screen, and is an example of a screen displayed on the image display device. It is a figure explaining the response | compatibility method to aged deterioration. It is a block diagram which shows the 2nd Example of the abnormality diagnosis apparatus of this invention. It is a figure explaining the example of an operation result of the reference signal determination means in 2nd Example of the abnormality diagnosis apparatus of this invention. It is a flowchart figure explaining operation | movement of the classification | category means in 2nd Example of the diagnostic apparatus of this invention. It is a figure explaining the aspect of the data preserve | saved in the classification result database in 2nd Example of the abnormality diagnosis apparatus of this invention. It is a figure explaining one Example of the screen displayed on the image display apparatus in 2nd Example of the abnormality diagnosis apparatus of this invention.

  The abnormality diagnosis apparatus of the present invention will be described with reference to the drawings.

  FIG. 1 is a block diagram showing a first embodiment of the abnormality diagnosis apparatus of the present invention. The plant 100 is diagnosed with the abnormality diagnosis device 200.

  The abnormality diagnosis apparatus 200 includes a reference signal determination unit 400, a data classification unit 500, and a diagnosis unit 600 as arithmetic units. Moreover, the abnormality diagnosis apparatus 200 includes a measurement signal database 310, a reference signal database 320, and a classification result database 330 as databases.

  The abnormality diagnosis apparatus 200 includes an external input interface 210 and an external output interface 220 as interfaces with the outside.

  The abnormality diagnosis device 200 can be operated by operating the external input device 900 including the measurement signal 1 obtained by measuring various state quantities of the plant from the plant 100 and the keyboard 910 and the mouse 920 via the external input interface 210. The external input signal 2 to be created is input. Also, the image display data 11 is output from the abnormality diagnosis device 200 to the image display device 950 via the external output interface 220.

  The measurement signal 3 input via the external input interface 210 is stored in the measurement signal database 310. The reference signal determining unit 400 extracts normal data from the measurement signal 5 stored in the measurement signal database 310 and divides the normal data into two or more data groups in terms of time. This divided data group is called a reference signal, and the reference signal 6 is stored in the reference signal database 320.

  The data classification unit 500 processes the reference signal 7 stored in the reference signal database 320 and outputs a classification result 8. In the data classification unit 500, the learning unit 510, the test unit 520, and the resolution parameter adjustment unit 530 are operated repeatedly for a predetermined number of times, thereby adjusting the resolution parameter and creating the reference signal classification result 8. The learning unit 510 classifies one data group (learning data) included in the reference signal into a category (= cluster). The classification result 501 obtained by processing the learning data by the learning unit 510 is transmitted to the test unit 520. The test unit 520 uses the learning data classification result 501 to classify data groups (test data) that were not used by the learning unit into categories. The classification result 502 obtained by processing the learning data and the test data by the test unit 520 is transmitted to the resolution parameter adjustment unit 530. The resolution parameter adjustment unit 530 adjusts the resolution parameter using the classification result 502 of the learning data and the test data, and transmits the adjusted resolution parameter value 503 to the learning unit 510 and the test unit 520. The learning unit 510 and the test unit 520 classify the learning data and the test data into categories using the resolution parameter values.

  The classification result 8 is stored in the classification result database 330. The diagnosis unit 600 processes the current measurement signal 4 stored in the measurement signal database 310 and the classification result 9 stored in the classification result database 330. The current measurement signal 4 is processed by the learning unit and the test unit. Diagnose whether the plant is normal or abnormal based on the quantity that does not belong to the category that occurred. The diagnosis unit 600 outputs the diagnosis result as the diagnosis result 10 to the external output interface 220. The external output interface 220 converts the diagnosis result 10 into the image display data 11 and outputs it to the image display device 950. The quantity that does not belong to the category used for diagnosis is the ratio of the quantity that does not belong to the category divided by the quantity that belongs to the category in order to grasp the quantity that does not belong to the category, or whether there is a quantity that does not belong to the category that is 1 as the quantity Including.

  Further, diagnostic device information 50 including data stored in the measurement signal database 310, the reference signal database 320, and the classification result database 330 can be displayed on the image display device 950. Moreover, these data can also be corrected using the external input device 900 as needed.

  In this embodiment, the reference signal determining means 400, the data classification means 500, the diagnosis means 600, the measurement signal database 310, the reference signal database 320, and the classification result database 330 are all inside the abnormality diagnosis apparatus 200. A part may be arranged outside the abnormality diagnosis apparatus 200 so that only data is communicated. In this embodiment, there is one plant to be diagnosed, but a plurality of plants can be diagnosed by the abnormality diagnosis apparatus 200.

  FIG. 2A is a diagram for explaining an embodiment of a method of classifying into categories in the diagnostic unit 600, the learning unit 510, the test unit 520, and the resolution parameter adjustment unit 530. Here, an embodiment in which adaptive resonance theory (ART) is applied will be described, but other clustering methods such as vector quantization may be used.

  As shown in FIG. 2A, the ART is composed of a data preprocessing device 610 and an ART module 620.

  The data preprocessing device 610 converts the operation data into input data for the ART module 620. The steps will be described below.

  First, the maximum value and the minimum value are calculated for each measurement item. Data is normalized using the calculated maximum and minimum values.

The normalization method will be described by taking xi as an example. The number of data of xi is N and the nth measurement value is xi (n). Further, when the maximum value and the minimum value in N data are Maxi and Min_i, respectively, normalized data Nxi (n) is calculated by the following equation.
[Equation 1]
Nxi (n) = α + (1−α) × (xi (n) −Min_i)
/ (Max_i-Mini) (Formula 1)

Here, it is a constant of α (0 ≦ α <0.5), and the data is normalized in the range of [α, 1−α] by Equation 1. In the present embodiment, α = 0.2. The complement CNxi (n) of the normalized data Nxi (n) is calculated by the following equation.
[Equation 2]
CNxi (n) = 1−Nxi (n) (Formula 2)

  Next, data consisting of data Nxi (n) and CNxi (n) is input to the ART module 620 as input data.

  The ART module 620 classifies input data into a plurality of categories. The ART module 620 includes an F0 layer 621, an F1 layer 622, an F2 layer 623, a memory 624, and an orientation subsystem 625, which are coupled to each other. The F1 layer 622 and the F2 layer 623 are coupled via a weighting factor, and the weighting factor represents a prototype (prototype) of a category into which input data is classified.

  Next, the algorithm of the ART module will be described. The outline of the algorithm when input data is input to the ART module is as shown in the following steps 1 to 5.

  Step 1: The F0 layer is normalized so that the size of the input vector becomes 1, and noise is removed.

  Step 2: A suitable category candidate is selected by comparing the input data input to the F1 layer with a weighting factor.

  Step 3: The validity of the category selected in the Orienting Subsystem is evaluated by the ratio with the parameter ρ. If it is determined to be valid, the input data is classified into the category, and the process proceeds to step 4. If it is not judged to be valid, the category is reset, and an appropriate category candidate is selected from the other categories (the process of step 2 is repeated). When the value of the parameter ρ is increased, the number of categories increases and the data is classified finely. Conversely, if ρ is reduced, the number of categories decreases, and data is classified roughly. In ART, the parameter ρ is called a warning parameter. When ART is used for the clustering technique, the resolution parameter adjustment unit 530 adjusts the warning parameter ρ. Hereinafter, the alert parameter is referred to as a resolution parameter.

  Step 4: When all the existing categories are reset in Step 2, it is determined as a new category, and a new weighting factor representing a prototype of the new category is generated.

Step 5: When the input data is classified into the category J, the weighting factor WJ (new) corresponding to the category J is the next from the past weighting factor WJ (old) and the input data p (or data derived from the input data). Updated with expression.
[Equation 3]
WJ (new) = Kw · p + (1−Kw) · WJ (old) (Formula 3)

  Here, Kw is a learning rate parameter and determines the degree to which the input vector is reflected in the new weighting factor.

  The feature of the data classification algorithm of the ART module 620 is the processing in step 4. When input data different from the stored pattern is input by the process of step 4, the stored pattern is not changed and a new pattern can be stored. For this reason, it is possible to store a new pattern while storing patterns learned in the past.

  As described above, when the operation data given in advance is given as input data, the ART module 620 learns the given pattern. This process corresponds to the process of the learning unit 510 in FIG. Therefore, when new input data is input to the learned ART module 620, it is possible to determine which pattern is near by the above algorithm. If the pattern has never been experienced before, it is classified into a new category. This process corresponds to the process of the test unit 520 in FIG.

  FIG. 2B illustrates an example of the classification result. In FIG. 2B, two items of measurement data are displayed and represented by a two-dimensional graph. The data is divided into a plurality of categories 630 (circles in FIG. 2B).

  FIG. 3 is a flowchart showing a procedure from when the measurement signal of the plant 100 is acquired until abnormality information is displayed on the image display device 950. As shown in FIG. 3, this flowchart is executed by combining Steps 1000, 1010, 1020, and 1030.

  First, in step 1000, the measurement signal 1 from the plant 100 is taken into the abnormality diagnosis apparatus 200 via the external input interface 210, and the measurement signal 3 is stored in the measurement signal database 310.

  In step 1010, the diagnostic means 600 is operated. In step 1020, if an abnormality is diagnosed in step 1010, the process proceeds to step 1030. If no abnormality is diagnosed, the process returns to step 1000.

  In step 1030, the diagnosis result 10 output from the diagnosis unit 600 is converted into image display data 11 by the external output interface 220 and output to the image display device 950. This notifies the plant operator of the abnormality.

  FIG. 4 is a diagram illustrating an example of the plant 100 and an operation example when the abnormality diagnosis apparatus 200 is applied to the plant 100.

  FIG. 4A is a diagram illustrating a gas turbine plant that is an example of the plant 100.

  The plant 100 includes a gas turbine generator 110, a control device 120, and a data transmission device 130. The gas turbine generator 110 includes a generator 111, a compressor 112, a combustor 113, and a turbine 114. Compressed air from the compressor 112 is sent to the combustor 113, mixed with fuel, and burned in the combustor 113. The turbine 114 is driven by the high-pressure gas generated by the combustion, and the generator 111 generates power.

  The control device 120 generates an operation signal 101 for controlling the output of the gas turbine generator 110 according to the power demand. The control device 120 receives a measurement signal 102 measured by a sensor (not shown) installed in the gas turbine generator 110. The operation data is measured at one-second intervals with state quantities such as intake air temperature, amount of fuel input, turbine exhaust gas temperature, turbine speed, generator power generation, turbine shaft vibration, and the like. It also measures weather information such as atmospheric temperature. The data transmission device 130 transmits the measurement signal 103 measured by the control device 120 to the abnormality diagnosis device 200. Note that the measurement signal 103 can also include the operation signal 101.

  FIG. 4B is a diagram for explaining a result when the abnormality diagnosis apparatus 200 is applied to the plant 100. FIG. 4B shows the measurement signal input to the diagnostic unit 600, the result of classifying the measurement signal 4 into categories by the diagnostic unit 600, and the change over time of the occurrence rate of the new category. Note that the occurrence ratio of the new category is calculated using a smoothing technique such as a moving average using the number of data classified into the new category.

  As shown in FIG. 4B, the diagnosis unit 600 receives measurement signals of data I and data II, and performs an abnormality diagnosis using these signals. As a result of inputting the reference data composed of the data in the normal state into the data classification means 500 in advance and classifying the data into categories, the data was classified into categories 1 to 4 (reference category). Thereafter, the current measurement data is sequentially classified into categories by the diagnosis unit 600. In the time zone before the occurrence of an abnormality, the measurement signal 4 belongs to any one of category numbers 1 to 4, and after the occurrence of the abnormality, it belongs to a category that is not normal (new category). The occurrence rate of new categories calculated by the moving average of the number of new categories that occur in a certain period increases after the occurrence of an abnormality, and an abnormality is diagnosed when it exceeds a preset threshold value. In this example, since it is classified into a new category at the same time as the occurrence of an abnormality, it can be diagnosed with high accuracy.

  However, the desired diagnostic performance may not be obtained depending on the set value of the resolution parameter. FIG. 5 is a diagram illustrating an example of failure in abnormality diagnosis. 4B is different from the setting value of the resolution parameter, FIG. 5B is a diagnosis result when the resolution is rough (when ρ is small), and FIG. 5C is when the resolution is fine (when ρ is large). It is a figure explaining the diagnosis result of ().

  As shown in FIG. 5B, if the resolution is rough, even after the occurrence of an abnormality, the current measurement signal input to the diagnostic unit 600 belongs to the category generated at the reference time. Cannot be detected. On the other hand, as shown in FIG. 5C, if the resolution is fine, a new category is generated even if a minute data tendency is changed in a normal state. Therefore, there arises a problem that a normal state is misdiagnosed as abnormal. From the above, in order to detect an abnormality with high accuracy, it is necessary to set a resolution parameter so that a new category does not occur when normal and a new category occurs after an abnormality occurs.

  FIG. 6 is a diagram for explaining the operation mode and aging degradation of the gas turbine plant. In the gas turbine plant, an output changing operation is performed as shown in FIG. The data trends differ between the output change operation (ε1 to ε5) and the constant output operation (α, β, γ). Further, the data tendency varies depending on the output band. Also, the type of fuel used may change. Furthermore, as shown in FIG. 6 (b), with the aging of the plant, the efficiency decreases and the data trend also changes.

  Since such changes in operating conditions and aging are not abnormal, it is desirable to adjust the resolution parameter so that a new category does not occur even if the data trend changes due to this change.

  These problems can be solved by adjusting the resolution parameter using the reference signal determination unit 400 and the data classification unit 500 of the abnormality diagnosis apparatus 200 of the present invention. In the present invention, the resolution parameter is adjusted to be the finest within a range where a new category does not occur in a normal state. Details of the resolution parameter adjustment method will be described below.

  FIG. 7 is a diagram for explaining an operation example of the reference signal determination unit 400.

  FIG. 7A shows a screen displayed on the image display device 950. Using the external input device 900 while referring to this screen, a parameter used for determining the reference signal 6 by the reference signal determining means 400 is input.

  The reference signal determination unit 400 extracts normal data from the measurement signal database 310 and divides this data into two data groups. The normal period and the data division method are determined based on the information input in FIG.

  The normal period is determined by entering a date. In FIG. 7A, year, month, and day are entered, but information such as hours, minutes, and seconds can also be entered. The division method is selected from three types of “division every fixed period”, “division in the first half and second half”, and “division at random”.

  When “divide every predetermined period” is selected, the data is divided every predetermined period as shown in FIG. 7B (six divisions in the example of FIG. 7B), and these are alternately divided into groups A and B. Divide into groups of data groups. The period of data to be divided is input on the screen of FIG. 7A (in FIG. 7A, one month is input).

  When “divide into first half and second half” is selected, the data is divided into the first half and the second half as shown in FIG.

  When “Randomly Divide” is selected, a random number generated in the range of 0-1 is associated with a time, and the time data corresponding to a random number greater than 0.5 is group A, 0.5 or less. The time data corresponding to the random number is divided into the B group.

  In any of the division methods, the data may be exchanged so that all the operation mode data shown in FIG. 6A are always included in the A group and the B group. Thereby, both the data groups of A group and B group will have the data tendency of all the operation modes.

  FIG. 7C is a diagram for explaining a mode of data stored in the reference signal database 320. As shown in FIG. 7C, the measurement time and the measured data value are divided and stored for each of the data groups of the A group and the B group.

  In FIG. 7, the case where data is divided into two groups, group A and group B, has been described, but it may be divided into three or more data groups. Further, the reference signal determination unit 400 may perform a process such as a smoothing process and a Fourier transform on the measurement signal 5 and create the reference signal 6 using the result after the process. In addition, the reference signal determining unit 400 can have a function of dividing the data into data groups every other sample point. Further, it may be divided so that the measurement signals of the operation state such as the power generation output included in the normal data and the type of fuel to be used are included in the entire data group.

  FIG. 8 is a diagram for explaining the operation of the data classification unit 500, and is a flowchart for determining a resolution parameter. As shown in FIG. 8, this flowchart is executed by combining steps 1100, 1110, 1120, 1130, 1140, 1150, 1160.

  First, in step 1100, the reference signal 7 is divided into learning data and test data. For example, the data group A in FIG. 7C is used as learning data, and the data group B is used as test data. Next, in step 1110, the learning unit 510 is operated to classify the learning data into categories. In step 1120, the test unit 520 is operated to classify the test data into categories.

  In Step 1130, if the number of repetitions of Step 1110 and Step 1120 is less than or equal to a preset threshold value, the process proceeds to Step 1140. If the threshold value is exceeded, the process proceeds to Step 1150. In step 1140, the parameter is updated according to [Equation 4] when a new category is generated when step 1120 is operated, and according to [Equation 5] when a new category is not generated.

Ρ _new is an updated value of the resolution parameter, ρ _old is a previous value of the resolution parameter, ρ _min is a lower limit value of the resolution parameter search range, and ρ _max is an upper limit value of the resolution parameter search range.
[Equation 4]
ρ _new = ρ _old −0.5 × (ρ _old −ρ _min )
ρ _max = ρ _old (Formula 4)
[Equation 5]
ρ _new = ρ _old + 0.5 × (ρ _max −ρ _old )
ρ _min = ρ _old (Formula 5)

Thereafter, in steps 1110 and 1120, the reference data is classified into categories using ρ _new .

In step 1150, the combination of the data group and the data for learning and test is evaluated. If there is no combination that has not been implemented, the process proceeds to step 1160. Assign a reference signal to For example, when the data group A in FIG. 7B is the learning data and the data group B is the test data last time, the data group B is the learning data and the data group A is the test data. In step 1160, ρ _min is determined as a resolution parameter.

  In step 1140, the evaluation value is set to be higher when a new category does not occur in the test unit, and the evaluation value is set to be higher as the resolution is smaller, and the evaluation value is set using an optimization method such as a genetic algorithm. You may make it search for the resolution parameter in which becomes the maximum.

  FIG. 9 is a diagram for explaining the resolution parameter adjustment process.

The vertical axis in FIG. 9A represents ρ _max , ρ _min , ρ _new , and whether or not a new category has been generated in step 1120, and the horizontal axis represents the number of repetitions of steps 1110, 1120, 1130, and 1140 in FIG. 8. . FIG. 9A shows the result when the number of repetitions is set to five.

When the number of repetitions was zero, a new category was generated in step 1120 of FIG. In this case, ρ _new and ρ _max are updated according to [Equation 4]. In the first iteration, as a result of operating the learning unit and the test unit using the updated ρ _new , no new category was generated in step 1120 of FIG. In this case, ρ _new and ρ _min are updated according to [Equation 5]. Thereafter, the same procedure is repeated to update the resolution parameter.

FIG. 9B is a diagram for explaining a relationship between a combination of learning data and test data and a value of a candidate value (ρ _min ) of ρ. The candidate value of ρ when the learning data is set to the A group and the test data is set to the B group is 0.999, and the candidate of ρ when the learning data is set to the B group and the test data is set to the A group. The value is 0.99. In this case, in step 1160 of FIG. 8, a resolution parameter value (0.999 in the example of FIG. 9B) with the smallest resolution among these candidate values is used as the final resolution parameter value. decide. Thus, by changing the combination of the learning data and the test data and determining the resolution parameter so that the resolution becomes the finest among them, it is possible to deal with outliers of the data.

  FIG. 9C is a diagram for explaining outliers. If the outlier belongs to the A group, and the A group becomes the test data, a new category is generated when the outlier is input unless the resolution is set coarsely. Therefore, the resolution parameter adjustment unit 530 may adjust the resolution parameter so that the resolution becomes coarse, and may not be able to detect an abnormality.

  In order to avoid this phenomenon, in the embodiment of the present invention, the resolution data is adjusted by exchanging the data for learning and the data for testing, and the value with the finest resolution is selected.

  The result of diagnosis using the resolution parameter value adjusted in the flowchart of FIG. 8 is the same as that described with reference to FIG.

FIG. 10 is a resolution parameter confirmation screen, which is an example of a screen displayed on the image display device 950. The value of ρ _new in FIG. 9 in the image display device 950 and the rate at which a new category has occurred in the test unit at that time. By displaying these two relationships, the plant operator can validate the selected value of ρ. In addition to the rate of occurrence of new categories, the relationship between the amount of occurrence, the presence / absence of occurrence, and the value of ρ may be used.

  Further, it is possible to display the screen of FIG. 10 on the image display device 950 and manually change the resolution parameter value. In other words, the resolution parameter 1161 is selected in the resolution parameter determination (step 1160) in the flowchart of FIG. 8, but if it is desired to make a diagnosis with the coarse resolution parameter 1162 or the fine resolution 1163, an external display on the screen of FIG. It is also possible to operate the input device 900 and change the value of the resolution parameter. In this way, the abnormality diagnosis device stores the measurement signal of the plant in the measurement signal database, performs the learning to classify into the category using the measurement signal in the normal state, and uses the other measurement signal that is not used in the learning to perform the category. The reference signal is classified into categories according to the similarity of the data by executing the test classified into, and adjusting the resolution parameter used when classifying into the category using the learning and test classification results. When adjusting the resolution parameter, the relationship between the resolution parameter value set in the past by the resolution parameter adjustment and the new category generated by the test is displayed on the screen, and the measurement signal for the diagnosis is different from the normal measurement signal. If the ratio that does not belong to the category that occurred in the classification exceeds a certain value, it is decomposed by an abnormality diagnosis method that diagnoses an abnormality. Display the relationship between the parameter value and the percentage of new categories that occurred in the test, allowing plant operators to validate the selected ρ value and perform anomaly diagnosis with appropriately tuned parameters Can do.

  In the resolution parameter adjustment unit 530 of this embodiment, in addition to the function of adjusting the resolution parameter so that the resolution becomes the finest within the range where no new category occurs in the test unit, the categories to which two consecutive data groups belong are the same. It also has a function of adjusting the resolution parameter so that In this case, in step 1140 of FIG. 8, when step 1120 is operated, a new category does not occur, and when the category classified into it is the same as step 1110, the parameter is updated in [Equation 5], and the other categories are updated. At time, the parameter is updated by [Equation 4]. Thereby, it is possible to cope with the aged deterioration of the plant described in FIG.

  FIG. 11 is a diagram for explaining a method for dealing with aging deterioration, in which data values in January and February are plotted. It is assumed that both January and February are normal, but the data trend has changed slightly due to aging.

  FIG. 11A shows data for January, category 1201 includes data for January, and category 1202 includes data for February. Some of the February data is outside the category 1201.

  Therefore, if a category is created using January data as a reference signal and the February data is diagnosed, a new category is generated because a part of the February data does not belong to the category 1201. Thereby, although it is normal, it may be diagnosed as abnormal. When the resolution is adjusted so that the categories to which two consecutive data groups belong are the same, the categories 1203 and 1204 in FIG. 11B are generated. That is, even if the data trend changes due to aging for one month, a new category does not occur.

  Therefore, the resolution parameters are determined using the operation data (reference data) for January and February as learning data and test data, respectively. After that, the operation data for March is diagnosed using this resolution parameter. In this case, a new category does not occur even if the data trend changes due to aging for one month. If a sudden abnormality occurs and the data changes greatly, a new category is generated.

  Therefore, since a new category is generated only when a sudden abnormality occurs, the abnormality can be detected with high accuracy.

  FIG. 12 is a block diagram showing a second embodiment of the abnormality diagnosis apparatus of the present invention. The reference signal determining unit 400, the data classifying unit 500, and the diagnostic unit 600 in the first embodiment are different in the following points.

  The reference signal determining unit 400 divides normal data of the measurement signal 5 stored in the measurement signal database 310 into at least three groups.

  In the data classification unit 500, a threshold setting unit 540 is added to the learning unit 510, the test unit 520, and the resolution parameter adjustment unit 530. The learning unit 510 classifies the data into categories using one of the data groups. The test unit 520 classifies the data into categories using one of the remaining data groups that are not used by the learning unit 510. In addition, the threshold setting unit 540 uses one of the remaining data groups not used by the learning unit 510 and the test unit 520 as a threshold that does not belong to the category generated by the learning unit and the test unit. This threshold value is stored in the classification result database 330. For the quantity that does not belong to the category for the threshold, the ratio of the quantity that does not belong to the category divided by the quantity that belongs to the category in order to grasp the quantity that does not belong to the category, or whether there is a quantity that does not belong to the category that is 1 as the quantity Including.

  The diagnosis unit 600 diagnoses an abnormality when the ratio that the current measurement signal does not belong to the category generated by the data classification unit 500 exceeds the threshold obtained by the threshold setting unit 540.

  That is, in the first embodiment, the operator has to set a threshold value (see FIG. 4B) for diagnosing an abnormality with the diagnosis unit 600 in advance, but in the second embodiment, this threshold value is set. It is automatically determined by the threshold setting unit 540. Hereinafter, a method for determining the threshold will be described.

  FIG. 13 is a diagram for explaining an example of the operation result of the reference signal determining means 400 in the second embodiment of the abnormality diagnosis apparatus of the present invention. The normal data of the measurement signal 5 stored in the measurement signal database 310 is divided into at least three groups. FIG. 13 shows an example in which normal data is divided into three groups of A group, B group, and C group.

  FIG. 14 is a flowchart for explaining the operation of the data classification means 500 in the second embodiment of the diagnostic apparatus of the present invention.

  The flowchart of FIG. 14 is configured by adding step 1170 to the flowchart of FIG. In step 1170, the threshold value setting unit 540 is operated, and the occurrence ratio of new categories not belonging to the categories of the learning unit and the test unit when the resolution parameter determined in step 1160 is used is obtained, and this value is determined as the threshold value.

  New categories can occur due to outliers, measurement noise, and other effects. Since the threshold setting unit 540 performs classification using only normal data, a new category that occurs when the threshold setting unit 540 is operated occurs due to the influence of outliers, measurement noise, and the like. Therefore, in this embodiment, the diagnosis accuracy is improved by diagnosing the normal category when the ratio of the new category generated when the diagnostic unit 600 is operated is equal to or less than the threshold value, and diagnosing the abnormality when the ratio is equal to or greater than the threshold value. I am doing so.

  FIG. 15 is a diagram for explaining an aspect of data stored in the classification result database 330 in the second embodiment of the abnormality diagnosis apparatus of the present invention.

  As shown in FIG. 15A, the classification result database 330 stores the relationship between the resolution parameter and the threshold value. In addition, as shown in FIG. 15B, a threshold value can be stored for each category, and the threshold value is calculated for each category number before the new category is generated, and the value is stored. Also good.

  Thus, by setting a threshold value for each category, the threshold value can be changed for each operating condition, and a more accurate diagnosis is possible.

  FIG. 16 is a diagram illustrating an example of a screen displayed on the image display device 950 in the second embodiment of the abnormality diagnosis apparatus of the present invention.

  By displaying the new category occurrence rate and the change over time in the threshold, the transition of these parameters can be easily confirmed.

  The devices of the above-described embodiments can be implemented by a computer having a memory and a CPU, and the processing means as functions of the device are program modules. By reading the modules and causing the computer to execute them, Each function can be implemented.

  Each function can be implemented by causing a computer to read a recording medium on which a program module is recorded.

  In addition, this invention is not limited to an above-described Example, Various modifications are included. For example, the above-described embodiments have been described in detail for easy understanding of the present invention, and are not necessarily limited to those having all the configurations described. In addition, a part of the configuration of a certain embodiment can be replaced with the configuration of another embodiment, and the configuration of another embodiment can be added to the configuration of a certain embodiment. Further, it is possible to add, delete, and replace other configurations for a part of the configuration of each embodiment.

  In addition, each of the above-described configurations may be configured such that some or all of them are configured by hardware, or are implemented by executing a program by a processor. Further, the control lines and information lines indicate what is considered necessary for the explanation, and not all the control lines and information lines on the product are necessarily shown. Actually, it may be considered that almost all the components are connected to each other.

100 Plant 200 Abnormality Diagnosis Device 210 External Input Interface 220 External Output Interface 310 Measurement Signal Database 320 Reference Signal Database 330 Classification Result Database 400 Reference Signal Determination Unit 500 Data Classification Unit 510 Learning Unit 520 Test Unit 530 Resolution Parameter Adjustment Unit 600 Diagnosis Unit 900 External input device 910 Keyboard 920 Mouse 950 Image display device

Claims (21)

A measurement signal database that stores plant measurement signals;
A reference signal determining means for dividing the measurement signal at the time of normal plant into two or more data groups and creating a reference signal;
A learning unit that classifies into a category using one of the data groups, a test unit that classifies into a category using one of the remaining data groups not used in the learning unit, the learning unit, and the test unit A resolution parameter adjustment unit that adjusts a resolution parameter used when classifying into a category using the classification result, and a classification unit that classifies the reference signal into a category according to data similarity;
An abnormality diagnosing device comprising: a diagnosis unit that diagnoses an abnormality when a quantity that does not belong to a category generated by the classification unit exceeds a certain value with respect to a measurement signal to be diagnosed that is different from the measurement signal in the normal state. The abnormality diagnosis device according to claim 1,
The reference signal determining means divides a measurement signal when the plant is normal into three or more data groups,
Using one of the remaining data groups that are not used in the learning unit and the test unit, a threshold setting unit that obtains as a threshold a ratio that does not belong to the category generated by the learning unit and the test unit,
The diagnostic means diagnoses an abnormality when a quantity that does not belong to the category generated by the classification means exceeds the threshold for a measurement signal to be diagnosed different from the measurement signal at the normal time Diagnostic device. In the abnormality diagnosis apparatus according to claim 1 or 2,
The resolution parameter adjustment unit adjusts the resolution parameter so that the resolution becomes the finest within a range in which a category different from the category classified by the learning unit does not occur when the test unit classifies the category. A characteristic abnormality diagnosis device. In the abnormality diagnosis apparatus according to claim 3,
The abnormality diagnosis apparatus characterized in that the resolution parameter adjustment unit determines a resolution parameter with the finest resolution while changing a combination of data groups distributed to the learning unit and the test unit. In the abnormality diagnosis apparatus according to claim 3,
In the reference signal setting means, the data is divided into two or more groups at predetermined intervals,
An abnormality diagnosis apparatus, wherein the resolution parameter adjustment unit adjusts a resolution parameter using information on a quantity in which two consecutive data groups are classified into the same category. Applying the abnormality diagnosis device according to claim 1 or claim 2 to a thermal power plant,
The reference signal determining means uses the measurement signal of the operating state including the power generation output or the type of fuel to be used as the measurement data in the normal state, and divides the data group so as to be included in all data groups An abnormality diagnosis device for a thermal power plant. In the abnormality diagnosis apparatus according to claim 1 or 2,
An abnormality diagnosis apparatus characterized by using a clustering technique as a technique for classifying into categories. In the abnormality diagnosis apparatus according to claim 2,
An abnormality diagnosis apparatus, wherein the threshold value setting unit obtains a threshold value for each category that belonged immediately before a new category occurred. In the abnormality diagnosis apparatus according to claim 1 or 2,
The reference signal determining means has at least one of a function of dividing into data groups using random numbers, a function of dividing into data groups every other sample point, and a function of dividing into data groups in a predetermined period. An abnormality diagnosis device characterized by the above. In the abnormality diagnosis apparatus according to claim 1 or 2,
An abnormality diagnosing device comprising: an image display device for displaying a relationship between a resolution parameter value evaluated in an adjustment process in the resolution parameter adjustment unit and a new category generated in the test unit on a screen. In the abnormality diagnosis apparatus according to claim 8,
An abnormality diagnosis device comprising an image display device for displaying a change with time of the threshold on a screen. The abnormality diagnosis device
Store the measurement signal of the plant in the measurement signal database,
Divide the measurement signal when the plant is normal into two or more data groups to create a reference signal,
Performing learning to classify into a category using one of the data groups, performing a test to classify into a category using one of the remaining data groups not used in the learning, and By classifying the reference signal into categories according to the similarity of data by adjusting the resolution parameter used when classifying into categories using the classification results,
An abnormality diagnosis method comprising: diagnosing an abnormality when a measurement signal to be diagnosed different from the normal measurement signal exceeds a certain value that does not belong to the category generated in the classification. In the abnormality diagnosis method according to claim 12,
When creating the reference signal, the normal measurement signal of the plant is divided into three or more data groups,
Using one of the remaining data groups not used in the learning and the test, a quantity that does not belong to the category generated in the learning and testing is obtained as a threshold value,
In the case of the diagnosis, the measurement signal to be diagnosed that is different from the measurement signal at the normal time is diagnosed as abnormal when the quantity that does not belong to the category generated in the classification exceeds the threshold value Diagnosis method. In the abnormality diagnosis method according to claim 12 or claim 13,
An abnormality diagnosis method characterized by adjusting the resolution parameter so that the resolution becomes the finest within a range in which a category different from the learning result category is not included in the test result category when adjusting the resolution parameter . In the abnormality diagnosis method according to claim 14,
A method for diagnosing abnormality, wherein the resolution parameter is determined so that the resolution becomes the finest among the combinations of data groups used to obtain the learning result and the test result. In the abnormality diagnosis method according to claim 14,
When creating the reference signal, the data is divided into two groups at predetermined intervals,
A method for diagnosing an abnormality, wherein when adjusting a resolution parameter, the resolution parameter is adjusted using information on the quantity of two consecutive data groups classified into the same category. Applying the abnormality diagnosis method according to claim 12 or claim 13 to a thermal power plant,
When creating the reference signal, the measurement signal of the operation state including the power generation output or the type of fuel to be used is used as the normal measurement data and is divided so as to be included in the entire data group of the data group. An abnormality diagnosis method. In the abnormality diagnosis method according to claim 12 or claim 13,
An abnormality diagnosis method characterized by using clustering technology as a method of classifying into categories. In the abnormality diagnosis method according to claim 13,
A method for diagnosing abnormalities, characterized in that, when the threshold value is obtained, a threshold value is obtained for each category that belonged immediately before a new category occurred. In the abnormality diagnosis method according to claim 12 or claim 13,
When generating a reference signal, at least one of dividing into data groups using random numbers, dividing into data groups every other sample point, and dividing into data groups at a predetermined period is executed. An abnormality diagnosis method characterized by the above. The abnormality diagnosis device
Store the measurement signal of the plant in the measurement signal database,
Perform learning to classify into categories using measurement signals at normal time, perform tests to classify into categories using other measurement signals not used in the learning, and use the learning and the classification results of the tests By adjusting the resolution parameter used when classifying into categories, the reference signals are classified into categories according to the similarity of data,
When adjusting the resolution parameter, display the relationship between the resolution parameter value set in the past by the resolution parameter adjustment and the new category generated by the test on the screen.
An abnormality diagnosis method comprising: diagnosing an abnormality when a measurement signal to be diagnosed different from the normal measurement signal exceeds a certain value that does not belong to the category generated in the classification.

Images