JPH0728770A - Knowledge generating device - Google Patents

Abstract

PURPOSE:To objectively and quantitatively generate the knowledge for operation support without depending upon operator's subjectivity at the time when the operator generates this knowledge by interview by classifying recorded data of a recording means into clusters and calculating the statistical values of each cluster. CONSTITUTION:First, sample data is gathered from each observation object like a plant and is displayed on a CRT or the like. The operator sees the display of this sample data to determine the cluster of the observation object. Next, the sample time and multiple variable data in sample data are recorded by a recording means 3, and the discrimination result including the cluster determined by the operator is recorded. Further, recorded data of the recording means 3 is classified into clusters by a concept skeleton calculating means 4 to calculate the statistical values of each cluster, and the calculation result is subjected to the operation including the calculation of the Mahalanobis parallel distance to determine the boundary face of the cluster. The support knowledge for the operator to discriminate the condition of the observation object is generated from sample data by this boundary face.

Description

Detailed Description of the Invention

[0001]

BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention provides an operator who operates a plant or the like with an instruction based on knowledge for supporting failure diagnosis, abnormality monitoring, and operation, etc., based on the empirical rule of a skilled operator who operates the plant or the like. The present invention relates to a knowledge generation device for

[0002]

2. Description of the Related Art As a conventional knowledge generation device, for example, "Joint work module for situation recognition" (written by Hiromitsu Shimakawa and Teruo Usami, Proceedings of the 6th Annual Meeting of the Software Science Society, 19
There is a knowledge generation device for driving support described in 1989). This device is designed to monitor anomalies during operation in large-scale plants such as electric power and steel. In a large-scale plant, multivariate data such as temperature, pressure, current, and voltage are collected from a plant monitoring device, and the size of these data changes with time. Therefore, a skilled operator always recognizes the state transition on the time axis of the observation target when performing fault diagnosis, abnormality monitoring, operation, etc. of a large-scale plant.

FIG. 21 is an overall configuration diagram of a conventional operation support knowledge generation apparatus in a plant. In FIG. 21,
Reference numeral 1a is an input / output device for displaying the operating state of the plant and inputting data, 1b is a central processing unit (CPU) for controlling the entire plant, 2 is a monitoring target of the input / output device 1a, and judgment is made, and an observation target in the plant An operator for deciding and describing a cluster indicating the situation of 6 is a monitoring device for a plurality of plants for monitoring the state of the observation target in the plant. The state of each observation target sent from the monitoring device 6 of these plants is displayed as a plant operating state display on the CRT screen of the input / output device 1a. FIG. 22 is a diagram showing a structure of data displayed as an operating state display of the plant. In FIG. 22, 100 is a characteristic variable, which is a variable expressing the state quantity sampled from the observation target. Assuming that the characteristic variables 100 are temperature variables T1 and T2 that change with time, for example, the temperature variable T
1 and T2 are shown. 101 is an instance,
It is a collection of feature variables 100 sampled from one observation target, and represents the state of the observation target. A scene 102 is composed of a set of sample time and instance 101. 103 is a time series, and scene 102
Are arranged in order of time within a certain period, and are data representing state transitions during a certain period of the observation target.

A pattern for recognizing the above characteristic variables is a skeleton. This skeleton is a set of instances 101 that satisfy the condition consisting of a plurality of conceptual skeletons and time-series skeletons, and is recorded in the database for recognizing the situation of the observation target. Concept skeletons describe a wide variety of multivariate data (eg, temperature, voltage, current, etc.) from many plant monitoring devices as "high temperature, normal voltage" or "normal voltage "The voltage is abnormal" is described, and the state of one sample data taken from an observation target such as a plant is determined. When the boundary surface of the cluster showing the situation of the observation target is decided, the condition of the parameter indicating the establishment of the conceptual skeleton is quantitatively decided by this decision. However, this conceptual skeleton cannot judge temporal changes in sample data. Therefore, a time-series skeleton is prepared for this purpose. The time-series skeleton is a condition indicating a change in state of the plant with the passage of time, and is equivalent to a so-called conceptual skeleton arranged in the time axis direction.

FIG. 23 is a skeleton showing changes in the inflow velocity and temperature of the liquid in a device for putting the liquid of high temperature into the bath. Note that, in the example of FIG. 23, the amount of change in the previous sample data between the actual value of each manipulated variable / controlled amount of the liquid and the set value of the inflow velocity or temperature of the liquid is recorded. To do. In FIG. 23, 1
Reference numeral 03 denotes a concept, and the concept tank indicates a characteristic variable for recording the inflow velocity of the liquid, a change in temperature, and the like. Reference numerals 104 and 105 are conceptual skeletons and time-series skeletons, which show a tank representing a bathtub to be observed and a transition from a manipulation limit value to a state in which an inflow velocity, which is a manipulated variable of the liquid, flows. The outline of each skeleton description is as follows. In the concept 103, first, the type of the characteristic variable in the characteristic variable tank is described. For example, it is int (integer type) or float (real number type). Next, the manipulated variable of the liquid, the inflow rate,
Describe the variable name at temperature. For example, c_fb, v
x_fb, tmp_chg, etc. Furthermore, the range of possible values of those variables is described. For example, c_fb:>
= 10, <= 100, etc., and in this case the variable c_fb
Indicates a value from 10 to 100. Next, in the conceptual skeleton 104, clusters and inequalities corresponding to the characteristic variable tank and their logical products, logical sums, and assessments are represented. For example, normal, repairable, tm
pIncremented is a cluster that is an adjective of a conceptual skeleton, and c_fb:> = 30, <= 40;
Is an inequality. That is, when the sample data of the variable c_fb satisfies the conditions of> = 30 and <= 40, the cluster is determined to be normal. Further, the time-series skeleton 105 shows a state in which the time-series skeleton toBeRepaired of the characteristic variable tank transits over time. For example, the characteristic variable tank for 45 seconds is a state of conceptual skeleton tmpIncremented at 0 second (at (0)), a state of conceptual skeleton normal from 0 to 30 seconds, and a state of conceptual skeleton normal from 30 seconds to 45 seconds. Concept skeleton Re
That is, it is in a pairable state.

Next, the operation of the conventional driving support knowledge generating apparatus will be described with reference to FIGS.
First, the experienced operator 2 who has a long experience in plant operation first matches the normal (normal) and replaceable (regular) that match each scene while monitoring the input / output device 1a for each time-series scene sampled from the plant monitoring apparatus 6. (Repair) etc. are determined and described. Also, time-series skeleton 1
The same applies to 05. This description is, for example, a dialogue (interview) between the input / output device 1a and the operator 2.
While done. This reveals the conceptual skeleton that matches each scene. As a result, in the conceptual skeleton 104, if the magnitude of the variable c_fb of the manipulated variable in the sample data input from the monitoring device 6 of the plant is 35, the possible value of the variable is> = 3.
Since 0, <= 40, the cluster is nom
It is judged as al (normal), and it is notified to the operator 2 who has little experience in actually operating the plant. The operator 2 looks at the result and performs a predetermined operation on the plant. Also in the time-series skeleton 105, similarly to the above, for example, when the characteristic variable tank to be observed is from 0 second to 30 seconds, the concept skeleton 104>
When the conditions of = 30 and <= 40 are satisfied, the cluster is judged to be normal (normal) and the operator 2 is notified.
Next, the operator 2 receives the transmitted cluster normal.
(Normally) That is, it is checked whether or not the concept skeleton 104 matches the concept skeleton specified in the time series skeleton 105. Then, the operator 2 performs this operation for all the conceptual skeletons 104 in the time series skeleton 105, and if all match, it is determined that the time series skeleton 105 matches. By the above operation, the operator 2 can determine whether or not the time series skeleton representing the characteristics of the state transition in the observation target of the plant matches the time series 103 in FIG.

[0007]

In the conventional driving support knowledge generation apparatus, a skilled operator interacts with a personal computer or the like to describe a concept skeleton (cluster boundary) which is a component of the driving support knowledge (interview). ) However, it was done by one's subjectivity. Therefore, since the operator's own subjectivity is added, the knowledge description based on the objective criteria is not made, and there is a problem that it is impossible to quantitatively determine the performance of the driving support knowledge. Further, when using these driving support knowledge to educate an operator with a low degree of skill, there is a problem that it is difficult to understand because the knowledge cannot be visually confirmed.

The present invention has been made in order to solve such a problem, and when the operator creates the knowledge for driving support by an interview, it is objective and quantitative without depending on the subjectivity of the operator. The purpose is to obtain a knowledge generation device that can be created. Another object of the present invention is to obtain a knowledge generation device capable of judging the performance of the driving support knowledge created by the operator and displaying the driving support knowledge to easily educate a less skilled operator.

[0009]

As shown in FIG. 1, a knowledge generation apparatus according to the first aspect of the present invention records sample time and multivariate data of sample data in a plant, and a cluster determination result by an operator. Recording means 3 for recording, and recording data of the recording means 3 are classified into clusters to calculate a statistic amount for each cluster,
The calculation result includes a knowledge calculation unit (conceptual skeleton calculation unit 4) that determines the boundary surface of the cluster by performing calculation including calculation of Mahalanobis square distance. As shown in FIG. 15, the knowledge generating device according to the second aspect of the present invention is based on the determination result of the boundary surface by the knowledge calculating means (conceptual skeleton calculating means 4) and the determination result of the recording data by the recording means 3. The discriminating performance determining means 7 for quantitatively determining the discriminating performance based on the support knowledge for the operator is provided in the configuration of the first invention. As shown in FIG. 18, the knowledge generation device according to the third invention is
The support knowledge display means 8 is provided to display the boundary surface determined by the knowledge calculation means (conceptual skeleton calculation means 4) so that the boundary surface can be visually confirmed.

[0010]

The knowledge generating device according to the first aspect of the present invention operates as follows. First, sample data is collected from each observation target such as a plant and displayed on a CRT or the like. The operator determines the cluster to be observed by looking at the display of the sample data. Next, the recording means 3 records the sample time and the multivariate data in the sample data, and records the determination result including the cluster by the operator. Further, the knowledge calculation means (conceptual skeleton calculation means 4) classifies the recording data of the recording means into clusters and calculates the statistic amount for each cluster, and the calculation result includes the calculation of the Mahalanobis square distance. Determine the boundary surface of the cluster. With this boundary surface, knowledge for supporting an operator who judges the situation of the observation target from the sample data is generated. In the knowledge generation device according to the second aspect of the present invention, the discrimination performance judgment means 7 is based on the judgment result of the boundary surface by the knowledge calculation means (conceptual skeleton calculation means 4) and the judgment result of the operator recorded by the recording means 3. , It works by quantitatively obtaining the discrimination performance of the support knowledge by the boundary surface. The knowledge generation device according to the third aspect of the invention operates by the support knowledge display means 8 displaying the boundary surface determined by the knowledge calculation means (conceptual skeleton calculation means 4) on coordinates such as CRT. Then, an operator with a low degree of skill visually confirms the boundary surface.

[0011]

Embodiments of the present invention will now be described with reference to the drawings. FIG. 1 is an overall configuration diagram of an operation support knowledge generation device in a plant showing an embodiment (embodiment 1) of the first invention. In FIG. 1, 1a is a terminal for displaying the operating state of the plant and inputting data, and 1b is a CPU
(Central processing unit) 2 is an operator who monitors the input / output device 1a to judge a cluster of sample data. 3 is a data record including the cluster judged by the operator 2 (sample time of sample data and multivariate data). A recording means 4, a concept skeleton calculating means as a knowledge calculating means for objectively and quantitatively generating the concept skeleton from the record data 3a of the recording means 3, and 5 a concept skeleton which is a source of knowledge for supporting the operation of the plant. Is.

FIG. 2 is a functional block diagram showing in detail the conceptual skeleton calculating means shown in FIG. In FIG. 2, 31A
Is based on the recording data 3a input from the recording means 3,
Data classification means for classifying sample data into clusters, 32A means for calculating statistics from the sample data classified for each cluster, 33A means for calculating square distance of Mahalanobis from the statistics. , 34A are boundary surface determining means for determining the boundary surface of the cluster which is the source of 5 from the calculation result.

FIGS. 3 and 4 are flow charts showing the operation of the apparatus of the first embodiment. Next, this Example 1
The operation of this device will be described with reference to the flowcharts of FIGS. First, the sample data of the observation target of the plant is taken into the CPU 1b from the monitoring device 6 of the plant (step S21) and displayed on the input / output device 1a (step S22). The skilled operator 2 looks at the displayed sample data, determines a cluster of adjectives such as "low temperature" and "normal temperature" (step S23), and describes from the keyboard of the input / output device 1a. To do. This description includes the sampling time (observation time) of the sample data, the multivariate data, and the input / output device 1
It is recorded in the recording means 3 via a and the CPU 1b (step S24). Next, the recording data 3a output from the recording means 3 is input to the conceptual skeleton calculating means 4.
In the conceptual skeleton calculation means 4, the data classification means 31A
Classifies the recorded data 3a for each cluster (step S3
1), recorded data 3 classified by the statistical amount calculating means 32A
The statistical amount of the multivariate data for each cluster is calculated from b (step S32). This statistic is calculated by calculating the centroid vector and the variance-covariance matrix. The formula of this center of gravity vector is

[0014]

[Equation 1]

It is represented by the above equation (1). Also, the arithmetic expression of the variance-covariance matrix is

[0016]

[Equation 2]

It is expressed by the above equation (2). Next, the square distance calculating means 33A obtains the Mahalanobis square distance from the centroid vector for each cluster for any value of any multivariate data (step S33). Here, the Mahalanobis square distance is a function of the value of the multivariate data, and the arithmetic expression for obtaining this Mahalanobis square distance is

[0018]

[Equation 3]

It is expressed by the above equation (3). Next, the boundary surface between each cluster is determined (step S34). When determining this boundary surface, the values of the multivariate data are classified into clusters having the smallest Mahalanobis square distance, and the cluster to which all the values of the multivariate data in the multivariate space belong are determined. Then, the boundary surface of the conceptual skeleton in the multivariate space is determined, and the driving support knowledge is generated.

FIG. 5 is a circuit block diagram showing a concrete structure of a simple plant of a cooling device, for example. In FIG. 5, 50 is a sensor A or B for detecting temperature, 51 is a controller, 52 is a driver, and 53 is a fan. The sensor A50 detects the temperature of the inlet for cooling the heating element, and the sensor B50 detects the temperature of the outlet for cooling the heating element. Reference numeral 51 is a controller that controls the driver 52 based on the results detected by the sensors 50A and 50B and changes the rotation speed of the fan 53 to cool the heating element. FIG. 6 is a diagram showing sample data times, multivariate data (temperature), and operator's judgment (cluster) in each part of the cooling device of FIG. In FIG. 6, 6
0 is the time when the sample data is sampled, 61 is the temperature T1 detected by the sensor 50A in the cooling device of FIG.
(= X1), 62 indicates the change in temperature T2 (= x2) detected by the sensor 50B in the cooling device, and 63 indicates the judgment of a skilled operator.

Next, the operation when the device of the first embodiment is used for the cooling device of FIG. 5 will be described. First, Fig. 1
In, the plant monitoring device 6 monitors the temperature detected by the sensors 50A and 50B in the cooling device of FIG. 5 at regular time intervals. The temperature data for each time is C
It is taken into the PU 1b (step S21 in FIG. 3) and displayed on the input / output device 1a. This display is displayed, for example, as shown in FIG. 6 (step S22 in FIG. 3). The operator 2 judges the cluster while watching the temperature data for each time (step S24 in FIG. 3), describes the judgment result, and inputs it from the input / output device 1a. This description result is the operator's decision 63 in FIG. The data of FIG. 6, that is, the sample time of the sample data (time 6
0), multivariate data (temperatures T1 and T2), and operator judgment results (operator judgment 63) are recorded in the recording means 3 (step S24 in FIG. 3).

Next, in FIG. 2, 31A classifies the recorded data 3a. This classification is performed for each judgment made by the operator (step S31 in FIG. 4).
That is, time 13:40:10 to time 13:14:
The temperatures T1 and T2 up to 40 are “hot” of the cluster, and the times 13:40:50 to 13:41:20 are “normal” of the cluster. As a result, hot {x1
(T1), x2 (T2) is

[0023]

[Equation 4]

The above equation (4) is obtained, and the normal {x1 (T
1) and x2 (T2) are

[0025]

[Equation 5]

The above equation (5) is obtained. Next, (1),
The statistic for each cluster is calculated using the equation (2) (step S32 in FIG. 4). Here, the number of samples S = 4. As a result, according to the equation (1), the center of gravity vector χ (bar) when the cluster is hot is

[0027]

[Equation 6]

The above equation (6) is obtained. Next, using the equation (2), the elements c11, c22, c12, c21 of the covariance matrix C when the cluster is hot are calculated.
The result of this calculation is

[0029]

[Equation 7]

The above equations (7) and (8) are obtained, and the calculation result of the variance-covariance matrix C is

[0031]

[Equation 8]

The above equation (9) is obtained. Similarly, when the statistic when the cluster is normal is calculated using the equations (1) and (2), the center of gravity vector χ (bar) and the covariance matrix C when the cluster is normal are

[0033]

[Equation 9]

The above equation (10) is obtained. Next, the Mahalanobis square distance of each cluster is calculated using equation (3).
As a result, the square of the Mahalanobis square distance D when the cluster is hot is

[0035]

[Equation 10]

The above equation (11) is obtained. Also, the square of the Mahalanobis square distance D when the cluster is normal is

[0037]

[Equation 11]

The above equation (12) is obtained. When these clusters are hot, the magnitude is compared from the calculation result of the Mahalanobis square distance in the normal case, and the boundary surface of the cluster is determined from this calculation result (step S34 in FIG. 4). Therefore, the magnitude comparison of the square of the Mahalanobis square distance D is

[0039]

[Equation 12]

When the above expressions (13) and (14) are satisfied, the square of the Mahalanobis square distance D is

[0041]

[Equation 13]

The above expressions (15) and (16) are satisfied. As a result, the temperature of the cooling device in FIG. 5 is T1 (x1) ≦ T2.
When (x2) is satisfied, it is hot, and the temperature is T1 (x
When 1) <T2 (x2) is satisfied, the condition is normal, and the conceptual skeleton 5 in FIG. 1 is described as in FIGS. In the conceptual skeleton of FIG. 7, temperature T1 (x1) ≦ T2
Under the condition of (x2), it is described as hot, and the temperature T1 (x1)> T in the conceptual skeleton of FIG.
When the condition is 2 (x2), it is described as normal. Further, the boundary surface of the cluster is shown on the coordinate axes as shown in FIG. In FIG. 9, 64 is a hot area, 65 is a normal area, and 66 is a boundary surface (temperature T1).
(X1) = T2 (x2), where x and y axes are x1 and x
The sample data in 2 is shown. With the conceptual skeleton created in this way, sample data at a certain time from the cooling device of FIG. 5, for example, temperature T1 = 40 ° C.,
When the sample data of T2 = 30 ° C. is input to the driving support knowledge generation device, from the condition of the temperature T1 (x1)> T2 (x2) of the boundary surface of the cluster by the conceptual skeleton,
The cluster is determined to be normal, and the operator 2 of FIG. 1 is indicated by the input / output device 1a. The operator 2 performs a predetermined process on the cooling device of FIG. 5 while looking at this cluster.

The case where the operation support knowledge generation device of the first embodiment is used in the plant of the simple cooling device has been described above, but the number of sample data is very large in an actual large-scale plant. Next, the case where the apparatus of Example 1 is used in a general plant will be briefly described. First, Fig. 1
0 is the recording result by the recording means 3 of FIG. 1, and is the processing result of step S24 in the flowchart of FIG. In FIG. 10, 70 is a time variable of the sample time of the sample data, and variables Xt = 0 to Xt = 5 are described. 71 is the time, and 70 variables Xt = 0 to Xt =
It corresponds to each of 5. 72 is multivariate data, and x1 = 14.0 to xn = 134.0 corresponding to the time.
There are n multivariate data of. Reference numeral 73 is a cluster judgment made by the operator, and is COOL (cold), NOMA
L (normal) and HOT (hot). Next, the recording data is classified into each cluster by the process of step S31 of FIG. As a result, as shown in FIG. 11, the cluster is HO.
T (hot) is Xt = 3, Xt = 4, cluster is N
OMAL (normal) ones are classified into Xt = 2 and Xt = 5, and clusters COOL (cold) are classified into Xt = 0 and Xt = 1.

Next, by the processing of step S32 of FIG. 4, the statistic amount for each classified cluster is calculated according to the flowchart of FIG. First, the number of sample data K = 0 and the center of gravity vector χ (bar) = 0 are initialized (step S41), and the sum of K = K + 1 is calculated (step S42). Next, the calculation based on the above-mentioned formula (1),

[0045]

[Equation 14]

That is, the calculation of the equation (17) is performed (step S43). Then, it is determined whether K = S,
If YES, the process ends, and if NO, the process returns to step S42. Next, the variance / covariance matrix C, which is a statistic, is obtained.
The method of obtaining this matrix is shown in the flowchart of FIG. In FIG. 13, first, the number of sample data K =
0, the element Cij = 0 of the variance-covariance matrix C is initialized (step S51), and the sum of K = K + 1 is calculated (step S52). Next, the calculation based on the above equation (2),

[0047]

[Equation 15]

That is, the above equation (18) is calculated (step S53). Then, it is determined whether or not K = S (step S54). If YES, the process ends, and if NO, the process returns to step S52. Next, the process of step S33,
That is, the Mahalanobis square distance of each cluster is calculated according to the flowchart of FIG. First, the number i = j = indicating the element of the covariance matrix C, that is, the index.
0, the square of Mahalanobis square distance Dm = 0 is initialized (step S61). Next, j = j + 1 and i = i +
The sum of 1 is calculated (steps S62 and S63), and the calculation based on the above equation (3) is performed.

[0049]

[Equation 16]

That is, the above equation (19) is calculated (step S64). Next, it is judged whether i = n and Y
If ES, the process proceeds to step S66, and if NO, the process returns to step S63. Similarly, it is determined whether or not j = n, and if YES, the process ends, and if NO, step S6.
Return to processing of 2.

FIG. 15 shows a plant operation support knowledge generation apparatus showing an embodiment (Embodiment 2) of the second invention. In FIG. 15, reference numeral 7 is a quantitative determination of the support knowledge discrimination performance based on the boundary surface based on the cluster boundary surface based on the conceptual skeleton 4 as the knowledge generation means and the operator situation judgment result based on the recorded data 3a of the recording means 3. It is a determination performance determination means that is obtained in a desired manner. The other parts are the same as those of the device of the first embodiment and are designated by the same reference numerals. FIG. 16 is a more detailed functional block diagram of the determination performance determination means. In FIG. 16, reference numeral 80 is a multivariate data input means for inputting multivariate data from the recorded data 3a, 81 is an operator judgment result input means for inputting a judgment result by the operator of the recorded data 3a, that is, a cluster, and 82 is a cluster sample data. Cluster-based determination result input means for inputting the determination result, 83 is multivariate data input means 80, operator determination result input means 8
1, misjudgment matrix generation means for generating a misjudgment matrix as shown in FIG. 17 from the cluster judgment result input means 82, and 84 is operated using the matrix generated by the misjudgment matrix generation means 83 to calculate the misjudgment rate. It is a Pij calculating means for calculating Pij. FIG. 17 is a diagram showing an example of a matrix which is generated by the discrimination performance determination means 7 and which calculates an erroneous discrimination rate. The matrix of FIG. 17 shows the result of classifying the recorded multivariate data according to the boundary surface of the cluster of the conceptual skeleton. In FIG.
Reference numeral 85 denotes a cluster index determined from the boundary surface, and 86 denotes a cluster index determined by the operator. In this figure, the number of clusters (i, j) is 3. The number of sample data belonging to the cluster 1 is 35 + 10 + 5 = 50 in the first line, and 35 out of 50 boundary surfaces are clusters,
It is determined that 10 clusters and 5 clusters. Then, each component of the matrix shows a correspondence relationship between the determination result by the operator of the recording data 3b and the determination result by the boundary surface of the conceptual skeleton obtained by the conceptual skeleton calculating means 4 of FIG. Further, the diagonal components of the matrix indicate the number of pieces of data in which the judgment result by the operator of the recorded data 3b and the judgment by the boundary surface obtained by the conceptual skeleton calculation means 4 are exactly the same. Therefore,
The greater the number of diagonal components of the matrix, the higher the accuracy of the determination by the boundary surface.

Next, the operation of the apparatus according to the second embodiment will be described. The operation is the same as that of the apparatus of the first embodiment until 5 is generated from the sample data of the plant. Next, the discrimination performance determination means 7 inputs the determination result by the boundary surface of the conceptual skeleton cluster created by the conceptual skeleton calculation means 4 and the determination result of the operator by the recording data 3b from the recording means 3. In the discrimination performance determination means 7, the multivariate data input means 80 inputs the multivariate data from the recording data 3b, the operator determination result input means 81 inputs the operator's determination result from the recording data 3b, and the cluster determination result input means. 82 inputs the determination result of the sample data by the cluster from the conceptual skeleton 5.
Then, the erroneous discrimination matrix generating means 83 generates the erroneous discrimination display matrix shown in FIG. Pij calculation means 84
Is based on this misidentification display matrix,

[0053]

[Equation 17]

The calculation according to the above equation (20) is performed to calculate Pij, which serves as an index quantitatively indicating the performance of the determination by the boundary surface of the cluster determined by the conceptual skeleton calculating means. In addition, Pii (i = 1, 2, ..., K) is used to calculate the misjudgment matrix as a numerator, but this shows the correct answer rate, and the accuracy of the judgment by the boundary surface of the cluster is shown. Shows. Therefore, the higher the Pii, the better. On the other hand, Pij (i ≠ j) indicates the erroneous determination accuracy of the determination based on the boundary surface of the cluster. When Pii and Pij are calculated using the equation (18) based on the example of FIG. 17, the results are as follows. Pii is P11 (%) =
{35 / (35 + 10 + 5)} × 100 = 70 (%),
Pij is P12 (%) = {10 / (35 + 10 +
5)} × 100 = 20 (%), P13 (%) = {5 /
(35 + 10 + 5)} × 100 = 10 (%) ... With this Pij, it is possible to quantitatively obtain the actual data determination performance of the conceptual skeleton, which is not obtained by the conventional method, the so-called driving support knowledge determination performance.

FIG. 18 is a plant operation support knowledge generation apparatus showing an embodiment (Embodiment 3) of the third invention. In FIG. 18, reference numeral 8 denotes a driving support knowledge generation device that displays the boundary surface of the cluster of the conceptual skeleton 5 on the coordinate axes. The functions of the other parts are the same as those of the device of the first embodiment, and the same reference numerals as those of the first embodiment are attached. FIG. 19 is a detailed functional block diagram of the driving support knowledge display means 8. 90
Is a multivariate data storage means for storing the multivariate data from the recording means 3, 91 is a recording means for storing the boundary surfaces, names, statistics, etc. of the clusters in the conceptual skeleton, and 92 is the multivariate data recording means 90 or the recording means. When the operator 2 selects the scale of coordinates for displaying the data 91, the coordinate axis selecting means 93, and when the plotting period of the sample data is designated, 93 synthesizes the coordinates selected by the coordinate axis selecting means 92 based on the plotting period. Display coordinate composition means, 9
Reference numeral 4 is a display means for displaying on the input / output device 1a of FIG. 18 based on the data from the display coordinate composition means 93. FIG. 20 shows a display example on the input / output device 1a by the display means 94. In FIG. 20, 95 is a knowledge display screen for driving support of the input / output device 1a, 96 is a coordinate selection display input unit, 97 is a plot period input unit for inputting a plot period, and 98 is coordinates at which the boundary surface of the cluster is displayed. is there.

Next, the operation of the apparatus according to the third embodiment will be described with reference to FIGS. Recording means 3
When the recorded data 3b is input to the driving support knowledge display means 8, the multivariate data storage means 90 stores time-series multivariate data. On the other hand, the storage unit 91 stores the boundary surface, the name, and the statistic of the cluster of the conceptual skeleton 5 calculated by the conceptual skeleton calculating unit 4. The multivariate that can be displayed at this time is a bivariate because it is a flat screen output. The operator 2 instructs the coordinates in the bivariate set from the coordinate axis selection input unit 96. This instruction is input from the input / output device 2a to the coordinate axis selection means 92 of the driving support knowledge display means 8. Further, the operator instructs the plotting period input unit 97 to the display coordinate synthesizing unit 93 of the driving support knowledge display unit 8 about the time-series plotting period of the start time and the end time. The display coordinate combination means 93 combines the coordinate data from the coordinate axis selection means 92, and the display means 94 is the input / output device 1a.
Instruct to display coordinates. As a result, the coordinates 98 and the boundary surface 9 of the cluster are displayed on the driving support knowledge display screen 95 of FIG.
8a is displayed. This display allows an inexperienced operator to visually confirm the boundary surface of the conceptual skeleton judged by the skilled operator, and can refer to the actual operation.

[0057]

According to the first aspect of the present invention, it is possible to objectively and quantitatively determine the boundary surface of the cluster from which the knowledge for supporting the operator is generated, without depending on the subjectivity of the operator. Therefore, the boundary planes that differ depending on the subjectivity of each operator can be unified, and it is possible to obtain highly reliable knowledge for supporting the operator. According to the second aspect of the invention, since the performance of determining the support knowledge by the boundary surface of the cluster can be quantitatively obtained, in addition to the effect of the first invention, the support knowledge can be obtained. There is an effect that it can quantitatively evaluate how much performance it has. According to the third invention, since the boundary surface of the cluster can be displayed on the CRT or the like,
In addition to the effect of the first aspect of the invention, an operator with a low degree of skill visually confirms the boundary surface of the cluster, so that it is possible to reasonably obtain support knowledge that requires experience. .

[Brief description of drawings]

FIG. 1 is an overall configuration diagram of a plant operation support knowledge generation device showing an embodiment according to the first invention.

FIG. 2 is a functional block diagram of a conceptual skeleton calculation means in the device of FIG.

FIG. 3 is a flowchart showing an operation in the apparatus of FIG.

FIG. 4 is a flowchart showing the operation of the apparatus of FIG.

5 is a circuit block diagram of a specific plant used in the apparatus of FIG.

6 is a diagram showing specific plant data of the plant of FIG.

7 is a diagram showing a description example of a conceptual skeleton obtained by using the apparatus of FIG. 1 in the plant of FIG.

FIG. 8 is a diagram showing a description example of a conceptual skeleton similar to FIG.

FIG. 9 is a diagram showing sample data on coordinates based on the conceptual skeletons of FIGS. 7 and 8.

10 is a diagram showing specific sample data by a general plant used in the apparatus of FIG.

FIG. 11 is a diagram when the sample data of FIG. 10 is divided into clusters.

12 is a flowchart in the case of calculating a statistic for each cluster using the sample data of FIG.

FIG. 13 is a flowchart for calculating a statistic amount for each cluster, similar to FIG.

14 is a flowchart in the case of calculating the Mahalanobis square distance of each cluster using the plant data of FIG.

FIG. 15 is an overall configuration diagram of a plant operation knowledge generation device showing an embodiment of the second invention.

16 is a functional block diagram of discrimination performance determination means in the apparatus of FIG.

FIG. 17 is a diagram showing an erroneous discrimination display matrix generated by the discrimination performance judging means of FIG. 16;

FIG. 18 is an overall configuration diagram of a plant operation knowledge generation apparatus showing an embodiment according to the third invention.

19 is a functional block diagram of driving support knowledge display means in the apparatus of FIG. 18. FIG.

20 is a diagram showing a display example by the driving support knowledge display means of FIG.

FIG. 21 is an overall configuration diagram of a conventional plant operation support knowledge generation device.

22 is a diagram for explaining the data structure in the device of FIG. 21. FIG.

FIG. 23 is a diagram for explaining a concept, a conceptual skeleton, and a time-series skeleton in a conventional plant operation support knowledge generation device.

[Explanation of symbols]

 1a Input / output device 1b CPU 2 Operator 3 Recording means 3a, 3b Recorded data 4 Conceptual skeleton calculation means 5 Conceptual skeleton 6 Plant monitoring device 7 Discrimination performance determination means 8 Operation support knowledge display means 31A Data classification means 32A Statistics amount calculation means 33A Square distance calculation means 34A Boundary surface determination means 80 Multivariate data input means 81 Operator judgment result input means 82 Judgment result input means by cluster 83 Misjudgment matrix generation means 84 Pij calculation means 90 Multivariate data storage means 91 Storage means 92 Coordinate axes Selection means 93 Display coordinate composition means 94 Display means

Claims (3)

[Claims] 1. When an operator determines a cluster indicating the situation of the observation target from the displayed sample data of the observation target, the boundary surface of the cluster is objectively and quantitatively determined based on the determined result. A knowledge generating device for generating support knowledge for an operator by determining, and recording means for recording the sample time and the multivariate data of the sample data, and recording the determination result including the cluster by the operator, The recording data of the recording means is classified into each cluster, the statistic amount of each cluster is calculated, and the calculation result includes a calculation including the calculation of the Mahalanobis square distance to determine the boundary surface of the cluster. A knowledge generation device characterized by being provided. 2. When an operator determines a cluster indicating the situation of the observation target from the displayed sample data of the observation target, the boundary surface of the cluster is objectively and quantitatively determined based on the determined result. A knowledge generating device for generating support knowledge for an operator by determining, and recording means for recording the sample time and the multivariate data of the sample data, and recording the determination result including the cluster by the operator, Knowledge calculating means for categorizing the recording data of the recording means to calculate the statistic amount for each cluster, and performing a calculation including calculation of Mahalanobis square distance in the calculation result to determine the boundary surface of the cluster, Based on the judgment result of the boundary surface of the cluster by this knowledge calculation means and the judgment result by the operator of the recorded data, A knowledge generation apparatus comprising: a discrimination performance determination unit that quantitatively obtains the performance of discrimination of support knowledge by the boundary surface. 3. When an operator determines a cluster indicating the situation of the observation target from the displayed sample data of the observation target, the boundary surface of the cluster is objectively and quantitatively determined based on the determined result. A knowledge generating device for generating support knowledge for an operator by determining, and recording means for recording the sample time and the multivariate data of the sample data, and recording the determination result including the cluster by the operator, Knowledge calculating means for categorizing the recording data of the recording means to calculate the statistic amount for each cluster, and performing a calculation including calculation of Mahalanobis square distance in the calculation result to determine the boundary surface of the cluster, A knowledge generating device, comprising: a supporting knowledge display means for displaying the boundary surface.