JP2020149181A - Data classification device - Google Patents

Abstract

To provide a data classification device which can visualize a relation between a state of a plant and an evaluation index in consideration of qualitative information of the plant and allows the relation to be used for operation guidance.SOLUTION: The data classification device of this invention has: an operation data database in which operation data of a plant is stored; an evaluation index database in which an evaluation index of the plant is stored; a qualitative data database in which qualitative data about the plant is stored; data classification means which uses a data clustering technique to classify operation data into a category and associates the category, the evaluation index, and the qualitative data with each other and learns relations between them; and classification result display means which displays a result of learning in the data classification means in a three-dimensional graph.SELECTED DRAWING: Figure 2

Description

The present invention relates to a data classification device that classifies operating data such as temperature, pressure, and flow rate.

Many sensors such as thermometers, pressure gauges, and flow meters are installed in power plants and chemical plants for the purpose of plant monitoring and control. In recent years, there has been an increasing demand for improving the operational efficiency of plants and the yield of products produced in plants by utilizing the measurement data of these sensors.

In order to improve evaluation indexes such as plant operation efficiency and product yield, it is necessary to model the relationship between the plant condition and the evaluation index.

As a background technique in this technical field, for example, Japanese Patent Application Laid-Open No. 2018-49314 (Patent Document 1) is available. In Patent Document 1, a data classification unit that classifies operation data into categories according to the degree of similarity, an evaluation index calculation unit that calculates an evaluation index of a category from the value of operation data, and an operation data included in each category are used. , The representative value of the operation data is calculated for each category, the identification information of each category is mapped two-dimensionally according to the similarity of the representative value of the operation data, and the identification information of the category is the first axis and the second. A plant data display processing device including a classification result display processing unit that generates three-dimensional image data in which the evaluation index of the category obtained by the evaluation index calculation unit is represented on the third axis while being represented on a plane consisting of the axes of. Is described (see summary).

JP-A-2018-49314

Patent Document 1 describes to visualize the relationship between the state of the plant and the evaluation index.

However, Patent Document 1 describes that the relationship between the state of the plant and the evaluation index is visualized by adding qualitative information of the plant, which does not change in conjunction with the fluctuation of the evaluation index. Not.

Therefore, the present invention provides a data classification device that can be used for operation guidance by visualizing the relationship between the state of the plant and the evaluation index by adding qualitative information of the plant.

In order to solve the above problems, the data classification device of the present invention includes an operation data database that stores plant operation data, an evaluation index database that stores plant evaluation indexes, and a qualitative data database that stores qualitative data related to the plant. , Operation data is classified into categories using data clustering technology, and the classified categories are associated with evaluation indexes and qualitative data, and the data classification means for learning these relationships and the data classification means It is characterized by having a classification result display means for displaying the learned result on a three-dimensional graph.

According to the present invention, it is possible to provide a data classification device that can be used for operation guidance by visualizing the relationship between the state of the plant and the evaluation index by adding qualitative information of the plant.

Issues, configurations and effects other than those described above will be clarified by the explanation of the following examples.

It is a graph which shows the relationship between the category number described in Patent Document 1 and an evaluation index. It is explanatory drawing which shows the data classification apparatus described in this Example. It is a graph which shows the relationship between the category number described in this Example, and an evaluation index. It is a system diagram of the plant described in this Example. It is explanatory drawing which shows the operation data stored in the operation data database. It is explanatory drawing which shows the evaluation index stored in the evaluation index database. It is explanatory drawing which shows the qualitative data stored in a qualitative data database. It is explanatory drawing which shows the qualitative data converted into the time series data. It is explanatory drawing which shows the classification method of the operation data using the adaptive resonance theory (ART). It is a schematic diagram when two-dimensional operation data is classified by ART. It is a schematic diagram in the case of classifying two-dimensional operation data in consideration of qualitative data. It is a flowchart which shows the data classification algorithm. It is a graph which shows the relationship between the category number classified by taking qualitative data into consideration, and the evaluation index. It is explanatory drawing which shows the data classification apparatus described in Example 2. It is a schematic diagram which shows unlearned data together with the classification of two-dimensional operation data in consideration of qualitative data. It is a graph which shows the estimation result together with the relationship between the category number classified by taking qualitative data into consideration, and the evaluation index.

Hereinafter, examples of the present invention will be described with reference to the drawings. The same or similar configurations are designated by the same reference numerals, and when the explanations are duplicated, the explanations may be omitted.

First, before explaining the embodiment of the present invention, the graph described in Patent Document 1 will be described.

FIG. 1 is a graph showing the relationship between the category number described in Patent Document 1 and the evaluation index.

The Z axis of this graph is an evaluation index, and the number displayed on the graph indicates a category number. In FIG. 1, it is shown that the evaluation index is high in the case of category number 1, and the evaluation index is low in the case of category numbers 3, 5, and 8. Further, since the position on the XY plane indicates the positional relationship between the category numbers, for example, it is visually that category number 5 is close to category number 8 and category number 1 and category number 7 are far from category number 8. It is confirmed.

Therefore, by using this information, for example, when the current state of the plant is category number 3, it is sufficient to aim for category number 1 which has a high evaluation index and is close to the position. Can be used for.

However, when it is used for operation guidance in this way, for example, even if the evaluation index is lowered to the same extent, the cause is the effect of changes in the composition of raw materials or the effect of deterioration of equipment performance. Qualitative effects of the plant may also be considered. The qualitative impact (information) of such a plant does not change in conjunction with the fluctuation of the evaluation index, and information cannot be simply added to this graph.

Therefore, the data classification device described in this embodiment takes into consideration the qualitative influence (information) of such a plant, visualizes the relationship between the state of the plant and the evaluation index, and utilizes it for operation guidance.

FIG. 2 is an explanatory diagram showing a data classification device described in this embodiment.

The data classification device described in this embodiment includes an operation data database 3, an evaluation index database 4, a qualitative data database 5, a data classification means 6, and a classification result display means 7. In the following, the "database" will be referred to as a "DB".

Then, the plant 1 and the control device 2 are connected to the data classification device described in this embodiment. Further, the data classification device, the plant 1, and the control device 2 constitute a plant control system.

The plant 1 has equipment, pipes connecting the equipment, valves, and the like. Further, in order to monitor and control the state of the plant 1, sensors such as a thermometer, a pressure gauge and a flow meter are installed.

The control device 2 monitors and controls the state of the plant 1 by using measurement data such as temperature, pressure and flow rate measured by sensors such as a thermometer, a pressure gauge and a flow meter installed in the plant 1. Then, using these measurement data, the operation amount such as the opening degree of the valve is determined, and the plant 1 is monitored and controlled.

The operation data DB 3 stores measurement data such as temperature, pressure and flow rate measured from the plant 1, operation amount data such as valve opening degree, and set value data for control as time series data. In addition, "measurement data, operation amount data, set value data" will be referred to as "operation data" below. Then, this operation data shows the state of the plant 1. That is, the operation data DB 3 stores the operation data of the plant 1.

The evaluation index DB 4 stores an evaluation index indicating the performance evaluation of the plant 1. In this embodiment, as the performance evaluation of the plant 1, the operational efficiency of the plant 1 (hereinafter, referred to as “plant efficiency”) required by the control device 2 is stored. The plant efficiency (evaluation index) may be obtained from the operation data stored in the operation data DB3. Further, for example, the composition of raw materials, the performance of equipment, and the like may be analyzed offline, and the analysis result may be stored as an evaluation index.

The qualitative data DB 5 stores qualitative data (qualitative information of the plant 1 that does not become a numerical value) with respect to the plant 1. In this embodiment, as qualitative data, information (inspection data) such as "valve A failure" and "reactor R4 deterioration" acquired by the inspection of the plant 1 is stored.

The data classification means 6 learns the relationship between the operation data stored in the operation data DB 3, the plant efficiency stored in the evaluation index DB 4, and the inspection data stored in the qualitative data DB 6. Specifically, multidimensional operation data is classified into categories using data clustering technology, and the classified categories are linked with plant efficiency (evaluation index) and inspection data (qualitative data). Learn these relationships.

The classification result display means 7 displays the result (learned result) classified by the data classification means 6 in a three-dimensional graph.

FIG. 3 is a graph showing the relationship between the category number described in this embodiment and the evaluation index.

Each axis of this three-dimensional graph is the same as in FIG. 1, the Z axis is an evaluation index (plant efficiency in this embodiment), and the XY plane is a plane showing the positional relationship of the categories in which the operation data is classified.

However, in FIG. 1, the category number is displayed as a legend in the representative value of each category, but in this embodiment, in addition to the category number, inspections such as "valve A failure" and "reactor R4 deterioration" are performed. Data (qualitative data) is displayed. The representative value of each category is the average value or weighted average (center of gravity) of the operation data included in the category.

That is, in the three-dimensional graph shown in this embodiment, the predetermined axis (for example, the Z axis) indicates the evaluation index (plant efficiency), and the other two axes (for example, the XY plane) are in the classified category. It is a graph showing representative values, and the category number and the qualitative data (inspection data) of the category number are shown in the legend of each plotted point.

As described above, in the present embodiment, for example, in the case of category numbers 3, 5 and 8, the information that the plant efficiency is lowered and the qualitative information of the plant 1 that causes the information (“valve A defect” and Inspection data such as "reactor R4 deterioration") is also displayed. This makes it possible to visualize information useful for driving guidance.

FIG. 4 is a system diagram of the plant described in this embodiment.

This system diagram is, for example, an electronic file created by CAD (computer aided design) software for creating a system diagram, and the system diagram includes equipment, pipes connecting the equipment, valves, thermometers, pressure gauges, and flow meters. Sensors (measuring equipment) such as are described.

The system diagram is also referred to as a piping instrumentation diagram or a P & ID (Piping & Instrument Flow Diagram). Information on equipment, pipes connecting equipment, valves, measuring equipment, etc. is described in the system diagram. In FIG. 4, the measuring device is tagged with F1, P1, T1, or the like. In this embodiment, F is a flow meter, P is a pressure gauge, and T is a thermometer. In addition, R indicates a reactor. In this embodiment, 28 measuring instruments, four reactors (R1 to R4), and three valves (A to C) are installed.

FIG. 5 is an explanatory diagram showing operation data stored in the operation data database.

The operation data stored in the operation data DB 3 is time-series data of flow rate, pressure, and temperature for each time. In this embodiment, each part (28 measuring instruments) of the system diagram (see FIG. 4) of the plant 1, such as the inlet flow rate T1 of the reactor R1, the inlet pressure P1 of the reactor R1, and the inlet flow rate F1 of the reactor R1). The measurement data (operation data) measured in is stored, for example, at 1-minute intervals.

FIG. 6 is an explanatory diagram showing an evaluation index stored in the evaluation index database.

The evaluation index DB 4 stores the plant efficiency indicating the performance evaluation of the plant 1 as an evaluation index. In this embodiment, the plant efficiency for each time (for example, 1 minute interval) required by the control device 2 is stored as time series data. In this embodiment, the plant efficiency and the operation data are recorded separately, but the plant efficiency may be recorded in the operation data shown in FIG.

FIG. 7 is an explanatory diagram showing qualitative data stored in the qualitative data database.

The qualitative data DB 5 stores qualitative data as qualitative information of the plant 1. In this embodiment, as qualitative data, inspection data such as "valve A defect" and "reactor R4 deterioration" acquired by the inspection of the plant 1 are stored. Then, in this embodiment, the time (start time) at which the valve A defect or the reactor R4 deterioration is confirmed by the inspection and the time (end time) at which the countermeasure is completed are stored.

However, when these measurement data (operation data), plant efficiency (evaluation index), and inspection data (qualitative data) are input to the data classification means 6, the inspection data (qualitative data) is input to the measurement data shown in FIG. It is necessary to convert to a data format for each time, such as (operation data) and plant efficiency (evaluation index) shown in FIG. Therefore, this inspection data (qualitative data) is converted into time series data.

FIG. 8 is an explanatory diagram showing qualitative data converted into time series data.

In this embodiment, for example, in the case of "valve A defective", "valve A defective" is stored from 10:01: 00 to 2018/3/10 15:15: 00, 2018/ "Normal" is stored from 3/10 15:16: 00 to 2018/4/1 20:00. In this way, the inspection data (qualitative data) is also stored as time-series data at the same time (for example, 1 minute interval) as the measurement data (operation data) and the plant efficiency (evaluation index). ..

In this embodiment, the inspection data is used as the qualitative data, but information indicating the composition of the raw material such as the production area of the raw material may be used as the qualitative data.

Next, the data classification means 6 will be described. In the data classification means 6, a learning type data clustering technique having teacher data is used to learn the relationship between multidimensional operation data and plant efficiency and inspection data. Before explaining this learning algorithm, a prerequisite data clustering technique will be described.

In this embodiment, Adaptive Resonance Theory (ART), which is one of unsupervised learning, is used as a data clustering technique. The data clustering technique is not limited to ART, and other data clustering techniques may be used.

ART is a model that simulates a human pattern recognition algorithm, and multidimensional data can be classified into a plurality of categories based on their similarity. Since the structure of ART is known, the description thereof will be omitted, and the method of classifying data using ART will be described.

FIG. 9 is an explanatory diagram showing a method of classifying operation data using adaptive resonance theory (ART).

In the upper figure of FIG. 9, the horizontal axis shows the time and the vertical axis shows the operation data. Further, in the lower figure of FIG. 9, the horizontal axis shows the time and the vertical axis shows the category number (state of plant 1).

Generally, the operation data to be input to the ART is four-dimensional or more data, but here, when the data is simplified and the two-dimensional time series data (data 1 and data 2) shown in the upper figure of FIG. 9 are classified. Will be described. The operation data for each time is two-dimensional time series data (input data) of data 1 and data 2. Here, for convenience of explanation, two operation data of data 1 and data 2 are shown in the upper figure of FIG. 9, but in this embodiment, the measurement is measured by 28 measuring devices. Since it has data (operation data), the operation data of data 1 to data 28 will be described. That is, 28 operation data (corresponding to the measurement data (operation data) measured by the 28 measuring devices) are input to the ART as input data.

When this two-dimensional time series data (operation data) is input to the ART as the data classification means 6, for example, the data in the region 1 in which the value of the data 1 is large and the value of the data 2 is small is in a certain category ( It is classified into category 1) (see the figure below in Fig. 9). Further, the data in the area 2 is classified into another category (category 2) because the relationship between the data in the area 1 and the data 1 and the data 2 is different (see the figure below in FIG. 9). Similarly, the data in the area 3 and the data in the area 4 are classified into different categories.

In this way, ART classifies the input data into a plurality of categories in chronological order based on the similarity. In this embodiment, the category number (number) is used when classifying the categories, but the category number is not limited to the number (number). In addition, "○" in the lower figure of FIG. 9 indicates the state of the plant 1 for each time (for example, 1 minute interval) in each category.

However, since ART is an unsupervised learning type algorithm, it is difficult to learn these three relationships of driving data, evaluation index, and qualitative data.

Therefore, in this embodiment, a supervised learning algorithm is further used. Before showing the learning algorithm used in this embodiment, the concept of the learning algorithm will be described with reference to FIGS. 10 and 11.

FIG. 10 is a schematic diagram when two-dimensional operation data is classified by ART.

In FIG. 10, each point indicates the operation data, and the shade of color of the operation data indicates the evaluation index. The darker the color, the lower the evaluation index (plant efficiency), and the lighter the color, the higher the evaluation index (plant efficiency).

In FIG. 10, the operation data is classified into four categories. However, the operation data classified into each category contains a mixture of high evaluation index and low evaluation index, and it is not possible to associate the category with the evaluation index. Further, the operation data included in the category number 1 includes the data of the abnormality A and the data of the abnormality B, and the category and the qualitative data (inspection data) cannot be associated with each other. Here, "abnormality A" and "abnormality B" are types of inspection data, and indicate "valve A failure", "reactor R4 deterioration", and the like.

Therefore, in this embodiment, the range of categories (size: adjusted by the caution coefficient described later) is reduced, the driving data included in one category is reduced, and the range of evaluation indexes (degree of variation described later) is set to a constant value. Below, the ratio of certain qualitative data (matching degree described later) is classified so as to be a certain value or more. In addition, "the range of the evaluation index is set to a certain value or less" means that the range of the evaluation index is set to a certain ratio or less, and "the ratio of a certain qualitative data is set to a certain value or more" means, for example, a certain category. The ratio of the abnormality A (for example, the abnormality having the highest ratio) contained in is set to a certain value or more.

That is, in this embodiment, the data classification means 6 sets the ratio of the qualitative data having the highest ratio among the qualitative data associated with the operation data classified into each category to a certain value or more, and the operation is classified into each category. Classify so that the range of evaluation indexes associated with the data is below a certain value.

FIG. 11 is a schematic diagram when two-dimensional operation data is classified in consideration of qualitative data.

Also in FIG. 11, each point indicates the operation data, and the shade of color of the operation data indicates the evaluation index. The darker the color, the lower the evaluation index (plant efficiency), and the lighter the color, the higher the evaluation index (plant efficiency).

In FIG. 11, the operation data is classified into eight categories. The operation data included in one category shown in FIG. 11 is less than the operation data included in one category shown in FIG. As a result, the variation in the evaluation index of the driving data classified into each category becomes small. For example, category number 1 in FIG. 10 is a mixture of low to high evaluation indexes, but in FIG. 11, the operation data included in category number 1 in FIG. 10 is classified into category numbers 1, 2, and 8. Category number 1 is a category with a relatively low evaluation index, category number 2 is a category with a medium evaluation index, and category number 8 is a category with a relatively high evaluation index. ..

Further, in the category number 1 of FIG. 10, the data of the abnormality A and the data of the abnormality B are mixed, but in FIG. 11, the operation data included in the category number 1 of FIG. 10 is the category numbers 1 and 2. , 8 is classified into three categories, so that the category number 1 contains only the operation data of the abnormality A, and the category number 8 contains only the operation data of the abnormality B. As a result, for example, the driving data classified into category number 1 can be associated with the driving data, the evaluation index, and the qualitative data, such as "driving data having a low evaluation index and an abnormality A". That is, it is possible to learn the relationship between driving data, evaluation index, and qualitative data.

Here, the details of the data classification algorithm that enables such classification will be described.

FIG. 12 is a flowchart showing a data classification algorithm.

In step 1, the operation data, the evaluation index, and the qualitative data are read from the operation data DB3, the evaluation index DB4, and the qualitative data DB5. Since these data are time-series data for each time as shown in FIGS. 5, 6 and 8, the relationship between the operation data at a certain time and the evaluation index and the qualitative data is shown.

In step 2, setting a warning coefficient p _j for each category (a j). Here, vigilance coefficient p _j is a parameter which determines the range of categories, takes a value between 0 and 1. The closer it is to 1, the smaller the category range, and the closer it is to 0, the larger the category range. In this embodiment, the alert coefficient _pj is set for each category j, and the range of the category j is adjusted. The alert coefficient _pj is set by the operator as an initial value.

In step 3, using the alert coefficient p _j that is set in step 2, using the ART (cluster analysis techniques) shown in FIG. 9, classifies the operation data to each category.

In step 4, the degree of matching of qualitative data and / or the degree of variation in the evaluation index are calculated for the operation data classified into each category.

Matching degree M _j qualitative data in category j is defined by Equation (1) it is determined from equation (1).

M _j = Nm / N _j (1)
Here, N _m is the number of the largest number (highest proportion) of qualitative data in the operation data (N _j ), and N _j is the number of operation data of the category j.

For example, the number of operation data classified into category j is 100 (N _j ), of which the number of operation data that was abnormal A is 70 (Nm), the number of normal operation data is 25, and the abnormality B. When the number of operating data is 5, the qualitative data having the highest ratio is anomaly A, so the number of qualitative data is 70, and the matching degree M _j of the qualitative data is 0.7.

Further, variation of V _j of metrics, the metrics and Z, variation of V _j of metrics, the maximum value of the evaluation index Z which is categorized j - minimum (hereinafter referred to as "range" Will be explained).

For example, if the maximum value of the evaluation index (plant efficiency) Z of the operation data classified into the category j is 0.954 and the minimum value of the evaluation index (plant efficiency) Z is 0.941, the degree of variation V _j of the evaluation index is 0.013. ..

In step 5, M _j and V _j obtained in step 4 for each category j are compared with the set threshold value (set value), and it is determined whether or not the condition is satisfied.

Specifically, assuming that the set value of the matching degree of the qualitative data is Mset, when the matching degree M _j of the qualitative data is larger than the set value Mset, the matching degree is high, and it is determined that the condition is satisfied. Further, assuming that the set value of the degree of variation of the evaluation index is Vset, if the degree of variation V _j of the evaluation index is smaller than the set value Vset, the degree of variation is low, and it is judged that the condition is satisfied.

Here, the set value Mset is a "constant value" that determines the ratio of certain qualitative data, and the set value Vset is a "constant value" that determines the range of the evaluation index. The set value Mset and the set value Vset are set in advance by the operator according to the purpose of use of the graph.

When both the conditions of the matching degree and the variation degree are satisfied (YES), the state shown in FIG. 11 is obtained, and the relationship between the category (driving data) and the evaluation index and the qualitative data can be learned. move on.

On the other hand, if any one of the conditions matching degree and variation degree does not satisfy (NO) are the state shown in FIG. 10, it is necessary to change the alert coefficients p _j, the process proceeds to step 6.

In step 6, since the alert coefficient p _j is changed, the amount of change in the alert coefficient p _j is calculated.

Here, the alert coefficient p _j before the change is p _j (k), and the alert coefficient p _j after the change is p _j (k + 1). In this embodiment, equations (2) to (4) are used to obtain _pj (k + 1).

p _j (k + 1) = max (pV _j · pM _j ) (2)

That is, pV _j determined from the degree of variation in the evaluation index and pM _j determined from the degree of matching of the qualitative data are obtained, and the larger of these is defined as p _j (k + 1), and the changed alert coefficient p _j. And. That is, pV _j and pM _j indicate the update value of the warning coefficient obtained according to the degree of "deviation" from each set value.

In the equation (3), the maximum value of p is p _max, and the larger the degree of variation of the evaluation index, the closer pV _j becomes to p _max . Note that a _V is the adjustment coefficient.

Similarly, in the equation (4), the maximum value of p is p _max, and the smaller the matching degree of the qualitative data, the closer pM _j becomes to p _max . Note that a _M is the adjustment coefficient.

In step 7, the alert coefficient p _j is changed by using the p _j obtained in step 6.

The data classifying section 6 as obtains vigilance coefficient of categories p _j, the matching degree M _j qualitative data, the variation of V _j metrics.

Then, the process returns to step 3, step 3, using the alert coefficient p _j to be changed in step 7, classifying the operation data to each category.

By repeating the above steps, the relationship between the multidimensional operation data and the evaluation index and the qualitative data can be learned. In this embodiment, a range is used as the degree of variation of the evaluation index, but a standard deviation, a variance, or the like may be used.

Finally, the classification result display means 7 will be described. The classification result display means 7 displays (outputs) the results classified by the data classification means 6 in a three-dimensional graph.

As described above, according to this embodiment, the relationship between the operation data and the evaluation index and the qualitative data can be learned at the same time, and these relationships can be visualized. Therefore, the operation considering both the evaluation index and the qualitative data. Guidance can be provided.

FIG. 13 is a graph showing the relationship between the category numbers classified in consideration of qualitative data and the evaluation index. FIG. 13 shows the classification result shown in FIG. 11 in a three-dimensional graph.

The Z-axis of the three-dimensional graph is the evaluation index (plant efficiency), and the XY plane is the plane showing the positional relationship of the categories in which the operation data is classified.

In general, the operation data is three-dimensional or more, and in that case, the positional relationship of the categories cannot be accurately reproduced on the XY plane. In this embodiment, the positional relationship of data of three or more dimensions is simulated on a two-dimensional XY plane by the multidimensional scaling method.

In the example of FIG. 11, for the sake of simplicity, the operation data is shown in two dimensions, and the positional relationship of each category is shown in two dimensions. Therefore, the positional relationship of each category in FIG. 11 shown in two dimensions is strictly shown in the XY plane of FIG.

In addition to category numbers 1 to 8, inspection data such as "abnormality A" and "normal" are also displayed in the legend. By displaying the evaluation index (plant efficiency) and qualitative data (inspection data) corresponding to each category number in this way, the state of plant 1 can be visualized more clearly and utilized for clearer operation guidance. be able to.

In this embodiment, the classification result display means 7 displays the relationship between the operation data and the evaluation index by using a three-dimensional graph with the evaluation index as the Z axis, but plots based on the value of the evaluation index. You may use a two-dimensional graph in which the color of the point to be used is changed. Further, when the number of dimensions of the operation data is three or more, the positional relationship of the categories may be represented by a three-dimensional graph, and the color of the points to be plotted may be changed based on the value of the evaluation index.

In this way, according to this embodiment, the relationship between the plant state (for example, the category classified based on the operation data) and the evaluation index is visualized by adding the qualitative information (qualitative data) of the plant. However, it can be used for driving guidance.

In Example 1, the relationship between the driving data acquired in advance and the evaluation index and qualitative data is learned and visualized and used for driving guidance. However, in this embodiment, only the driving data can be obtained and the evaluation index is used. And an example when the qualitative data is unknown.

That is, the data classification device described in this embodiment learns the relationship between the operation data and the evaluation index and the qualitative data offline in advance, so that the evaluation index and the qualitative data are newly obtained. It estimates the data.

FIG. 14 is an explanatory diagram showing the data classification device described in the second embodiment.

In the data classification device described in the present embodiment, the unlearned data estimation means 8 is added to the data classification device described in the first embodiment.

In this embodiment, the performance index and the qualitative data are estimated from the newly obtained driving data by the step of learning and visualizing the relationship between the driving data and the evaluation index and the qualitative data offline and the unlearned data estimation means 8. It consists of two steps. Since the former is as described in the first embodiment, the description thereof is omitted here. The latter unlearned data estimation means 8 will be described below.

The processing of the unlearned data estimation means 8 differs depending on the category in which the operation data is classified by the data classification means 6.

First, a case where the obtained driving data is classified into one of the categories classified at the time of learning will be described.

For example, when the classification result (learning category) of the driving data learned from the offline data (driving data, evaluation index, qualitative data) has the relationship shown in FIG. 11, new driving data (used in the learning category). If the non-driving data) is classified into any of the learning categories of category numbers 1 to 8, the evaluation index and qualitative data associated with the learning category can be estimated.

That is, in the unlearned data estimation means 8, when the learning category classified by the data classification means 6 includes the estimated estimation category, the evaluation index and the qualitative data associated with the classified learning category are displayed. Output as an estimated value.

Next, a case where new driving data is not in the learning category and is classified into a new category will be described.

For example, when the new driving data is not classified into any learning category and is classified into the new category number 9, since there is no evaluation index and qualitative data associated with the category number 9, the evaluation index and qualitative data are used. It cannot be estimated.

Therefore, when the new driving data is not classified into any learning category, the unlearned data estimation means 8 uses the evaluation value and qualitative data of the learning category in the vicinity of the new category to obtain the evaluation index and qualitativeness of the new category. Estimate the data.

The specific algorithm will be described below. Let J be the number of categories classified at the time of learning, x _{j be} the representative value (average value of driving data) of the driving data of the learning category j (j = 1,2, .., J), and the representative of the evaluation index. _Let the value be Z _Vj and the value of the qualitative data be Z _Mj .

The representative value (estimated value) Z _{V (J + 1)} of the evaluation index of the new (estimated) category (J + 1 ₎ is calculated by the equations (5) and (6).

Note that d _j is the distance from the new category (J + 1) to each learning category j in the same multidimensional space as the dimension of the driving data, and f (x) is a function that decreases as x increases. Yes, in this embodiment, equation (7) is used.

f (x) = 1 / x (7)
That is, the representative value of the evaluation index of the new category (J + 1) is obtained from the weighted average of the learning category at the time of learning. Then, the coefficient becomes a value inversely proportional to the distance. With this algorithm, the estimated value (representative value) of the evaluation index can be obtained even in a new category. Note that f (x) is not limited to the equation (7).

In addition, the value (estimated value) Z _{M (J + 1)} of the qualitative data of the new (estimated) category (J + 1) has a probability P (estimated value) that the estimated value is Z _{M (J + 1)} because the qualitative data is not a numerical value. Z _{M (J + 1)} ) is also calculated. In this embodiment, the number of categories in which the type of qualitative data is k (k = 1,2, .. K) is Jk, and the probability P (k) is calculated by the equation (8).

Here, f (x) uses equation (7), and d _j is the distance from the new category (J + 1) to each learning category j. Then, the coefficient becomes a value inversely proportional to the distance.

That is, the value of the qualitative data of the new category (J + 1) is obtained from the weighted average of the frequency of the learning category at the time of learning. Then, this value is a probability that considers the distance from the new category (J + 1) to the frequency of the type k of the qualitative data.

In this way, the unlearned data estimation means 8 evaluates the new driving data associated with the category number 9 even if the new driving data is not classified into any of the learning categories and is classified into the new category number 9. Indicators and qualitative data can be estimated.

FIG. 15 is a schematic diagram showing unlearned data combined with a classification of two-dimensional operation data in consideration of qualitative data.

For example, when the representative value of the new driving data category is point A, it is close to learning category number 7 (abnormal B) and learning category number 8 (abnormal B), and then close to learning category number 1 (abnormal A). .. That is, in this case, the qualitative data is presumed to be anomalous B.

FIG. 16 is a graph showing the relationship between the category numbers classified in consideration of qualitative data and the evaluation index, and the estimation results.

As shown in FIG. 16, the coordinates on the three-dimensional graph are specified from the XY coordinates of the point A and the estimated value Z _{Vj of the} evaluation index, and the estimated value Z _{M (J + 1) of the} qualitative data and its probability are combined. Display P (Z _{M (J + 1)} ).

In this way, even when the new driving data is not classified into any learning category but is classified into the new category number 9, the evaluation index and the qualitative data can be estimated and visualized.

The XY coordinates of the point A were determined so that the coordinates of the representative values of the learning category on the XY plane were fixed and e _{(J + 1)} in the equation (9 ₎ was minimized.

Here, d' _j is the distance on the XY plane from the representative value of the new category (J + 1) to the representative value of the learning category, and d _j is the representative value of the new category (J + 1) as described above. The distance in multidimensional space to the representative value of the learning category. As a result, the XY coordinates of the point A can be obtained without changing the XY coordinates of the representative value of the learning category.

In this embodiment, the case where one new category is generated has been described, but the processing when a plurality of operation data are classified and a plurality of new categories are generated is also the representative value of the learning category. It can be obtained so that the XY coordinates of are not changed.

If the XY coordinates of the representative value of the learning category may change, all (J + 1) XY coordinates of the new category and the learning category may be obtained by multidimensional scaling.

In this way, the unlearned data estimation means 8 uses the weighted average of the evaluation indexes associated with the classified learning categories and the weighted average of the evaluation indexes when the estimated estimation categories are not included in the learning categories classified by the data classification means 6. The weighted average of the frequency of qualitative data is output as an estimated value.

That is, in the data classification device described in this embodiment, in particular, the data classification means 6 classifies the operation data (measurement data) into categories by using the data clustering technique, and the classified categories and evaluation indexes ( By associating plant efficiency) and qualitative data (inspection data), the relationship between operation data and evaluation index and qualitative data is learned offline in advance, and newly obtained operation data based on the learned relationship. Estimate evaluation index (plant efficiency) and qualitative data (inspection data) for (measurement data).

Further, the classification result display means 7 displays the results learned by the data classification means 6 and the unlearned data estimation means 8 on a three-dimensional graph.

As described above, according to this embodiment, the state of the plant (for example, the learning category classified based on the operation data) in which the qualitative information (qualitative data) of the plant obtained in advance offline is added. Based on the relationship between and the evaluation index, newly obtained driving data can be visualized and used for driving guidance. Therefore, it can also be used for operation guidance when plant operation data is obtained in real time.

The present invention is not limited to the above-mentioned examples, and includes various modifications. For example, the above-described embodiment is described in detail in order to explain the present invention in an easy-to-understand manner, and is not necessarily limited to those having all the configurations described. Further, a part of the configuration of one embodiment can be replaced with a part of the configuration of another embodiment, and the configuration of another embodiment can be added to the configuration of one embodiment.

1: Plant 2: Control device 3: Operation data database 4: Evaluation index database 5: Qualitative data database 6: Data classification means 7: Classification result display means 8: Unlearned data estimation means

Claims (11)

An operation data database that stores plant operation data and
An evaluation index database that stores the evaluation index of the plant and
A qualitative data database that stores qualitative data about the plant,
The operation data is classified into categories using a data clustering technique, and the classified categories are associated with the evaluation index and the qualitative data, and a data classification means for learning the relationship between the classified categories and the data classification means.
A data classification device comprising: a classification result display means for displaying the result learned by the data classification means on a graph. The graph is a three-dimensional graph in which the evaluation index is shown on a predetermined axis and the positional relationship of representative values of the classified categories is shown on the other two axes, and the category number and the qualitative data of the category number are displayed. The data classification apparatus according to claim 1, wherein the data classification device is shown. The graph is a two-dimensional graph or a three-dimensional graph showing the positional relationship of representative values of the classified categories and changing the color of the points plotted based on the value of the evaluation index, and is a category number and the category. The data classification device according to claim 1, wherein qualitative data of numbers are shown. The data classification device according to claim 1, wherein the qualitative data is stored as time-series data at the same time as the operation data and the evaluation index. The data classification device according to claim 1, wherein the data clustering technique is an adaptive resonance theory. The data classification means sets the ratio of the qualitative data having the highest ratio among the qualitative data associated with the operation data classified into each category to a certain value or more, and is an evaluation index associated with the operation data classified into each category. The data classification device according to claim 1, wherein the data is classified so that the range is set to a certain value or less. The data classification device according to claim 1, wherein the data classification means obtains a caution coefficient of a category, a matching degree of qualitative data, and a variation degree of an evaluation index. An operation data database that stores plant operation data and
An evaluation index database that stores the evaluation index of the plant and
A qualitative data database that stores qualitative data about the plant,
The operation data is classified into categories using data clustering technology, and the classified categories are associated with the evaluation index and the qualitative data, and the operation data, the evaluation index, and the qualitative data are linked in advance offline. Data classification means to learn the relationship between
An unlearned data estimation means that estimates the evaluation index and qualitative data from newly obtained driving data based on the relationship between the driving data learned in advance offline and the evaluation index and qualitative data.
A data classification device comprising the data classification means and the classification result display means for displaying the results learned by the unlearned data estimation means on a graph. The data classification device according to claim 8, wherein the unlearned data estimation means estimates a category of driving data obtained in real time. When the category classified by the data classification means includes the estimated category, the unlearned data estimation means outputs an evaluation index and qualitative data associated with the classified category as an estimated value. The data classification device according to claim 9. When the unlearned data estimation means does not include the estimated category in the categories classified by the data classification means, the weighted average of the evaluation indexes associated with the classified categories and the frequency of the qualitative data. The data classification apparatus according to claim 8, wherein the weighted average is output as an estimated value.

Images