JP7173395B1 - Analysis device, analysis method and program - Google Patents

Abstract

A technique capable of analyzing the cause of an abnormality is provided. An analysis device according to one aspect of the present disclosure classifies performance data collected from a target into a plurality of groups including an abnormality group composed of performance data when an abnormality has occurred in the target. and a grouping unit configured to calculate, by causal inference, an item that causes a difference between the abnormal group and other groups with respect to items included in the performance data. a difference factor calculation unit configured to rank the items in descending order of relevance to the abnormality using the analysis result of the causal inference; an anomaly condition calculation unit configured to calculate, with respect to the item, conditions under which the anomaly occurs. [Selection drawing] Fig. 16

Description

The present disclosure relates to an analysis device, an analysis method, and a program.

2. Description of the Related Art Techniques for analyzing factors of anomalies when anomalies occur in facilities such as plants, production facilities, and industrial equipment have been known for some time. For example, in Patent Document 1, targeting the energy efficiency of equipment, the data related to the energy of the equipment is classified into a group during normal operation and a group other than that, and the factors that cause the difference from the group during normal operation are determined. A tree-based analysis technique is disclosed.

JP 2020-47245 A

However, in the prior art, the accuracy of estimating a factor that causes an abnormality is not necessarily high, and for example, even a factor that has changed under the influence of the abnormality may be estimated as a factor of the abnormality. This is because the conventional technology considers only the correlation between factors using a machine learning method such as a decision tree, and does not consider the relationship between the cause and the result when an abnormality occurs.

The present disclosure has been made in view of the above points, and an object thereof is to provide a technique capable of analyzing the cause of an abnormality.

An analysis device according to an aspect of the present disclosure is configured to classify performance data collected from a target into a plurality of groups including an abnormality group composed of performance data when an abnormality occurs in the target. and a difference factor calculation configured to calculate, by causal inference, an item that causes a difference between the abnormal group and other groups with respect to the items included in the performance data. a ranking calculation unit configured to rank the items in descending order of relevance to the abnormality using the analysis results based on the causal inference; an anomaly condition calculator configured to calculate a condition for the anomaly to occur.

A technology is provided that can analyze the cause of the abnormality.

It is a figure showing an example of the whole analysis system composition concerning this embodiment. It is a figure which shows an example of the operation performance data stored in operation performance DB. It is a figure which shows an example of the abnormality track record data stored in abnormality track record DB. It is a figure which shows an example of the hardware constitutions of the analysis apparatus which concerns on this embodiment. It is a figure showing an example of functional composition of an analysis device concerning this embodiment. It is a figure which shows an example of the grouping performance data memorize|stored in the grouping performance data memory|storage part. 4 is a flow chart showing an operation example of the analyzer according to the present embodiment; 8 is a flowchart showing an example (part 1) of grouping calculation processing according to the present embodiment; 9 is a flowchart showing an example (part 2) of grouping calculation processing according to the present embodiment; FIG. 10 is a diagram showing an example of area designation on a scatter diagram; 9 is a flow chart showing an example (part 3) of grouping calculation processing according to the present embodiment; 9 is a flowchart showing an example of difference factor calculation processing according to the embodiment; It is a figure which shows an example of a causal graph. 6 is a flowchart showing an example of ranking calculation processing according to the present embodiment; 6 is a flowchart showing an example of abnormal condition calculation processing according to the present embodiment; FIG. 10 is a diagram showing an example of items ranked in terms of relevance to anomalies and their anomaly conditions;

An embodiment of the present invention will be described below. In the following embodiments, an analysis system 1 capable of analyzing the factors of anomalies (in particular, analyzing the factors of anomalies in consideration of the relationship between causes and effects when an anomaly occurs) will be described.

<Overall Configuration Example of Analysis System 1>
FIG. 1 shows an example of the overall configuration of an analysis system 1 according to this embodiment. As shown in FIG. 1, the analysis system 1 according to the present embodiment includes an analysis device 10, a facility 20, an input/output device 30, an operation record DB 40, an abnormality record management device 50, and an abnormality record DB 60. included.

The analysis device 10 is a general-purpose server that analyzes the causes of anomalies that have occurred in the equipment 21 in the facilities 20 using the operation performance data stored in the operation performance DB 40 and the abnormality performance data stored in the abnormality performance DB 60. and various information processing devices such as PCs (personal computers).

The facility 20 is, for example, various facilities such as a factory. Facility 20 includes facility 21 , sensor 22 , actuator 23 , and measurement control device 24 . The facilities 21 are, for example, plants, production facilities, industrial equipment (for example, boilers, heat exchangers, etc.), various devices, and the like. However, these are only examples, and the equipment 21 is not limited to these, and any equipment, equipment, apparatus, or the like that is the object of analysis may be used. The sensor 22 is a measuring device that measures information representing various states of the facility 21 (for example, pressure, temperature, flow rate, current, voltage, etc.). Actuator 23 is a driving device that operates facility 21 . The actuator 23 also measures the operation of the equipment 21 (for example, the presence or absence of some kind of operation such as a stirring operation, and measures the speed of the operation, etc.). The measurement control device 24 controls the sensor 22 and the actuator 23 , acquires (collects) measured values measured by the sensor 22 and the actuator 23 , and stores these measured values in the operation record DB 40 . The measurement control device 24 also transmits these measured values to the abnormality record management device 50 .

The input/output device 30 is various devices that function as an input/output interface for the analysis device 10 . Examples of the input/output device 30 include a keyboard, mouse, display, and touch panel.

The operational performance DB 40 stores operational performance data. A specific example of the operation performance data will be described later.

The abnormality record management device 50 detects an abnormality by a known abnormality detection technique using the measured value received from the measurement control device 24, and stores the detection result in the abnormality record DB 60 as abnormality record data.

The abnormality record DB 60 stores abnormality record data. A specific example of the abnormal record data will be described later.

Note that the overall configuration of the analysis system 1 shown in FIG. 1 is an example, and is not limited to this. For example, at least one of the analysis device 10, the input/output device 30, the operation record DB 40, the abnormality record management device 50, and the abnormality record DB 60 may be integrally configured.

<Operating performance data>
FIG. 2 shows an example of operation performance data stored in the operation performance DB 40. As shown in FIG. As shown in FIG. 2, the operation result data includes a plurality of items, an item number of each of the plurality of items, a measurement value of each of the plurality of items, and a time indicating the date and time when the measurement values were received. Includes stamps.

For example, the operation result data shown in FIG. 2 includes items from item "item 1" with item number "001" to item "item N" with item number "N". In addition, the operational performance data shown in FIG. 2 includes measurement values of each item for each time stamp. In addition, in the example shown in FIG. 2, the time stamp is every one second, but this is just an example and the present invention is not limited to this.

Here, the item is information representing the state, operation, etc. of the facility 21 . Items include, for example, pressure, temperature, flow rate, current, voltage, presence/absence of stirring operation, and stirring speed. Therefore, a measured value of a certain item is a value obtained by measuring information related to this item by the sensor 22, the actuator 23, or the like.

Hereinafter, a set of each time stamp and a measurement value associated therewith will be referred to as an "operation result record". In the operation result data shown in FIG. 2, the operation result record can be expressed in a format of (time stamp, measurement value of item 1, measurement value of item 2, . . . , measurement value of item N). It should be noted that, with this notation, the operation result data can also be regarded as time-series data of the operation result record.

<Abnormal performance data>
FIG. 3 shows an example of the abnormality record data stored in the abnormality record DB 60. As shown in FIG. As shown in FIG. 3, the abnormality record data includes an abnormality number identifying an abnormality, the contents of the abnormality corresponding to the abnormality number, and the date and time when the contents of the abnormality occurred.

For example, the abnormality record data shown in FIG. 3 includes the contents of the abnormality from the abnormality content "power amount abnormality" of the abnormality number "0001" to the abnormality content "steam flow rate excessive" of the abnormality number "L" and the date and time of occurrence of the abnormality. ing.

<Hardware Configuration Example of Analyzer 10>
FIG. 4 shows a hardware configuration example of the analysis device 10 according to this embodiment. As shown in FIG. 4, the analysis device 10 according to the present embodiment includes an external I/F 101, a communication I/F 102, a RAM (Random Access Memory) 103, a ROM (Read Only Memory) 104, and an auxiliary storage device. 105 and a processor 106 . Each of these pieces of hardware is communicably connected via a bus 107 .

The external I/F 101 is an interface with an external device such as the recording medium 101a. Examples of the recording medium 101a include a CD (Compact Disc), a DVD (Digital Versatile Disk), an SD memory card (Secure Digital memory card), a USB (Universal Serial Bus) memory card, and the like.

Communication I/F 102 is an interface for connecting analysis device 10 to a communication network. A RAM 103 is a volatile semiconductor memory (storage device) that temporarily holds programs and data. The ROM 104 is a non-volatile semiconductor memory (storage device) that can retain programs and data even when power is turned off. The auxiliary storage device 105 is, for example, a non-volatile storage device such as a HDD (Hard Disk Drive) or SSD (Solid State Drive), and stores programs and data. The processor 106 is, for example, various arithmetic devices such as a CPU (Central Processing Unit) and a GPU (Graphics Processing Unit).

Note that the hardware configuration shown in FIG. 4 is an example, and the analysis device 10 may have other hardware configurations. For example, the analysis device 10 may have a plurality of auxiliary storage devices 105 and a plurality of processors 106, and various hardware other than the illustrated hardware (for example, input devices such as keyboards and mice, displays, output device such as a display panel, etc.).

<Example of functional configuration of analysis device 10>
FIG. 5 shows an example of the functional configuration of the analysis device 10 according to this embodiment. As shown in FIG. 5, the analysis device 10 according to this embodiment includes a grouping unit 201, a difference factor calculation unit 202, a ranking calculation unit 203, an abnormal condition calculation unit 204, and an information presentation unit 205. . These units are implemented by, for example, processing that one or more programs installed in the analysis apparatus 10 cause the processor 106 to execute. The analysis device 10 according to this embodiment also has a grouping result data storage unit 206 and a model data storage unit 207 . These units are realized by, for example, the auxiliary storage device 105 or the like.

The grouping unit 201 groups the operation result records included in the operation result data using the operation result data stored in the operation result DB 40 and the abnormality result data stored in the abnormality result DB 60. For example, for a certain abnormality, the grouping unit 201 divides each operation result into a group composed of operation result records when the abnormality occurs and a group composed of operation result records when the abnormality does not occur. Group records. Hereinafter, a group composed of operation result records when a certain abnormality occurs is called an "abnormality group", and a group composed of operation result records when the abnormality does not occur is called a "normal group". do.

Further, the grouping unit 201 stores the grouped operation result records as grouped result data in the grouped result data storage unit 206 . Details of the grouping result data will be described later.

The difference factor calculation unit 202 uses the grouped result data representing the operation result records grouped by the grouping unit 201 to determine items (called factors) that are difference factors between groups by causal inference. ) are analyzed. For example, the difference factor calculation unit 202 analyzes the items (factors) that are the difference factors between the abnormal group and the normal group by causal inference. Here, as a causal inference algorithm, LiNGAM (Linear Non-Gaussian Acyclic Model) described in References 1 and 2, for example, can be used. In LiNGAM, as an output of the algorithm, a causal graph represented by a weighted directed graph composed of nodes representing items (factors) and edges representing causal relationships between nodes and their strengths is obtained. The causal relationship between nodes is represented by the direction of the edge, and the strength of the causal relationship is represented by the weight given to the edge.

In addition, the difference factor calculation unit 202 stores model data representing a causal graph obtained as a result of causal inference in the model data storage unit 207 .

The ranking calculation unit 203 ranks items (factors) in descending order of relevance to anomaly using grouped performance data and model data, taking causality into consideration.

The abnormal condition calculation unit 204 calculates a condition under which an abnormality occurs (hereinafter also referred to as an abnormal condition) with respect to items (factors) with high rankings.

The information presentation unit 205 presents various information to the input/output device 30 such as a display. For example, the information presenting unit 205 displays the top ranking items (factors) and their abnormal conditions on the input/output device 30 such as a display.

The grouping performance data storage unit 206 stores grouping performance data. The model data storage unit 207 stores model data.

<Grouping performance data>
An example of grouping result data stored in the grouping result data storage unit 206 is shown in FIG. As shown in FIG. 6, the grouping result data includes a time stamp indicating the reception date and time of the measured value, a plurality of items, the measured value of each of the plurality of items, and a group label. That is, the grouped performance data is obtained by assigning a group label to each operation performance record.

For example, the grouped performance data shown in FIG. Labels and included. Note that in the example shown in FIG. 6, operation result records with a group label of "0" belong to the normal group, and operation result records with a group label of "1" belong to the abnormal group. However, this is just an example, and the possible values of the group label are not limited to 0 or 1. For example, when the operational performance data are grouped into three or more groups, the group label can take any one of three values.

<Example of Operation of Analyzer 10>
An operation example of the analyzer 10 according to this embodiment will be described below with reference to FIG.

First, the analysis device 10 executes grouping calculation processing using the operation performance data stored in the operation performance DB 40 and the abnormality performance data stored in the abnormality performance DB 60, and creates grouping performance data ( step S101). Next, the analysis device 10 executes difference factor calculation processing using the grouped performance data stored in the grouped performance data storage unit 206, and creates model data representing a causal graph (step S102). Next, the analysis apparatus 10 executes ranking calculation processing using model data stored in the model data storage unit 207, and ranks items (factors) highly related to abnormality (step S103). Then, the analysis device 10 executes the abnormal condition calculation process, calculates the abnormal conditions of the items (factors) with the highest ranking, and displays the items and the abnormal conditions on the input/output device 30 such as a display ( step S104).

The details of each process of steps S101 to S104 will be described below.

<Grouping calculation processing>
Grouping calculation processing (part 1) to grouping calculation processing (part 3) will be described below as examples of grouping calculation processing.

<<Grouping calculation processing (Part 1)>>
Grouping calculation processing (1) according to the present embodiment will be described below with reference to FIG.

The grouping unit 201 receives an anomaly as a reference for grouping (step S201). For example, the grouping unit 201 causes the input/output device 30 such as a display to display the abnormality number, the content of the abnormality, the date and time of occurrence, and the like included in the abnormality record data stored in the abnormality record DB 60, and then selects it by the user. It is only necessary to accept the error of the specified error number. However, this is only an example, and in addition to this, the grouping unit 201 may also receive, for example, a preset abnormality, an abnormality with a high frequency of occurrence, an abnormality with a large number of occurrences, an abnormality that occurred most recently, Abnormalities and the like may be accepted. In the following, as an example, it is assumed that "abnormality A" has been received.

Next, the grouping unit 201 receives target periods to be grouped (step S202). The grouping unit 201 may, for example, accept a period input by the user through the input/output device 30 such as a keyboard or mouse as the target period. However, this is only an example, and in addition to this, the grouping unit 201 may also include, for example, a preset period, a period in which the frequency of occurrence of an abnormality is high, a period in which the frequency of occurrence of an abnormality is high, and A period or the like may be accepted as a target period.

Next, the grouping unit 201 receives a target item as a basis for grouping and a threshold for the target item (step S203). The grouping unit 201 may accept items and thresholds selected or specified by the user through the input/output device 30 such as a keyboard and a mouse as target items and thresholds. However, this is only an example, and in addition to this, the grouping unit 201 may accept, for example, preset items and thresholds as target items and thresholds. may be set and only the other may be selected or specified by the user. In the following, as an example, it is assumed that abnormality A has occurred when the value of the target item exceeds the threshold, and abnormality A has not occurred when the value of the target item is equal to or less than the threshold.

Next, the grouping unit 201 compares the target item values included in the operation result records within the target period with the threshold value, and groups these operation result records (step S204). That is, the grouping unit 201 assigns the group label “0” to the operation result records whose target item value is equal to or less than the threshold value for the operation result records within the target period, for example, and classifies them into the normal group. A group label of "1" is given to the operation result record exceeding the threshold value, and the record is classified into the abnormal group.

Then, the grouping unit 201 stores the operation result record to which the group label is assigned in step S204 as grouped result data in the grouped result data storage unit 206 (step S205).

In addition, in the above grouping calculation process (part 1), the operation result records are grouped into two groups, but the grouping is not limited to this, and may be grouped into three or more groups. This also applies to the grouping calculation process (part 2) and the grouping calculation process (part 3), which will be described later.

<<Grouping calculation processing (Part 2)>>
Grouping calculation processing (2) according to the present embodiment will be described below with reference to FIG.

The grouping unit 201 receives an anomaly as a reference for grouping (step S301), as in step S201 of FIG.

Next, the grouping unit 201 receives target periods to be grouped (step S302), as in step S202 of FIG.

Next, the grouping unit 201 receives two target items (step S303). For example, the grouping unit 201 displays a list of items included in the operation result record within the target period on the input/output device 30 such as a display, and accepts two items selected by the user as target items. All you have to do is However, this is only an example, and in addition to this, the grouping unit 201 may accept, for example, two preset items as target items, or only one target item may be selected or specified by the user. However, the other target item may be determined according to the target item. In the following, as an example, it is assumed that "item 1" and "item 2" are accepted as two target items.

Next, the grouping unit 201 causes the input/output device 30, such as a display, to display a scatter diagram with the two target items on the vertical axis and the horizontal axis, respectively, for the operation result record within the target period (step S304).

Next, the grouping unit 201 receives an area specification operation on the scatter diagram (step S305). The area specifying operation is an operation for specifying a connected area on the scatter diagram displayed on the input/output device 30 such as a display. A typical example is an operation of designating a connected region using a circle, an ellipse, a rectangle, or the like. Such operations can be performed using, for example, a pointing device such as a mouse, a touch panel, or the like. Also, for example, a keyboard or the like may be used to specify an expression, a numerical range, or the like representing a connected region. In addition, in the region designation operation, together with designation of a connected region, a value of a group label to be assigned to each operation result record within the connected region may be designated. For example, a scatter diagram 1100 shown in FIG. 10 shows a case where an area specifying operation for specifying an area 1101 and an area specifying operation for specifying an area 1102 are performed.

Next, the grouping unit 201 groups the operation performance records corresponding to the points within the area (closed area) specified by the area specifying operation into the same group (step S306). For example, in the example shown in FIG. 10, it is assumed that the group label "1" is specified by the area specifying operation for specifying the area 1101, and the group label "0" is specified by the area specifying operation for specifying the area 1102. . In this case, the grouping unit 201 assigns the group label “0” to the operation result records corresponding to the points in the area 1102 to classify them into the normal group, and classifies the operation result records corresponding to the points in the area 1101 into the normal group. A group label of "1" is given to it, and it is classified into an abnormal group.

8, the grouping unit 201 stores the operation result record to which the group label is assigned in step S306 as grouped result data in the grouped result data storage unit 206 (step S307). ).

<<Grouping calculation processing (Part 3)>>
The grouping calculation process (part 3) according to this embodiment will be described below with reference to FIG. 11 .

The grouping unit 201 receives an abnormality as a grouping criterion (step S401), as in step S201 of FIG.

Next, the grouping unit 201 receives target periods to be grouped (step S402), as in step S202 of FIG.

Next, the grouping unit 201 receives an arithmetic expression including one or more items included in the operation performance record within the target period and a threshold value for the arithmetic expression (step S403). The grouping unit 201 may, for example, display a list of arithmetic expressions and a list of threshold values on the input/output device 30 such as a display, and then receive the arithmetic expression and threshold values selected by the user. However, this is only an example, and in addition to this, the grouping unit 201 may accept, for example, a preset arithmetic expression and threshold value, or may accept an arbitrary arithmetic expression and threshold value input by the user. may Here, the calculation formula is a formula for calculating some value from the values of one or more items included in the operation result record. For example, if there is a facility 21 that uses input energy to do some work and emits output energy, and the operation result record includes items “input energy” and “output energy”, input energy and An arithmetic expression for obtaining energy efficiency from output energy can be used. In the following, as an example, if the value calculated by the arithmetic expression exceeds the threshold, it is assumed that abnormality A has occurred, and if the value calculated by the arithmetic expression is equal to or less than the threshold, abnormality A has not occurred. do.

Next, the grouping unit 201 calculates the value of the arithmetic expression received in step S403 for the operation performance record within the target period (step S404).

Next, the grouping unit 201 compares the value of the arithmetic expression calculated in the above step S303 with the threshold value received in the above step S403 with respect to the operation result records within the target period, and divides these operation result records into Group (step S405). That is, the grouping unit 201 assigns the group label “0” to the operation result records for which the value of the arithmetic expression is equal to or less than the threshold value for the operation result records within the target period, for example, and classifies them into the normal group. A group label "1" is given to the operation result record whose value exceeds the threshold, and it is classified into the abnormal group.

Then, as in step S205 of FIG. 8, the operation result record to which the group label is assigned in step S405 is stored as grouped result data in the grouped result data storage unit 206 (step S406).

<Difference Factor Calculation Processing>
Difference factor calculation processing according to the present embodiment will be described below with reference to FIG. 12 .

The difference factor calculation unit 202 acquires grouped performance data from the grouped performance data storage unit 206 (step S501). Note that, for example, when a plurality of grouped performance data with different target periods are stored in the grouped performance data storage unit 206, the difference factor calculation unit 202 acquires the grouped performance data for the target target period.

Next, the difference factor calculation unit 202 creates a causal model by causal inference using the group label of the grouped performance data as the objective variable and the other items as explanatory variables (step S502). Note that the difference factor calculation unit 202 may use, for example, LiNGAM described in References 1 and 2 as an algorithm for causal inference. As a result, a causal model is obtained in which the group label is the objective variable and the items other than the group label are explanatory variables. This causal model is represented by a weighted directed graph composed of nodes representing objective variables or explanatory variables and edges representing causal relationships between nodes and their strengths. and its strength.

For example, FIG. 13 shows an example of a causal model when the objective variable is a group label related to "abnormality A" and the explanatory variables are the items "flow rate" and "steam differential pressure". In the example shown in _FIG . ₁₃ , w1 represents the strength of the causal relationship from the item "flow rate" to "steam differential pressure", and w2 represents the group label for the item "steam differential pressure" to "abnormality A". It represents the strength of the causal relationship between This makes it possible to consider not only the causal relationship, but also its strength and causal order.

Next, the difference factor calculation unit 202 stores the model data representing the causal model created in step S502 in the model data storage unit 207 (step S503).

Then, the information presenting unit 205 displays the causal model created in step S502 on the input/output device 30 such as a display (step S504). This allows the user to confirm the causal model. However, this step does not necessarily have to be executed.

<Ranking calculation process>
The ranking calculation process according to this embodiment will be described below with reference to FIG.

The ranking calculation unit 203 acquires grouping result data from the grouping result data storage unit 206 (step S601). For example, when a plurality of grouped performance data with different target periods are stored in the grouped performance data storage unit 206, the ranking calculation unit 203 acquires the grouped performance data for the target target period.

Next, the ranking calculation unit 203 acquires model data stored in the model data storage unit 207 (step S602). Note that, for example, when a plurality of model data are stored in the model data storage unit 207, the ranking calculation unit 203 acquires the model data of the causal model corresponding to the grouped performance data acquired in step S601. .

Next, the ranking calculation unit 203 performs regression analysis on the causal model using the group label of the grouped performance data as the objective variable and the other items as explanatory variables (step S603). As such regression analysis, for example, the method described in reference 3 can be used. By using the method described in reference 3, it is possible to perform regression analysis considering the causal relationship (causal order and its strength) represented by the causal model.

Next, the ranking calculation unit 203 sorts the items corresponding to the explanatory variables, using the coefficients (regression coefficients) of the explanatory variables in the regression analysis results executed in step S603 as the importance of the explanatory variables. (Step S604). That is, for example, if the regression coefficient of the explanatory variable corresponding to item _n is an ( _n =1, . . . , _N ), the ranking calculation unit 203 calculates the Sort item n. As a result, the item _n corresponding to each explanatory variable is ranked in descending order of importance (that is, in descending order of relevance to abnormality A) using the regression coefficient an as the importance of the explanatory variable. Since the regression analysis in the above step S603 is performed on the causal model, the relationship with the abnormality here is the causal relationship (causal order and It should be noted that it is a relevance that considers its strength).

Then, the information presentation unit 205 displays the ranking of each item obtained in step S604 on the input/output device 30 such as a display (step S605). As a result, the user can check the items (factors) highly related to the abnormality in a ranking format, considering the causal relationship with the abnormality. However, this step does not necessarily have to be executed.

<Abnormal condition calculation processing>
Abnormal condition calculation processing according to the present embodiment will be described below with reference to FIG. 15 .

The abnormal condition calculation unit 204 acquires the above M items from among the items ranked in the ranking calculation process (step S701). That is, the abnormal condition calculation unit 204 acquires the top M items in the ranking. M (where M≦N) may be a preset value, or may be a value selected or specified by the user.

Next, the abnormal condition calculation unit 204 acquires grouped performance data from the grouped performance data storage unit 206 (step S702). For example, when a plurality of grouped performance data with different target periods are stored in the grouped performance data storage unit 206, the abnormal condition calculation unit 204 acquires the grouped performance data for the target target period.

Next, the abnormal condition calculation unit 204 executes decision tree analysis using the group label of the grouped performance data as the objective variable and the items acquired in step S701 (that is, the top M items in the ranking) as the explanatory variables. (step S703). In this decision tree analysis, a decision tree is created in which the values that the objective variable can take for the leaf nodes (the values that the group label can take) and the branching conditions for the explanatory variables are set for the nodes other than the leaf nodes. .

Next, the abnormal condition calculation unit 204 calculates the branch set in the node included in the route from the root node to the leaf node (hereinafter also referred to as an abnormal node) to which the group label value indicating the abnormality is set. A condition and items included in the branch condition are acquired (step S704). For the sake of simplicity, it is assumed below that the route from the root node to the abnormal node satisfies the branch condition set for the nodes included in the route. If there are multiple routes from the root node to the abnormal node, the abnormal condition calculation unit 204 calculates the branch condition set for each node included in the route and the branch condition included in the branch condition for each of these multiple routes. Items should be acquired. The branch condition acquired in this step represents a condition under which an abnormality occurs with respect to the items (factors) included in the branch condition. Therefore, the branch condition acquired in this step is an abnormal condition.

Then, the information presenting unit 205 displays the items and abnormal conditions acquired in step S705 (that is, the top M ranking items and their abnormal conditions) on the input/output device 30 such as a display (step S705). ). As an example, when M=3, the top three items in the ranking are "flow", "steam differential pressure", and "outside temperature", and the abnormal conditions for each of these items are "x>35.5", " FIG. 16 shows a display example when x>0.1" and "x<25.5". As shown in FIG. 16, the input/output device 30 such as a display displays the top three ranking items and their abnormal conditions. This allows the user to know the items (factors) that are highly relevant to the anomaly and the conditions that cause an anomaly due to the item, taking into consideration the causal relationship with the anomaly.

<Summary>
As described above, the analysis apparatus 10 according to the present embodiment can analyze the cause of the abnormality in consideration of the causal relationship (causal order and its strength) with the abnormality. In addition, the analysis device 10 according to this embodiment can present the analysis results to the user in a ranking format. Therefore, the user can know the factors highly related to the anomaly in a ranking format, considering the causal relationship with the anomaly. Therefore, for example, when the abnormality occurs, the user can know which factor (item) should be preferentially dealt with, and can appropriately and promptly deal with the abnormality.

The present invention is not limited to the specifically disclosed embodiments described above, and various variations, modifications, combinations with known techniques, etc. are possible without departing from the scope of the claims. be.

[References]
Reference 1: Shohei Shimizu et al., A Linear Non-Gaussian Acyclic Model for Causal Discovery, Journal of Machine Learning Research 7 (2006) 2003-2030.
Reference 2: Shohei Shimizu et al., DirectLiNGAM: A Direct Method for Learning a Linear Non-Gaussian Structural Equation Model, Journal of Machine Learning Research 12 (2011) 1225-1248.
Reference 3: Patrick Blobaum et al., ESTIMATION OF INTERVENTIONAL EFFECTS OF FEATURES ON PREDICTION, 2017 IEEE INTERNATIONAL WORKSHOP ON MACHINE LEARNING FOR SIGNAL PROCESSING.

1 analysis system 10 analysis device 20 facility 21 equipment 22 sensor 23 actuator 24 measurement control device 30 input/output device 40 operation performance DB
50 Abnormality record management device 60 Abnormality record DB
101 External I/F
101a recording medium 102 communication I/F
103 RAM
104 ROMs
105 auxiliary storage device 106 processor 107 bus 201 grouping unit 202 difference factor calculation unit 203 ranking calculation unit 204 abnormal condition calculation unit 205 information presentation unit 206 grouping result data storage unit 207 model data storage unit

Claims (10)

a grouping unit configured to classify performance data collected from a target into a plurality of groups including an anomaly group composed of performance data when an abnormality occurs in the target;
a difference factor calculation unit configured to calculate, by causal inference, an item that causes a difference between the abnormal group and another group, with respect to the items included in the performance data;
a ranking calculation unit configured to rank the items in descending order of relevance to the abnormality using the analysis result of the causal inference;
an abnormality condition calculation unit configured to calculate a condition under which the abnormality occurs with respect to a predetermined number of the items at the top of the ranking;
analyzer. 2. The analysis apparatus according to claim 1, further comprising an information presentation unit configured to display on a display unit a predetermined number of items ranked high in the ranking and conditions under which the abnormality occurs with respect to the items. The grouping unit
2. The analyzing apparatus according to claim 1, wherein said performance data is classified into said plurality of groups according to magnitude relationship between a value of a predetermined item included in said performance data and a predetermined threshold value. The grouping unit
2. The method according to claim 1, wherein said performance data is classified into said plurality of groups according to areas specified by a user on a scatter diagram relating to values of two predetermined items included in said performance data. Analysis equipment. The grouping unit
2. The performance data is classified into the plurality of groups according to a magnitude relationship between a calculated value relating to the value of one or more predetermined items included in the performance data and a predetermined threshold value. The analyzer described in . The difference factor calculation unit
2. The analyzer according to claim 1, configured to calculate a causal model using the group label representing the group as an objective variable and the items included in the performance data as explanatory variables by the causal inference. 7. The analyzer of claim 6, wherein the causal model is LiNGAM. The ranking calculation unit
Regression analysis is performed on the causal model using the group label as an objective variable and items included in the performance data as explanatory variables, and the items are ranked in descending order of the regression coefficient of the regression analysis. 8. The analyzer according to claim 6 or 7, wherein A grouping procedure for classifying performance data collected from a target into a plurality of groups including an abnormality group composed of performance data when an abnormality occurs in the target;
a difference factor calculation procedure for calculating, by causal inference, an item that causes a difference between the abnormal group and other groups with respect to the items included in the performance data;
A ranking calculation procedure for ranking the items in descending order of relevance to the anomaly using the analysis results based on the causal inference;
an abnormality condition calculation procedure for calculating a condition under which the abnormality occurs with respect to a predetermined number of the items at the top of the ranking;
A computer-implemented analysis method. A grouping procedure for classifying performance data collected from a target into a plurality of groups including an abnormality group composed of performance data when an abnormality occurs in the target;
a difference factor calculation procedure for calculating, by causal inference, an item that causes a difference between the abnormal group and other groups with respect to the items included in the performance data;
A ranking calculation procedure for ranking the items in descending order of relevance to the anomaly using the analysis results based on the causal inference;
an abnormality condition calculation procedure for calculating a condition under which the abnormality occurs with respect to a predetermined number of the items at the top of the ranking;
A program that makes a computer run

Images