KR20140055311A

KR20140055311A - System and method for diagnosing status

Info

Publication number: KR20140055311A
Application number: KR1020120121952A
Authority: KR
Inventors: 공정민; 오규삼; 김형찬; 유창준
Original assignee: 삼성에스디에스 주식회사
Priority date: 2012-10-31
Filing date: 2012-10-31
Publication date: 2014-05-09
Also published as: KR102045874B1

Abstract

A state diagnostic system and method are disclosed. The condition diagnosis system according to an embodiment of the present invention may include a reasoning model generation unit that generates a reasoning model including nodes having at least one condition formula using data collected during a predetermined period and state values according to the collected data An inference model correcting unit for correcting the inference model so as to improve the accuracy of the inference model, and a state inference unit for inserting the input data into the corrected inference model and deducing the state value according to the input data.

Description

SYSTEM AND METHOD FOR DIAGNOSING STATUS [0001]

More particularly, the present invention relates to a state diagnosis system and method, and more particularly, to a state diagnosis system and method that generate a reasoning model using data collected for a predetermined period of time, To a state diagnostic system and method.

Diagnosis of existing facilities and equipments is done mainly by a specialist putting the data collected from the sensors installed in each facility and equipments into a rule set by experts. However, according to the conventional method, if the data collected from the sensors is similar to the condition of the predetermined rule but does not completely match, there is a problem that it is difficult to determine the state of the facility and equipment to be diagnosed.

According to the existing method, the accuracy and reliability of the diagnosis result are degraded because the probability of occurrence of the failure is not reflected according to the use history of each facility and equipment, and the failure to reflect the newly generated failure pattern . In addition, there is a problem in that if a highly trained specialist is input and the current state of each facility and equipment is not analyzed, the failure state of the facility and equipment can not be detected, so that it can not be immediately addressed.

Embodiments of the present invention provide a state diagnostic system and method that can automate state diagnosis of each facility and equipment and diagnose it in real time.

The condition diagnosis system according to an embodiment of the present invention includes a reasoning model for generating a reasoning model composed of nodes having at least one conditional expression using data collected for a predetermined period and state values according to the collected data, Generating unit; A reasoning model correcting unit for correcting the reasoning model so that the accuracy of the reasoning model is improved; And a state inference unit for inserting the input data into the corrected inference model to deduce a state value according to the input data.

A method for diagnosing a state according to an embodiment of the present invention is a method for diagnosing a state of an object by using an inference model comprising nodes having at least one conditional expression using data collected for a predetermined period and state values according to the collected data ; Correcting the inference model so that the inference model correcting unit improves the accuracy of the inference model; And a step of inferring the state value according to the input data by substituting the input data for the state inference unit into the corrected inference model.

According to an embodiment of the present invention, an inference model is generated using collected data for a predetermined period of time, and the state of the corresponding equipment or device is diagnosed according to the currently collected data through the generated inference model, The state diagnosis can be performed in real time and the state diagnosis can be performed in real time. In this case, it is possible to immediately deal with the equipment inferred to be in a failed state, thereby reducing energy waste. The error rate is minimized while directly inputting the verification data into the generated inference model, thereby improving the accuracy and reliability of the state diagnosis.

BRIEF DESCRIPTION OF THE DRAWINGS FIG. 1 is a diagram showing the configuration of a state diagnostic system according to an embodiment of the present invention; FIG.
Figure 2 illustrates data stored in a database in accordance with an embodiment of the present invention.
3 illustrates an inference model generated by an inference model generation unit according to an embodiment of the present invention.
4 is a diagram illustrating a state in which leaf nodes of a reasoning model correcting part inference model according to an embodiment of the present invention are pruned.
FIG. 5 is a diagram illustrating a reasoning model finally corrected by a reasoning model correcting unit according to an embodiment of the present invention. FIG.
6 is a diagram for explaining a method of diagnosing a state of a facility when an error rate of a leaf node of a finally corrected inference model according to an embodiment of the present invention exceeds a preset reference value.
FIG. 7 illustrates a method of diagnosing a state of a facility when an error rate of a leaf node of a finally corrected inference model according to another embodiment of the present invention exceeds a preset reference value. FIG.
8 is a flowchart illustrating a state diagnostic method according to an embodiment of the present invention.

Hereinafter, the status diagnosis system and method of the present invention will be described in detail with reference to FIG. 1 through FIG. However, this is an exemplary embodiment only and the present invention is not limited thereto.

In the following description, a detailed description of known functions and configurations incorporated herein will be omitted when it may make the subject matter of the present invention rather unclear. The following terms are defined in consideration of the functions of the present invention, and may be changed according to the intention or custom of the user, the operator, and the like. Therefore, the definition should be based on the contents throughout this specification.

The technical idea of the present invention is determined by the claims, and the following embodiments are merely a means for efficiently describing the technical idea of the present invention to a person having ordinary skill in the art to which the present invention belongs.

1 is a block diagram illustrating a configuration of a state diagnostic system according to an embodiment of the present invention. Hereinafter, the building management system will be described as an example in which the state diagnosis system is applied. However, the present invention is not limited thereto and may be applied to various other fields.

1, the state diagnosis system 100 includes a database 102, a data collection unit 104, an inference model generation unit 106, an inference model correction unit 108, a state inference unit 110, And an extracting unit 112.

The database 102 may store data collected during a past period of time by the data collecting unit 104 and state values of corresponding equipments (or devices) according to the collected data. 2 is a diagram illustrating data stored in a database according to an embodiment of the present invention. Here, 300 sensing values (s1, s2, s3, s4) and state values corresponding thereto are stored in the database 102 for convenience of explanation, but the present invention is not limited thereto. Although four sensors are installed in a specific facility or apparatus, the number of sensors is not limited thereto.

Referring to FIG. 2, the database 102 stores sensing values s1, s2, s3, and s4 (i) periodically collected from a data collection unit 104 including a first sensor to a fourth sensor ) Are stored. Here, the first sensor, the second sensor, the third sensor, and the fourth sensor may be, for example, a temperature sensor installed in a cooling apparatus in a building, but the present invention is not limited thereto.

The database 102 stores the status values of the corresponding cooling devices according to the sensed values s1, s2, s3, and s4 collected at respective predetermined time points. At this time, when the state value is 0, it indicates that the cooling apparatus is in the normal state. When the state value is 1, it indicates that the cooling apparatus is in the abnormal state. If the state value is 2, .

Here, it is described that the database 102 stores the sensing values s1, s2, s3, and s4 collected from the sensors installed in the air conditioner and the state values according to the sensing values collected at a certain point of time, And the database 102 may store sensed values collected from sensors installed in various facilities or devices as well as the state values of corresponding facilities (or devices).

The data collecting unit 104 collects data from each sensor (not shown) installed in a facility or equipment to diagnose the condition. For example, the data collecting unit 104 acquires sensing data from various sensors (for example, a temperature sensor, a pressure sensor, a humidity sensor, a motion sensor, and the like) installed in an air conditioner, a cooling device, a heating device, Can be collected. At this time, the data collecting unit 104 may periodically collect data from each sensor (not shown).

The inference model generation unit 106 generates an inference model using data collected during a past period of the database 102 and state values according to the collected data. At this time, the inference model may be composed of nodes having at least one conditional expression that can infer the state of the facility.

The inference model generation unit 106 may classify the collected data for a certain period of time into learning data and verification data, and then generate an inference model using the state values of the learning data and the learning data. Here, the inference model generation unit 106 may randomly classify the data collected during a certain period of time into learning data and verification data, and classify the learning data and the verification data so that they have the same number or a certain ratio. However, the present invention is not limited to this, and the inference model generation unit 106 may classify the collected data for a certain period of time in various other ways.

The inference model generation unit 106 may generate a decision tree using, for example, the state values of the learning data and the learning data. In this case, the generated decision tree is an inference model for inferring the state value of the corresponding equipment according to the data collected by the data collection unit 104. Here, the decision tree is described as an example of the inference model, The present invention is not limited to this decision tree, but may be formed in various other forms.

The inference model generation unit 106 can generate a decision tree using algorithms such as ID3, C4.5, C5.0, CART, CHAID, and the like. A method of generating a decision tree using such an algorithm is well known in the art, so a detailed description thereof will be omitted.

3 is a diagram illustrating an inference model generated by an inference model generation unit according to an embodiment of the present invention. Here, the decision tree is generated using the CART algorithm, but the present invention is not limited thereto. In addition, here, the number of learning data is set to 150 for convenience.

Referring to FIG. 3, the inference model includes one root node RN1 and nine leaf nodes LN1 through LN9. When the data collected by the data collecting unit 104 is substituted into the inference model, the state value of the facility at the present time can be inferred. That is, if sensor values (s1, s2, s3, s4) collected at a certain point in time from the root node (RN) to each leaf node (LN) are compared with the corresponding conditional expression, The state value of the facility can be deduced.

Specifically, the first leaf node LN1 satisfies the condition (1) (s3 <2.45), and the state value of the facility is 0 (that is, normal) and the number of learning data n . The second leaf node LN2 satisfies the conditional expression (2) (s4 <1.75), conditional expression (3) (s3 <4.95), and conditional expression (4) (s4 <1.65) The state value of the facility is 1 (that is, an abnormal symptom), and the number of learning data belonging to the facility is 47. The third leaf node LN3 satisfies the conditional expression 1 (s3 <2.45), the conditional expression 2 (s4 <1.75) and the conditional expression 3 (s3 <4.95) are satisfied and the conditional expression 4 (s4 <1.65) is not satisfied In this case, the state value of the facility is 2 (i.e., failure), and the number of learning data belonging to the facility is one.

The fourth leaf node LN4 does not satisfy the conditional expression 1 (s3 <2.45), the conditional expression 2 (s4 <1.75) is satisfied, the conditional expression 3 (s3 <4.95) And the conditional expression ⑥ (s1 <6.95), the state value of the corresponding equipment is 1, and the number of learning data belonging to the equipment is two. The fifth leaf node LN5 does not satisfy the conditional expression 1 (s3 <2.45), satisfies the conditional expression 2 (s4 <1.75), satisfies the conditional expression 3 (s3 <4.95) Is satisfied, and conditional expression ⑥ (s1 <6.95) is not satisfied. The state value of the corresponding facility is 2, and the number of learning data belongs to one. The sixth leaf node LN6 does not satisfy the conditional expression 1 (s3 <2.45), the conditional expression 2 (s4 <1.75) is satisfied, the conditional expression 3 (s3 <4.95) and the conditional expression 5 (s4 ≥ 1.55) In this case, the state value of the facility is 2, and the number of learning data pertaining to the facility is 3.

The seventh leaf node LN7 satisfies the conditional expressions ⑦ (s3 <4.85) and conditional expression ⑧ (s1 <5.95) without satisfying the conditional expressions (s3 <2.45) and conditional expressions The state value of the equipment is 1, and the number of learning data belonging to it is one. The eighth leaf node LN8 does not satisfy the conditional expressions (s3 <2.45) and conditional expressions (s4 <1.75), satisfies the conditional expression (7) (s3 <4.85) In this case, the state value of the facility is 2, and the number of learning data belonging to the facility is 2. The ninth leaf node LN9 has a condition value of 2 for the facility, and the learning condition belongs to the learning condition belonging to the ninth leaf node LN9. The ninth leaf node LN9 does not satisfy the conditional expressions (s3 <2.45), conditional expressions (s4 <1.75) The number of data is 43.

Here, if the data collected by the data collection unit 104, i.e., the sensing values s1, s2, s3, and s4, are substituted into the inference model shown in Fig. 3, . However, there is a limit to the accuracy of inferring the state value of the facility at the present time through the inference model.

That is, since the inference model is generated by using the data collected during a certain period of time in the past and the corresponding state values, it can only make inferences based on empirical and statistical probabilities of the state values of the facilities at the present time, The inferred state value includes a certain error rate. Therefore, a separate process is required to reduce the error rate. To this end, the inference model correcting unit 108 corrects the inference model.

The inference model correcting unit 108 may correct the inference model generated by the inference model generating unit 106 using data classified as verification data among the data collected during the past period of time in the database 102. [ Specifically, the inference model correcting unit 108 substitutes the inference model into the inference model generated by the inference model generating unit 106, and calculates the error rate with respect to the state value of the corresponding facility inferred by the inference model. That is, since the verification data includes the sensing values s1, s2, s3 and s4 and the corresponding state values of the corresponding equipment, the sensing values s1, s2, s3 and s4 of the verification data are substituted into the inference model, It is possible to calculate the error rate for the state value of the corresponding facility inferred by the inference model through whether the state value of the facility coincides with the state value of the corresponding facility according to the actual sensing values (s1, s2, s3, s4).

At this time, the inference model correcting unit 108 calculates the error rate of the state value of the corresponding facility estimated by the inference model while changing the number of the leaf nodes of the inference model generated by the inference model generating unit 106 . For example, the inference model correcting unit 108 can check the error rate of the state value of the corresponding facility deduced by the inference model while decreasing the number of leaf nodes of the inference model by one (referred to as "pruning" .

That is, the reasoning model shown in FIG. 3 has nine leaf nodes LN1 to LN9. Here, the inference model correcting unit 108 may reduce the number of leaf nodes by one by merging a pair of leaf nodes having the same parent node with the parent node. For example, the fourth leaf node LN4 and the fifth leaf node LN5 have the same parent node. Here, if the fourth leaf node LN4 and the fifth leaf node LN5 are merged into the parent node, the number of leaf nodes in the inference model is reduced from nine to eight.

4 is a diagram showing a state in which leaf nodes of a reasoning model correcting part inference model according to an embodiment of the present invention are pruned.

3 and 4, when the inference model correcting unit 108 merges the fourth leaf node LN4 and the fifth leaf node LN5 into a parent node, the corresponding parent node is a new leaf node, that is, 10 leaf node LN10. At this time, the state value of the tenth leaf node LN10 may be a state value of a leaf node to which a larger number of learning data among the fourth leaf node LN4 and the fifth leaf node LN5 belongs. That is, since the number of learning data belonging to the fourth leaf node LN4 is two and the number of learning data belonging to the fifth leaf node LN5 is one, the state value of the tenth leaf node LN10 is four The state value of the leaf node LN4 can be set to one.

Next, the number of leaf nodes of the inference model is decreased from 9 to 8, and the inference model is corrected. Then, the inference model correction unit 108 substitutes the verification data into the corrected inference model, Calculates the error rate for the inferred state values of the corresponding facility.

Next, the inference model correcting unit 108 merges the second leaf node LN2 having the same parent node and the third leaf node LN3 into the parent node, thereby changing the number of leaf nodes of the inference model from 8 to 7 You can reduce one more. In this case, the parent node becomes a new leaf node, that is, the eleventh leaf node LN11, and the state value of the eleventh leaf node LN11 becomes one. That is, since the number of learning data belonging to the second leaf node LN2 is 47 and the number of learning data belonging to the third leaf node LN3 is one, the state value of the eleventh leaf node LN11 is 2 The state value of the leaf node LN2 can be set to one.

Next, the number of leaf nodes of the inference model is reduced from 9 to 7, and the inference model is corrected. Then, the inference model correction unit 108 substitutes the verification data into the corrected inference model, And calculates the error rate for the state values of the equipment inferred by the system.

In this way, the inference model correcting unit 108 calculates the error rate for the state value of the corresponding facility estimated by the corrected inference model while continuously reducing the number of leaf nodes of the inference model. In this way, the inference model correcting unit 108 calculates the error rate of the inference model according to the correction of each step, and outputs the corrected inference model in the case where the error rate with respect to the state value of the corresponding facility becomes the minimum, . At this time, the error rate can be minimized by inferring the state value of the facility at the present time with the finally corrected reasoning model.

5 is a diagram illustrating an inference model finally corrected by the inference model correcting unit according to an embodiment of the present invention. Here, when the number of leaf nodes of the inference model generated by the inference model generation unit 106 is reduced to three, the error rate is minimized and the inference model is finally corrected.

Referring to FIG. 5, the finally corrected reasoning model includes a first leaf node LN1, a fourteenth leaf node LN14, and a fifteenth leaf node LN15. The state value of the first leaf node LN1 is 0, and the total number of the learning data belonging to the first leaf node LN1 is 50. [ At this time, since the actual state value of all learning data belonging to the first leaf node LN1 is 0, the first leaf node LN1 does not have an error rate.

The status value of the 14th leaf node LN14 is 1 and the total number of learning data belonging to the 14th leaf node LN14 is 54. [ At this time, the actual state value of 49 learning data among the learning data belonging to the 14th leaf node LN14 is 1, and the actual state value of the remaining 5 learning data is 2. Therefore, in the case of the fourteenth leaf node LN14, it has an error rate of 5/54.

The state value of the fifteenth leaf node LN15 is 2, and the total number of pieces of learning data belonging to the fifteenth leaf node LN15 is 46. At this time, the actual state value of 45 learning data among the learning data belonging to the 15th leaf node LN15 is 2, and the actual state value of the remaining one learning data is 1. Therefore, in the case of the fifteenth leaf node LN15, the error rate is 1/46.

That is, the 14th leaf node LN14 and the 15th leaf node LN15 are leaf nodes formed by merging leaf nodes, and the state values of the 14th leaf node LN14 and the 15th leaf node LN15 are merged The fourteen leaf node LN14 and the fifteenth leaf node LN15 include a certain error rate since they are state values of a leaf node to which a larger number of learning data belongs than the previous two leaf nodes.

Here, when the error rates of the fourteenth leaf node LN14 and the fifteenth leaf node LN15 exceed the preset reference value, the fourteenth leaf node LN14 and the fifteenth leaf node LN15, A data table storing at least one of data, status values according to the learning data, and acquisition time information of the learning data. In other words, the inference model correcting unit 108 calculates the error rate of each node of the corrected inference model, and when the calculated error rate exceeds the predetermined reference value, the inference model correcting unit 108 updates the data table including the data belonging to the relevant node, And stores them in association with each other. With this configuration, the state inference unit 108, which will be described later, can accurately deduce the state value of the facility using the inference model and the associated data table.

However, the number of leaf nodes of the inference model may be calculated in various other ways than the number of leaf nodes of the inference model. The error rate can be calculated.

The state inferring unit 110 substitutes the data collected by the data collecting unit 104 into the inference model finally corrected by the inference model correcting unit 108 and deduces the state value of the corresponding equipment according to the collected data. In this case, the status of the facility can be diagnosed at the present time through the status value of the facility.

For example, when the data collected by the data collecting unit 104 is the sensing values s1, s2, s3 and s4, the state inferring unit 110 determines that the sensing value s3 satisfies the conditional expression 1 (s3 <2.45) (That is, corresponding to the first leaf node LN1), it is inferred that the state value of the corresponding equipment according to the collected data is 0 (that is, normal). In this case, it is possible to diagnose that the current state of the facility is normal.

If the sensing value s3 satisfies the conditional expression (2) (s4 < 1.75) (i.e., the sensing value s3 does not satisfy the conditional expression (3) ), It is inferred that the state value of the equipment according to the collected data is 1 (that is, an abnormal symptom). In this case, it can be diagnosed that the current state of the facility is an abnormal symptom.

If the sensing value s3 does not satisfy the conditional expression 1 (s3 <2.45) and the sensing value s4 does not satisfy the conditional expression (2) (i.e., the 15th leaf node LN15) , It is inferred that the state value of the corresponding equipment according to the collected data is 2 (that is, failed). In this case, it can be diagnosed that the current state of the facility is faulty.

In this way, the state inferring unit 110 outputs the state value of the leaf node to which the data collected by the data collecting unit 104 among the leaf nodes of the finally corrected inference model belongs, to the state of the data collected by the data collecting unit 104 It can be inferred as a value.

If the error rates of the fourteenth leaf node LN14 and the fifteenth leaf node LN15 exceed the preset reference value, the state inferring unit 110 determines that the data collected by the current data collecting unit 104, The state value of the corresponding facility is not directly inferred as the state value of the fourteenth leaf node LN14 or the fifteenth leaf node LN15 even if it corresponds to the node LN14 or the fifteenth leaf node LN15, The state value of the facility can be inferred by calculating the similarities between the data collected by the leaf node 104 and the learning data stored in association with the corresponding leaf node. At this time, the state inferring unit 110 can confirm the learning data most similar to the data collected by the current data collecting unit 104, and finally infer the state value according to the learning data as the state value of the corresponding facility. In this case, the accuracy of the state value deduced by the corrected reasoning model can be improved.

For example, when the 14th leaf node LN14 or the 15th leaf node LN15 is associated with the learning data belonging to the leaf node and the data table storing the state value according to the learning data, the state inference unit 110 Calculates the degree of similarity between the data collected by the current data collection unit 104 and the learning data in the corresponding data table, and deduces the state value of the corresponding facility.

Here, the similarity between the data collected by the current data collection unit 104 and the learning data belonging to the corresponding leaf node can be obtained by measuring the Euclidean distance between the data, but not limited thereto, Various other similarity measurement methods can be used.

6 is a diagram for explaining a method of diagnosing a state of a facility when an error rate of a leaf node of a finally corrected reasoning model according to an embodiment of the present invention exceeds a preset reference value.

6, if the sensed values s1, s2, s3, s4 collected from the data collection unit 104 are (5.9, 3, 5.1, 1.8), substituting them into the finally corrected inference model When the sensing value s3 does not satisfy the conditional expression 1 (s3 <2.45) and the sensing value s4 does not satisfy the conditional expression 2 (s4 <1.75) (that is, corresponding to the fifteenth leaf node LN15) (110) deduces that the state value of the facility is two.

When the error rate of the fifteenth leaf node LN15 exceeds a preset reference value, the state inferring unit 110 updates the current sensing values s1, s2, s3, and s4 from the data collecting unit 104, (6.3, 2.7, 4.9), which is the most similar to the currently collected sensing values (5.9, 3, 5.1, 1.8), is obtained by calculating the similarities between the learning data belonging to the leaf node LN15 , 1.8) can be inferred as the state value of the corresponding facility.

When the error rates of the fourteenth leaf node LN14 and the fifteenth leaf node LN15 exceed the preset reference value, the state inferring unit 110 determines the time series of the data collected by the current data collecting unit 104 And the similarity between the time-series patterns of the learning data belonging to the corresponding leaf node, respectively, and deduce the state value of the corresponding facility. At this time, after confirming the learning data having the time-series thermal pattern most similar to the time-series pattern of the data collected by the current data collection unit 104, the state value according to the learning data can be finally inferred as the state value of the facility .

In the above example, the fourteenth leaf node LN14 or the fifteenth leaf node LN15 is a data table for storing learning data belonging to the leaf node, state values according to the learning data, and collection time information of the learning data The state inferring unit 110 calculates the similarities between the time series patterns of the data collected by the current data collection unit 104 and the time series patterns of the learning data in the corresponding data tables, .

FIG. 7 is a diagram for explaining a method of diagnosing a state of a facility when an error rate of a leaf node of a finally corrected reasoning model according to another embodiment of the present invention exceeds a preset reference value. Here, for convenience of description, only the time-series patterns of the two learning data belonging to the leaf node are shown.

Referring to FIG. 7A, the time 160 indicates the current time, and the sensing values (5.9, 3, 5.1, 1.8) collected at the time 160 are the data collected from the current data collection unit 104 to be. Here, the data collected up to the past time point (for example, time 151) based on the current time point (i.e., the time point 160) is referred to as a time-series pattern of the currently collected data. Here, the time-series pattern of the currently collected data includes a total of 10 data, but the present invention is not limited thereto.

Next, referring to FIG. 7 (b), a time series pattern of the first learning data belonging to the fifteenth leaf node LN15 is shown. Here, the learning data at time 110 is the first learning data belonging to the fifteenth leaf node LN15, and the actual state value according to the first learning data (7.2, 3.6, 6.1, 2.5) is two. At this time, the learning data collected up to the past time point (for example, time 101) based on the time 110 form a time-series pattern of the first learning data.

Next, referring to FIG. 7C, a time-series pattern of the second learning data belonging to the fifteenth leaf node LN15 is shown. Here, the learning data at time 60 is the second learning data belonging to the fifteenth leaf node LN15, and the actual state value according to the second learning data (5.2, 2.7, 3.9, 1.4) is one. At this time, the learning data collected up to the past predetermined time (for example, time 51) based on the time 60 form a time-series pattern of the second learning data.

Here, the state inferring unit 110 calculates the degree of similarity between the time-series pattern of the currently collected data and the time-series pattern of the first learning data and the second learning data, respectively, The state value of the learning data having the pattern can be finally inferred as the state value of the corresponding facility.

For example, if the similarity between the time-series pattern of the currently collected data and the time-series pattern of the first learning data is 0.0548245614035088 and the similarity between the time-series pattern of the currently collected data and the time-series pattern of the second learning data is 0.0328299409061064 , The state inferring unit 110 finally deduces the state value (i.e., 2) of the first learning data having a time-series thermal pattern most similar to the time-series pattern of the currently collected data to the state value of the corresponding facility.

The rule extracting unit 112 extracts a rule from the root node of the finally corrected reasoning model to each leaf node. In the embodiment of the present invention, a rule means text data obtained by converting the corrected reasoning model into a conditional form. For example, the rule extracting unit 112 extracts a conditional expression from the root node LN to the first leaf node LN1, the fourteenth leaf node LN14, and the fifteenth leaf node LN15 step by step The rules can be extracted. Hereinafter, the rules from the root node LN to the first leaf node LN1, the fourteenth leaf node LN14, and the fifteenth leaf node LN15 will be described.

Rule

IF (s3 < 2.45) Class 0

ELSE THEN

IF (s4 < 1.75) Class 1

ELSE Class 2

END

Here, the rule extraction unit 112 can store the extracted rule in the database 102. [ The inference model exists in the form of an array or the like on the memory, so that the capacity is relatively large and there is a problem that it must always reside in the memory for judgment. However, when the reasoning model is structured as a text type rule, the capacity of the reasoning model is much smaller than that of the reasoning model, and the text data can be easily stored in the database 102 or the like because the processing on the computer is also simple . Also, if necessary, the extracted rolls may be used later to reconstruct the inference model.

8 is a flowchart illustrating a state diagnostic method according to an embodiment of the present invention.

Referring to FIG. 8, the inference model generation unit 106 generates an inference model based on data collected during a predetermined period of time in the database 104 and state values according to the collected data (S 101). At this time, the inference model generation unit 106 may classify the collected data for a certain period of time into learning data and verification data, and then generate an inference model based on the state values of the learning data and the learning data. As described above, the reasoning model may be configured, for example, in the form of a decryption tree, but the present invention is not limited thereto.

Next, the inference model correcting unit 108 corrects the inference model by changing the number of leaf nodes of the inference model (S 103). For example, the inference model correction unit 108 may reduce the number of leaf nodes of the inference model one by one, substitute the inference data into the inference model, and then calculate the error rate with respect to the state value of the corresponding inference apparatus estimated by the inference model The inference model can be finally corrected. At this time, the inference model correcting unit 108 can finally determine the inference model when the error rate according to the change in the number of leaf nodes becomes the minimum, as the corrected inference model. In addition, as described above, when there is a leaf node whose error rate exceeds the preset reference value among the leaf nodes of the corrected inference model, the inference model correcting unit 108 corrects the learning data corresponding to the corresponding node, The state value of the data, and the collection time information of the learning data in association with the corresponding node.

Next, the state inferring unit 110 substitutes the data collected by the data collecting unit 102 for the finally corrected inference model and deduces a state value according to the data (S 105). At this time, the state inferring unit 110 calculates a state value of the leaf node to which the data collected by the data collecting unit 102 among the leaf nodes of the finally corrected inference model belongs, as the state value of the data collected by the data collecting unit 102 .

On the other hand, when the error rate of the leaf node to which the data collected by the data collection unit 102 belongs exceeds the preset reference value, the similarity between the data collected by the data collection unit 104 and the learning data stored in association with the leaf node And can finally infer the state value of the learning data most similar to the data collected by the data collection unit 104 to a state value according to the data collected by the data collection unit 104 according to the calculation result.

If the error rate of the leaf node to which the data collected by the data collecting unit 102 belongs exceeds the preset reference value, the state inferring unit 110 calculates the time-series pattern of the data collected by the data collecting unit 104, The degree of similarity between the time series patterns of the learning data stored in association with the leaf nodes is calculated and the state of the learning data having the time series thermal pattern most similar to the time series pattern of the data collected by the data collection unit 104, Can be inferred as a state value according to the data collected by the data collection unit 104. [

On the other hand, an embodiment of the present invention may include a computer-readable recording medium including a program for performing the methods described herein on a computer. The computer-readable recording medium may include a program command, a local data file, a local data structure, or the like, alone or in combination. The media may be those specially designed and constructed for the present invention or may be known and available to those of ordinary skill in the computer software arts. Examples of computer-readable media include magnetic media such as hard disks, floppy disks and magnetic tape, optical recording media such as CD-ROMs and DVDs, magneto-optical media such as floppy disks, and magnetic media such as ROMs, And hardware devices specifically configured to store and execute program instructions. Examples of program instructions may include machine language code such as those generated by a compiler, as well as high-level language code that may be executed by a computer using an interpreter or the like.

While the present invention has been particularly shown and described with reference to exemplary embodiments thereof, it is clearly understood that the same is by way of illustration and example only and is not to be taken by way of limitation, I will understand. Therefore, the scope of the present invention should not be limited to the above-described embodiments, but should be determined by equivalents to the appended claims, as well as the appended claims.

100: status diagnosis system 102: database
104: Data collection unit 106: Reasoning model generation unit
108: inference model correction unit 110: state inference unit
112:

Claims

An inference model generating unit that generates an inference model including nodes having at least one conditional expression using data collected over a predetermined period and state values according to the collected data;
A reasoning model correcting unit for correcting the reasoning model so that the accuracy of the reasoning model is improved; And
And a state inference unit for substituting input data into a corrected inference model to deduce a state value according to the input data.

The method according to claim 1,
The inference model generation unit may include:
Classifying the collected data into learning data and verification data, and generating the inference model using the learning data and the state values of the learning data.

3. The method of claim 2,
The reasoning model,
State diagnosis system configured in the form of a decision tree.

The method of claim 3,
The inference model correcting unit corrects,
And calculates the error rate of the inference model according to the number of leaf nodes of the inference model using the verification data and determines the number of leaf nodes of the inference model so that the calculated error rate is minimized.

5. The method of claim 4,
The inference model correcting unit corrects,
And stores the learning data corresponding to the corresponding leaf node in association with the corresponding node when the error rate among the leaf nodes of the corrected inference model exceeds a preset reference value.

6. The method of claim 5,
The state inferring unit,
And the inference unit deduces the state value of the leaf node to which the input data belongs among the leaf nodes of the corrected reasoning model as the state value of the input data.

The method according to claim 6,
The state inferring unit,
Calculating the degree of similarity between the learning data stored in association with the leaf node and the input data when the error rate of the leaf node to which the input data belongs exceeds a preset reference value, The state value of the learning data most similar to the input data is inferred as a state value according to the input data.

The method according to claim 6,
The state inferring unit,
Calculating similarities between the time series patterns of the learning data stored in association with the leaf node and the time series patterns of the input data when the error rate of the leaf node to which the input data belongs exceeds a predetermined reference value, And deduces the state value of the learning data having a time-series thermal pattern most similar to the time-series thermal pattern of the input data as a state value according to the input data according to the calculation result.

The method according to claim 1,
The condition diagnosis system includes:
And a rule extracting unit for extracting a rule from a root node of the corrected reasoning model to each leaf node.

Generating an inference model including nodes having at least one conditional expression using data collected during a predetermined period and state values according to the collected data;
Correcting the inference model so that the inference model correcting unit improves the accuracy of the inference model; And
And inferring a state value according to the input data by substituting the input data for the state inference unit into the corrected inference model.

11. The method of claim 10,
Wherein the generating the inference model comprises:
Classifying the collected data into learning data and verification data; And
And generating the inference model using the state value of the learning data and the learning data.

12. The method of claim 11,
The reasoning model,
Wherein the diagnosis is made in the form of a decision tree.

12. The method of claim 11,
Wherein the step of correcting the inference model comprises:
Calculating an error rate of the inference model according to a change in the number of leaf nodes of the inference model using the verification data; And
And determining the number of leaf nodes of the inference model so that the calculated error rate is minimized.

14. The method of claim 13,
Wherein the step of correcting the inference model comprises:
Further comprising the step of storing the learning data corresponding to the node in association with the corresponding node when the error rate among the leaf nodes of the corrected reasoning model exceeds a preset reference value .

15. The method of claim 14,
The step of inferring the state value comprises:
Wherein the state inference unit deduces the state value of the leaf node to which the input data belongs among the leaf nodes of the corrected inference model as the state value of the input data.

16. The method of claim 15,
The step of inferring the state value comprises:
Determining whether an error rate of a leaf node to which the input data belongs exceeds a preset reference value;
Calculating a degree of similarity between the learning data stored in association with the leaf node and the input data if the error rate of the leaf node exceeds a predetermined reference value; And
And inferring the state value of the learning data most similar to the input data to a state value according to the input data according to the calculation result.

16. The method of claim 15,
The step of inferring the state value comprises:
Determining whether an error rate of a leaf node to which the input data belongs exceeds a preset reference value;
Calculating similarities between the time series patterns of the learning data stored in association with the leaf nodes and the time series patterns of the input data when the error rate of the leaf nodes exceeds a predetermined reference value; And
And infering the state value of the learning data having a time-series thermal pattern most similar to the time-series thermal pattern of the input data to a state value according to the input data according to the calculation result.

11. The method of claim 10,
After correcting the inference model,
Further comprising the step of the rule extracting section extracting a rule from the root node of the corrected reasoning model to each leaf node.

A computer-readable recording medium recording a program for executing the state diagnostic method according to any one of claims 10 to 18.