KR20190001781A - Method for detecting abnormal situation and system for performing the same - Google Patents

Method for detecting abnormal situation and system for performing the same Download PDF

Info

Publication number
KR20190001781A
KR20190001781A KR1020170081846A KR20170081846A KR20190001781A KR 20190001781 A KR20190001781 A KR 20190001781A KR 1020170081846 A KR1020170081846 A KR 1020170081846A KR 20170081846 A KR20170081846 A KR 20170081846A KR 20190001781 A KR20190001781 A KR 20190001781A
Authority
KR
South Korea
Prior art keywords
abnormal
sensor data
abnormal situation
situation
data
Prior art date
Application number
KR1020170081846A
Other languages
Korean (ko)
Other versions
KR101970619B1 (en
Inventor
김익재
조정현
최희승
남기표
강기헌
권태송
Original Assignee
한국과학기술연구원
주식회사 휴먼아이씨티
Priority date (The priority date is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the date listed.)
Filing date
Publication date
Application filed by 한국과학기술연구원, 주식회사 휴먼아이씨티 filed Critical 한국과학기술연구원
Priority to KR1020170081846A priority Critical patent/KR101970619B1/en
Publication of KR20190001781A publication Critical patent/KR20190001781A/en
Application granted granted Critical
Publication of KR101970619B1 publication Critical patent/KR101970619B1/en

Links

Images

Classifications

    • G06N99/005
    • GPHYSICS
    • G06COMPUTING; CALCULATING; COUNTING
    • G06NCOMPUTER SYSTEMS BASED ON SPECIFIC COMPUTATIONAL MODELS
    • G06N20/00Machine learning
    • GPHYSICS
    • G06COMPUTING; CALCULATING; COUNTING
    • G06QDATA PROCESSING SYSTEMS OR METHODS, SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL, SUPERVISORY OR FORECASTING PURPOSES; SYSTEMS OR METHODS SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL, SUPERVISORY OR FORECASTING PURPOSES, NOT OTHERWISE PROVIDED FOR
    • G06Q50/00Systems or methods specially adapted for specific business sectors, e.g. utilities or tourism
    • G06Q50/10Services
    • GPHYSICS
    • G08SIGNALLING
    • G08BSIGNALLING OR CALLING SYSTEMS; ORDER TELEGRAPHS; ALARM SYSTEMS
    • G08B21/00Alarms responsive to a single specified undesired or abnormal operating condition and not elsewhere provided for
    • G08B21/18Status alarms
    • G08B21/187Machine fault alarms

Abstract

A system for detecting an abnormal situation comprises: a first sensing unit obtaining first sensor data from facility equipment; a first situation determination unit determining whether an abnormal situation occurs from the obtained first sensor data; a data for learning generation unit generating learning data from the obtained sensor data if it is determined that the abnormal situation has occurred; and a parameter for abnormal situation recognition learning unit updating a parameter for recognizing an abnormal situation through machine learning by using the data for learning. The first situation determination unit determines occurrence of next abnormal situation based on the parameter for abnormal situation recognition. The data for learning includes the sensor data obtained for a predetermined time range from the occurrence time of the abnormal situation.

Description

TECHNICAL FIELD [0001] The present invention relates to a method for detecting abnormal conditions and a system for performing the same,

The present invention relates generally to a method for detecting an abnormal situation and a system for performing the same, and more particularly relates to detecting and detecting an abnormal situation and detecting the occurrence of an abnormal situation in the future.

In designing processes to be performed by facilities installed in power plants or factories, the designer must designate a series of process steps taking into account the specific situation (hereinafter referred to as "abnormal situation") affecting the efficiency of the equipment or the operation result . However, due to reasons such as unexpected reasons that designers did not consider, or unexpected situations that the designers did not consider to occur even though designers did their best to prevent abnormal situations in the process of performing the process design, An abnormal situation may occur.

To cope with such an abnormal situation, an apparatus for detecting the occurrence of an abnormal situation can be used. However, when a new abnormal situation occurs, there is a limit in that the user can directly input the data.

In order to solve such a disadvantage, there is an attempt to detect the occurrence of an abnormal situation by applying a machine learning algorithm. 1 is a diagram of a system for detecting an abnormal situation by applying a machine learning algorithm.

1, the detected sensing unit start time (t start) from the end time (t end) sensor data from up equipment device of the process (e.g., video, audio, smoke, temperature, humidity, pressure, etc.) . When the situation determination unit detects the occurrence of an abnormal situation, it applies all of the sensor data up to before the occurrence to the machine learning algorithm and improves the ability to detect the occurrence of an abnormal situation based on the result. As shown in FIG. 1, an abnormal situation may occur at the time of midway (t b1 , t b2 ) at which the sensed sensing unit acquires the sensor data from the process start time (t start ) to the end time (t end ). In this case, in order to perform the machine learning based on the sensor data, the detection result data having the range (t start to t b1 ) and the detection result data having the range (t b1 to t b2 ) must be applied to the machine learning algorithm.

However, in general, the incidence of abnormal conditions is relatively less than the incidence of normal conditions. Therefore, if basic data sufficient for deriving a meaningful learning result is secured, unnecessary data (i.e., data relating to a normal situation) among the secured data is a large proportion. Therefore, there is a problem that the sensor data capacity applied to the machine learning is consumed relatively more than the actual required capacity, and the analysis and processing are also time-consuming.

In addition, judging the occurrence of an abnormal situation means that an abnormal situation has already occurred. Therefore, there is a problem that it is not possible to prevent the occurrence of a process stoppage in the process or a failure in the equipment, despite detecting an abnormal situation.

KR 10-2017-0022625 A

There is provided a method of detecting an abnormal situation using data substantially required for improving the detection capability of an abnormal situation and a system for performing the same.

An abnormal situation detection method according to an embodiment includes: obtaining first sensor data from a facility apparatus; Determining whether an abnormal condition has occurred from the obtained first sensor data; Generating training data from the obtained sensor data if it is determined that an abnormal situation has occurred; Updating parameters for detecting an abnormal situation through machine learning using the learning data; And determining the occurrence of the next abnormal situation based on the abnormal condition recognition parameter, wherein the learning data may include sensor data obtained for a predetermined time range from the occurrence time of the abnormal condition.

The time range may be determined based on the characteristics of the sensor data acquired before the occurrence time of the abnormal condition.

The method comprising: obtaining second sensor data of a different type than the obtained sensor data; And determining whether an abnormal condition has occurred from the obtained second sensor data.

The method may further include generating an alert when it is determined that an abnormal situation will occur from the obtained sensor data before the occurrence of an abnormal situation is determined from the second sensor data.

The abnormal condition recognition parameter may include at least one of support vector machine (SVM), support vector data description (SVDD), support vector regression (SVR), artificial neural network (ANN), and K-means clustering algorithm Lt; / RTI >

The abnormal condition recognition parameter may be updated from a plurality of learning data accumulated for a predetermined time duration.

The abnormal condition recognition parameter may be updated in real time from the generated learning data.

And calculating the probability of occurrence of the next abnormal situation by applying the sensor data obtained after the occurrence of the abnormal condition to the probability model.

A computer program for performing the abnormal situation detection method may be recorded in the computer-readable recording medium according to the embodiments.

An abnormal situation detection system according to an embodiment includes a first sensing unit for acquiring first sensor data from a facility apparatus; A first situation determiner for determining whether an abnormal situation occurs from the obtained first sensor data; A learning data generation unit for generating learning data from the obtained sensor data if it is determined that an abnormal situation has occurred; And an abnormal condition recognition parameter learning unit for updating an abnormal condition recognition parameter through machine learning using the learning data, wherein the first situation determination unit judges occurrence of the next abnormal situation based on the abnormal condition recognition parameter And the learning data may include sensor data acquired for a certain period of time from the occurrence time of the abnormal situation.

 The time range may be determined based on the characteristics of the sensor data acquired before the occurrence time of the abnormal condition.

The system comprising: a second sensing unit for acquiring second sensor data of a different type from the first sensor data; And a second status determiner for determining an abnormal status from the second sensor data.

The system may further include a warning unit for generating a warning when the first situation determining unit determines that the abnormal situation will occur before the second situation determining unit determines that the abnormal condition occurs.

The system may further include a possibility calculation unit for calculating an abnormality occurrence probability by applying the obtained sensor data to a probability model.

The abnormal condition recognition parameter may include at least one of support vector machine (SVM), support vector data description (SVDD), support vector regression (SVR), artificial neural network (ANN), and K-means clustering algorithm Lt; / RTI >

According to an aspect of the present invention, there is provided a method for detecting an abnormal situation and a system for performing the same, the method comprising: updating an abnormal situation recognition parameter by using a certain range of learning data every time an abnormal situation occurs; It is possible to improve the abnormal situation judgment ability.

In addition, since only a certain range of data is used from the time when an abnormal situation occurs, which is substantially required in the process of learning the abnormal situation, it is possible to learn the abnormal situation more rapidly at a smaller capacity.

Further, it is possible to catch the sign of the abnormal situation and warn in advance of the occurrence of the abnormal situation. This can prevent breakdowns and failures of the in-service process and reduce damage and repair costs incurred until the suspended process is restarted.

1 is a schematic view showing a conceptual diagram of a system for learning a conventional abnormal situation.
2 is a conceptual block diagram of an abnormal situation detection system according to an embodiment.
3 is a view schematically showing learning data used for updating an abnormal situation recognition parameter in an abnormal situation detection system according to an embodiment.
4 is a conceptual block diagram of an abnormal situation detection system that detects an abnormal situation in advance and generates a warning according to an embodiment.
5 is a flowchart of an abnormal situation detection method according to an embodiment.
6 is a flowchart of an abnormal situation detection method for detecting an abnormal situation in advance and generating an alarm according to an embodiment.
The figures depict various embodiments of the invention only for purposes of illustration. Those skilled in the art will readily appreciate from the following description that alternative embodiments of the structures and methods described herein may be used without departing from the principles of the invention described herein.

As used herein, an "abnormal situation" refers to a set of circumstances that affects a planned process result due to any condition, such as a failure of a plant, an outage, or the like.

As used herein, " data on an abnormal situation " includes, for example, information in which a user inputs a specific situation as an abnormal situation, time information in which an abnormal situation occurs, or numerical information that reaches a threshold for judging an abnormal situation Quot; refers to data associated with the occurrence of an abnormal situation. &Quot; Data on normal situation " refers to data not related to the occurrence of an abnormal situation.

Hereinafter, embodiments of the present invention will be described in detail with reference to the drawings.

2 is a block conceptual diagram of an abnormal situation detection system 100 according to an embodiment.

In one embodiment, the system 100 includes a sensing unit 110, a situation determination unit 130, a training data generation unit 150, and a parameter learning unit (hereinafter, " learning unit " . It will be apparent to those skilled in the art that the system 100 may include other components not described herein. The system 100 also includes other hardware elements necessary for the operations described herein, including network interfaces and protocols, input devices for data entries, and output devices for display, printing, or other data display do.

In one embodiment, the sensing unit 110 senses the facility device or equipment in real time to acquire sensor data. The sensor data may be image, voice, heat, temperature, humidity, pressure data, or a combination thereof depending on the type of sensing part. The sensor data can be used to determine the occurrence of an abnormal situation by comparing and analyzing the abnormal condition recognition parameter.

More specifically, the sensor data may be referred to as a feature vector f. The feature vector represents a feature set comprising a plurality of feature values. The characteristic value means data generated from a state change of the facility while the sensing unit 110 senses the facility. Any given feature value may be a numerical value such as an integer, floating point, or binary value, or it may be categorical. The common feature includes the sensor data value and time of the sensing unit. For example, when the sensing unit 110 includes a temperature sensor, the feature vector includes a time and a temperature K at the corresponding time.

The sensing unit 110 may include at least one of a gravity sensor, a geomagnetism sensor, a motion sensor, a gyro sensor, an acceleration sensor, a tilt sensor, a brightness sensor, an altitude sensor, an olfactory sensor, a temperature sensor, a depth sensor, , An image sensor, a GPS sensor, a touch sensor velocity sensor, and the like, or a combination thereof.

In one embodiment, the system 100 may further include a storage unit 120 for storing acquired sensor data. The storage 120 is implemented using non-volatile computer readable storage devices and an appropriate database management system for data access and retrieval.

The storage unit 120 stores the sensor data acquired by the sensing unit 110. [ In one embodiment, the storage unit 120 may temporarily store the sensor data sensed by the sensing unit 110 until generating learning data. In the above embodiment, the stored sensor data may be deleted when used for generation of learning data. Or if the abnormal state is not generated for a long time and the capacity of the storage unit 120 is saturated due to the data related to the normal situation, the storage unit 120 deletes a part of the sensor data stored starting from the sensor data having the oldest storage time It is possible.

The storage unit 120 configured with non-volatile computer-readable storage may be configured to include at least one of volatile memory (not shown), non-volatile memory (not shown), and external memory (not shown). For example, volatile memory 310 may be a dynamic random access memory such as read only memory (RAM) or synchronous dynamic read-only memory (SDRAM), double data rate (DDR) SDRAM, or Rambus DRAM The nonvolatile memory may be any one of a read only memory (ROM) and a flash memory. The external memory may be an MMC type, an SD type, and a CF type memory Card. In one embodiment, the storage unit 120 may include volatile memory and non-volatile memory. In this case, the volatile memory (for example, a RAM) may lose all stored data when a reset is performed due to power-off or the like, while a nonvolatile memory (e.g., ROM) Once saved data is not lost. If the storage unit 120 includes different types of memory, efficient data storage control may be possible.

It will be apparent to those skilled in the art that, in some embodiments, the storage 120 is not a component of a generic database, and may include other data stores not explicitly mentioned herein. But may be implemented using any suitable database management system, such as Microsoft SQL Server, Oracle, SAP, IBM DB2, and so forth.

The situation determination unit 130 determines occurrence of an abnormal situation. In one embodiment, the situation determination unit 130 compares and analyzes the obtained sensor data with a preset threshold value to determine occurrence of an abnormal situation.

The threshold value may be updated. As described below, when the parameter for abnormality recognition is (re) updated in the learning unit, the threshold is (re) updated with the parameter for the updated abnormality condition. In one embodiment, the situation determination unit 130 compares and analyzes the obtained sensor data with the updated threshold value to determine whether an abnormal situation occurs. Parameters for abnormal situation recognition are described in detail in the learning section 170 below.

Further, the threshold value may be changed by data input regarding an abnormal situation by the user. In some embodiments, the situation determination unit 130 may determine that an abnormal situation has occurred when the operation of the facility apparatus is stopped even if the sensor data value is less than the threshold value.

In one embodiment, when the situation determination unit 130 acquires sensor data to determine occurrence of an abnormal situation, the occurrence time (t b1 , t b2 , ..., t bn ) of the abnormal condition is also determined. The situation determination unit 130 provides the learning data generation unit 150 with the occurrence times (t b1 , t b2 , ..., t bn ) of the determined abnormal situations.

The learning data generation unit 150 generates training data from the obtained sensor data. The learning data is used by the learning unit 170 to generate parameters for detecting an abnormal situation. The learning data generation unit 150 provides the generated learning data to the learning unit 170. [

In one embodiment, when the situation determination unit 130 determines that an abnormal situation has occurred, the learning data generation unit 150 generates the abnormality occurrence time tb1 , tb2 , ..., tbn and the occurrence of an abnormal situation Sensor data between the previous arbitrary times t a1 , t a2 , ..., t an is generated as learning data (t an <t bn ). In one embodiment, any time (t a1 , t a2 , ..., t an ) prior to the occurrence of the abnormal situation may be a predetermined time. For example, at time (t b1) when the occurrence of abnormal situation is determined from, the training data producing unit 150 is time of 5 seconds time range to determine the five seconds before the time as t a1 from the (t b1) (t a1 to &lt; RTI ID = 0.0 &gt; tl). &lt; / RTI &gt; Alternatively, it is also possible to generate the training data with the time (t b1) one week prior to determining a time t a1 in a week time span (t a1 ~ t b1) from.

3 is a view schematically showing learning data used for generating an abnormal situation recognition parameter in the abnormal situation detection system 100 according to an embodiment.

In one embodiment, the time range may be determined based on the characteristics of the sensor data prior to the occurrence of the abnormal condition. Since the abnormal situation occurrence times t b1 , t b2 , ..., t bn are constants, the time range of the training data is also determined when the time t a1 , t a2 , ..., t an is determined. The time t a1 , t a2 , ..., t an may be a characteristic of the sensor data acquired by the sensing unit 110, for example, the type of sensor data, the sensor data value, the average and / , An estimated capacity of the learning data to be generated, and the like.

Referring to FIG. 3, for example, when the sensor data acquired by the sensing unit 110 is temperature data and time tb1 , tb2 (T a1 ) at the time when the amount of change in the temperature K continuously increases when an abnormal situation occurs in the temperature (K). Alternatively, an arbitrary time with the largest change in the average temperature K among the times before the time t b1 may be determined as the time t a1 .

In one embodiment, the learning data generation unit 150 may generate learning data from the sensor data stored in the storage unit 120. [

The learning unit 170 updates the parameter for detecting the abnormal situation from the learning data generated by the learning data generation unit 150. [ The abnormal condition recognition parameter is related to the threshold value of the situation determination unit 130. [ When the parameter for abnormal situation recognition is updated, the threshold value is also updated. The learning unit 170 may provide the abnormal condition recognition parameter to other components in the system 100. [ In one embodiment, the learning unit 170 may provide the situation determination unit 130 with a parameter for detecting an abnormal situation.

In one embodiment, the parameters for anomaly detection are support vector machine (SVM), support vector data description (SVDD), support vector regression (SVR), artificial neural network (ANN), and K- Algorithm. &Lt; / RTI &gt;

In one example, the abnormal condition recognition parameter may be generated by applying learning data to a support vector machine (SVM) algorithm. The SVM algorithm is one of the fields of machine learning. It is a map learning model for pattern recognition and data analysis. It is mainly used for classification and regression analysis. For example, given a set of data belonging to one of two categories, the SVM algorithm can generate a non-stochastic binary linear classification model that determines, based on a given set of data, have. In searching for a linear hyperplane that classifies given data into two classes, it is performed by finding a hyperplane that is farthest from the data.

More specifically, the SVM algorithm is characterized in that it looks for a hyperplane with the largest margin among the many candidate planes that can separate data. Here, the margin refers to the minimum value of the distance from the hyperplane to each point (data). In order to classify each point into two classes (1, -1) while maximizing such a margin, the SVM algorithm sets the minimum value of the distance between the points belonging to class 1 and the points belonging to class- . This hyperplane is called the maximum-margin hyperplane. Based on the maximum margin hyperplane, classification of the currently stored data into two classes and prediction of which of the two classes the newly entered data belongs to.

In the case of the above example, the SVM-based classifying process is repeatedly performed on the learning data generated at the time tb1 , tb2 , ..., tbn to classify a plurality of abnormal situations hierarchically into classes, The newly acquired sensor data (i.e., sensor data acquired after t bn ) determines which of the classified classes belongs. The learning unit 170 reflects the difference in distance with respect to the existing class (for example, based on classification error, loss, etc.) generated from the newly acquired sensor data in real time to update the parameter for detecting the abnormal situation do.

Parameters for abnormal condition recognition may differ depending on the machine learning algorithm as parameters associated with such classification and judgment, and are not limited to specific types of information. The abnormal condition recognition parameter is reflected in the threshold value of the situation determination unit 130 and is used as a new criterion for determining occurrence of an abnormal condition. Accordingly, when the abnormal condition recognition parameter is updated, the abnormal condition determination capability of the situation determination unit 130 is also updated.

In another example, the K average clustering algorithm is used in the class classification process, and the process of determining where the newly input data belongs to the classified class is applied to the class classification process as well as the regression analysis algorithm is used And the algorithms applied to the process of determining whether the newly input data belongs to the classified class may be different.

In some embodiments, the system 100 may further include a probability calculation unit that expresses likelihood of occurrence of a time-by-time anomaly from the training data as a probability (hereinafter, referred to as "probability Pe"). The probability calculator determines the probability of occurrence of the next abnormal situation by applying sensor data acquired at an arbitrary time (t f , t bn <t f <t bn + 1 ) after the time (t bn ) And calculates a risk probability Pe.

The probability model for calculating probability (Pe) may be modeled by various statistical methods. The statistical methods include non-splines, three-dimensional splines, arbitrary regression types, Dirac delta functions (for category and binary data), or any combination of these methods, . The probability calculation unit applies the learning data generated by the learning data generation unit 150 to at least one of the above statistical methods to model the fitted probability model. If an abnormal situation has never occurred and a probability model has not been generated, the probability calculator may calculate an arbitrary initial value as a probability P e (for example, 0).

When the feature vector f is applied to the fitted probability model, the feature values included in the feature vector f may be limited for the entire range. In some embodiments, the feature value may be normalized to limit the range to between zero and one.

The probability calculator may provide the calculated probability Pe to the user or other components. For example, letters, numbers, or combinations thereof may be displayed on the display, including probability (Pe) numbers.

4 is a block conceptual diagram of an abnormal situation detection system 100 that detects an abnormal situation in advance and generates a warning according to an embodiment. In this embodiment, the system 200 further includes a second sensing unit 210, a second situation determination unit 230, and a warning unit 240.

The second sensing unit 210 and the second situation determining unit 230 are one means for providing a criterion for determining an abnormal state intended by the designer. The second sensing unit 210 acquires sensor data in real time from the facility apparatus, such as the sensing unit 110. The type of sensor data (hereinafter referred to as &quot; second sensor data &quot;) acquired by the second sensing unit 210 is different from the type of sensor data acquired by the sensing unit 110. [

The second situation determination unit 230 determines whether or not an abnormal condition has occurred from the obtained second sensor data. The second situation determiner 230 compares and analyzes the obtained second sensor data with a threshold value, and determines that an abnormal condition has occurred when the second sensor data is equal to or greater than the threshold value. The second situation determination unit 230 can also determine the time at which the abnormal situation occurred. The situation determination unit 130 and the second situation determination unit 230 provide the determined abnormal condition occurrence time data to the warning unit 240. [

When the second determining unit 230 determines that the abnormal condition has occurred before the abnormal condition is determined, the warning unit 240 determines that the abnormal condition occurs . In one embodiment, when the situation determination unit 130 determines the abnormal condition, if the second condition determination unit 230 has not yet determined the occurrence of the abnormal condition, the warning unit 240 generates a warning. Means of providing a warning additional warning include various means, including voice, vibration, image, illumination, or any combination thereof.

5 is a flowchart of an abnormal situation detection method (S100) according to an embodiment.

The abnormal state detection method (S100) according to one embodiment may be performed in substantially the same configuration as the system 100 shown in FIG. Therefore, the same components as those of the system 100 shown in FIG. 2 are denoted by the same reference numerals, and repeated descriptions are omitted. In addition, the method for detecting an abnormal situation according to an embodiment may be implemented as a computer program including computer instructions for performing an abnormal situation detection.

The sensing unit 110 acquires sensor data from the equipment or equipment (S110). In one embodiment, the storage unit 120 may store the sensor data acquired by the sensing unit 110 (S120). The sensing unit 110 acquires sensor data in real time independently of the following steps S130 to S160.

In one embodiment, the storage unit 120 may store the acquired sensor data (S120).

The situation determination unit 130 processes the obtained sensor data in real time to determine occurrence of an abnormal situation (S130). In one embodiment, the situation determination unit 130 compares and analyzes the obtained real-time data with a threshold value in real time to determine whether an abnormal condition is issued (S130). If the obtained sensor data is equal to or greater than the threshold value, the situation determination unit 130 determines that an abnormal situation has occurred. If it is determined that an abnormal condition has occurred, the situation determination unit 130 determines the occurrence time of the abnormal condition (S135).

When the situation determination unit 130 determines that an abnormal situation has occurred, the learning data generation unit 150 generates training data based on the abnormal situation occurrence time (S150). For example, at time (t b1) an abnormal situation (hereinafter referred to as "a first abnormal condition"), the learning data having a training data generator 150 is in the range (t a1 ~ t b1) if occurred (hereinafter referred to in "The 1 learning data &quot;) (S150). In one embodiment, the time t a1 may be determined based on the sensor data obtained before the time t b1 . For example, when the amount of change in the acquired sensor data value continuously increases from a specific time (t, t <t b1 ), the specific time t is determined as time t a1 .

The learning unit 170 updates the parameter for detecting an abnormal situation from the first learning data (S170). The learning unit 170 may provide the situation determination unit 130 with a parameter for detecting an abnormal situation.

In one embodiment, the parameters for anomaly detection are support vector machine (SVM), support vector data description (SVDD), support vector regression (SVR), artificial neural network (ANN), and K- Algorithm (S170).

The learning unit 170 updates parameters for detecting an abnormal situation from learning data through various processing methods (S170). In one embodiment, the learning unit 170 may generate a parameter for detecting an abnormal situation in a batch learning manner. In this case, the abnormal situation learning unit may include m learning abnormal condition recognition parameters generated for a predetermined time interval, for example, first learning data t a1 to t b1 , second learning data t a2 to t b2 ), third learning data ( ta3 to tb3 ), ... , And the mth learning data (t am , t bm ) are collectively applied to an algorithm for generating a parameter for detecting an abnormal situation, thereby updating the parameter for detecting the abnormal situation.

In another embodiment, the learning unit 140 may update parameters for detecting an abnormal situation in an on-line learning manner. In this case, the learning unit generates an algorithm for generating the parameters for recognizing the abnormal situation in real time for each of the generated learning data (t a1 to t b1 ), (t a2 to t b2 ), ..., (t am , t bm ) To update the parameter for detecting an abnormal situation.

In some embodiments, the probability calculator applies the sensor data acquired after the time of occurrence of the abnormal condition ( tbl ) to the probability model to calculate the probability of occurrence of the next abnormal situation. When the first abnormal situation is issued at time (t b1), likelihood calculation unit at the time (t b2) a second until an abnormal situation occurs, an arbitrary time (t f, t b1 <t f <t b2) at And calculates the probability Pe. When the steps S150 to S170 are repeated, the learning data is regenerated, and therefore, the probability model is also remodeled.

Then, it is determined whether or not the next abnormal situation occurs according to the updated abnormal condition recognition parameter (S190). Thereafter, steps S130 to S170 are repeated based on the updated abnormality recognition parameter. More specifically, the status determination unit 130 updates the threshold value to the parameter for that the abnormal situation, and the time (t b1) after the comparison and analysis of the sensor data acquired by the sensing unit (110), time (t b1) Thereafter, the occurrence of an abnormal condition (second abnormal condition) is judged.

When the situation determination unit 130 determines that an abnormal situation (second abnormal situation) occurs at the time (t b2 , t b1 <t b2 ) through the updated threshold value, the learning data generation unit 150 sets the range t a2 to t b2 ) (S150). In one embodiment, the time t a2 may be determined based on the characteristics of the sensor data prior to time t b2 .

Learning unit 170 first and updates the parameters for that the abnormal situation from the second learning data (S170), the situation determination unit 130 is time (t b2) based on the updated threshold value based on a parameter for whether the abnormal situation Then, it is determined whether or not the next abnormal situation occurs at an arbitrary time (t f ) (S 130).

As described above, the system 100 repeatedly determines whether or not an abnormal situation occurs, generates training data repeatedly each time an abnormal situation occurs, and re-updates the parameter for detecting an abnormal situation based on the generated training data .

FIG. 6 is a flowchart of an abnormal situation detection method (S100) for detecting an abnormal situation in advance and generating an alarm according to an embodiment. The method S100 further includes a step S210 of obtaining the second sensor data by the second sensing unit 210 and a step of determining whether the second situation determination unit 230 generates an abnormal situation.

The system 200 acquires sensor data by the sensing unit 110 (S110), and acquires second sensor data of a different type by the second sensing unit 210 (S210).

The situation determination unit 130 determines occurrence of an abnormal situation from the sensor data (hereinafter, referred to as &quot; first sensor data &quot;) acquired by the sensing unit 110 (S130). The second situation determination unit 230 also determines the occurrence of an abnormal situation from the sensor data (hereinafter referred to as &quot; second sensor data &quot;) obtained by the second sensing unit 210 at step S230. In some embodiments, the situation determination unit 130 may compare and analyze both the first sensor data and the second sensor data to determine occurrence of an abnormal situation.

When the situation determination unit 130 determines occurrence of an abnormal situation from the first sensor data, the abnormal situation occurrence time t 1 is determined (S135). When the second situation determination unit 233 determines from the second sensor data When the occurrence of an abnormal situation is determined, the occurrence time t 2 of the abnormal situation is determined. The time t 1 determined by the situation determiner 130 may be the same as the time t 2 determined by the second situation determiner 230 before the first abnormal situation occurs.

Thereafter, when the time t1 determined by the situation determination unit 130 is provided to the learning data generation unit 150 every time an abnormal situation occurs, the steps S150 through S190 shown in FIG. 5 are performed. As a result, t 1 and t 2 may be different from each other.

In one embodiment, the method (SlOO) may further comprise the step of detecting an occurrence of an abnormal situation in advance and generating an alarm (S240). The warning unit 240 may be configured such that when the situation determination unit 130 determines that an abnormal situation has occurred (t 1 <t 2 ) before the second situation determination unit 230 determines that an abnormal condition has occurred, The control unit 130 determines that the detection is a prior detection and generates a warning (S240).

The process that occurs every time an abnormal situation occurs in the system 200 is similar to that of the method S100 shown in FIG. 5, and thus a detailed description thereof will be omitted.

The method for detecting abnormal conditions according to the embodiments described above can be at least partially implemented in a computer program and recorded in a computer-readable recording medium. A program for implementing the abnormal state detection method according to embodiments is recorded, and the computer-readable recording medium includes all kinds of recording devices for storing data that can be read by a computer. Examples of the computer-readable recording medium include ROM, RAM, CD-ROM, magnetic tape, floppy disk, optical data storage, and the like. The computer readable recording medium may also be distributed over a networked computer system so that computer readable code is stored and executed in a distributed manner.

While the present invention has been particularly shown and described with reference to exemplary embodiments thereof, it is to be understood that the invention is not limited to the disclosed embodiments, but, on the contrary, is intended to cover various modifications and equivalent arrangements included within the spirit and scope of the appended claims. . However, it should be understood that such modifications are within the technical scope of the present invention. Accordingly, the true scope of the present invention should be determined by the technical idea of the appended claims.

110: sensing unit
130:
150: training data generation unit
170: Parameter learning unit for detecting abnormal situation
210: second sensing unit
230: second situation determination unit
240: Warning part

Claims (15)

  1. A method for detecting an abnormal situation,
    Obtaining first sensor data from a facility device;
    Determining whether an abnormal condition has occurred from the obtained first sensor data;
    Generating training data from the obtained sensor data if it is determined that an abnormal situation has occurred;
    Updating parameters for detecting an abnormal situation through machine learning using the learning data; And
    Determining whether or not a next abnormal situation occurs based on the abnormal condition recognition parameter,
    Wherein the learning data includes sensor data acquired for a predetermined time period from an occurrence time of an abnormal situation.
  2. The method according to claim 1,
    Wherein the time range is determined based on the characteristics of the sensor data acquired before the occurrence time of the abnormal situation.
  3. The method according to claim 1,
    Obtaining second sensor data of a different type from the obtained sensor data; And
    And determining whether an abnormal situation has occurred from the obtained second sensor data.
  4. The method of claim 3,
    Further comprising the step of generating a warning when it is determined from the second sensor data that an abnormal condition will occur from the obtained sensor data before the occurrence of the abnormal condition is determined.
  5. The method according to claim 1,
    The abnormal condition recognition parameter may include at least one of support vector machine (SVM), support vector data description (SVDD), support vector regression (SVR), artificial neural network (ANN), and K-means clustering algorithm And updating the abnormal state detecting means.
  6. The method according to claim 1,
    Wherein the abnormal condition recognition parameter is updated from a plurality of training data accumulated during a predetermined time duration.
  7. The method according to claim 1,
    And the abnormal condition recognition parameter is updated in real time from the generated learning data.
  8. The method according to claim 1,
    And applying the sensor data acquired after the occurrence of the abnormal situation to the probability model to calculate the probability of occurrence of the next abnormal situation.
  9. 9. A computer-readable recording medium having recorded thereon a computer program for performing the method for detecting an abnormal situation according to any one of claims 1 to 8.
  10. An abnormal situation detection system, comprising:
    A first sensing unit for acquiring first sensor data from the facility apparatus;
    A first situation determiner for determining whether an abnormal situation occurs from the obtained first sensor data;
    A learning data generation unit for generating learning data from the obtained sensor data if it is determined that an abnormal situation has occurred; And
    And an abnormal condition recognition parameter learning unit for updating parameters for detecting an abnormal situation through machine learning using the learning data,
    Wherein the first situation determination unit determines whether or not a next abnormal situation occurs based on the abnormal condition recognition parameter, and the learning data includes sensor data acquired for a predetermined time range from the occurrence time of the abnormal condition. Situation detection system.
  11. 11. The method of claim 10,
    Wherein the time range is determined based on characteristics of the sensor data acquired before the occurrence time of the abnormal situation.
  12. 11. The method of claim 10,
    A second sensing unit for acquiring second sensor data of a type different from the first sensor data; And
    And a second status determination unit for determining an abnormal status from the second sensor data.
  13. 13. The method of claim 12,
    Further comprising a warning unit for generating a warning when the first situation determining unit determines that the abnormal situation will occur before the second situation determining unit determines that the abnormal situation occurs.
  14. 11. The method of claim 10,
    And a possibility calculating unit for calculating an abnormality occurrence probability by applying the obtained sensor data to a probability model.
  15. 11. The method of claim 10,
    The abnormal condition recognition parameter may include at least one of support vector machine (SVM), support vector data description (SVDD), support vector regression (SVR), artificial neural network (ANN), and K-means clustering algorithm Wherein the abnormality detection system comprises:
KR1020170081846A 2017-06-28 2017-06-28 Method for detecting abnormal situation and system for performing the same KR101970619B1 (en)

Priority Applications (1)

Application Number Priority Date Filing Date Title
KR1020170081846A KR101970619B1 (en) 2017-06-28 2017-06-28 Method for detecting abnormal situation and system for performing the same

Applications Claiming Priority (1)

Application Number Priority Date Filing Date Title
KR1020170081846A KR101970619B1 (en) 2017-06-28 2017-06-28 Method for detecting abnormal situation and system for performing the same

Publications (2)

Publication Number Publication Date
KR20190001781A true KR20190001781A (en) 2019-01-07
KR101970619B1 KR101970619B1 (en) 2019-04-22

Family

ID=65016754

Family Applications (1)

Application Number Title Priority Date Filing Date
KR1020170081846A KR101970619B1 (en) 2017-06-28 2017-06-28 Method for detecting abnormal situation and system for performing the same

Country Status (1)

Country Link
KR (1) KR101970619B1 (en)

Cited By (1)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
KR102101974B1 (en) * 2019-01-23 2020-04-17 주식회사 마키나락스 Anomaly detection

Citations (4)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
US20140039834A1 (en) * 2012-08-01 2014-02-06 Hitachi Power Solutions Co., Ltd. Method and apparatus for monitoring equipment conditions
KR20150113700A (en) * 2014-03-31 2015-10-08 삼성전자주식회사 System and method for diagnosis
KR20170022625A (en) 2015-08-21 2017-03-02 삼성전자주식회사 Method for training classifier and detecting object
KR101713985B1 (en) * 2016-09-02 2017-03-09 에이블맥스(주) Method and apparatus for prediction maintenance

Patent Citations (4)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
US20140039834A1 (en) * 2012-08-01 2014-02-06 Hitachi Power Solutions Co., Ltd. Method and apparatus for monitoring equipment conditions
KR20150113700A (en) * 2014-03-31 2015-10-08 삼성전자주식회사 System and method for diagnosis
KR20170022625A (en) 2015-08-21 2017-03-02 삼성전자주식회사 Method for training classifier and detecting object
KR101713985B1 (en) * 2016-09-02 2017-03-09 에이블맥스(주) Method and apparatus for prediction maintenance

Cited By (1)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
KR102101974B1 (en) * 2019-01-23 2020-04-17 주식회사 마키나락스 Anomaly detection

Also Published As

Publication number Publication date
KR101970619B1 (en) 2019-04-22

Similar Documents

Publication Publication Date Title
US10013303B2 (en) Detecting anomalies in an internet of things network
Gupta et al. Outlier detection for temporal data
US20190230108A1 (en) Cognitive information security using a behavioral recognition system
US9589181B2 (en) Person search method and device for searching person staying on platform
EP2905665B1 (en) Information processing apparatus, diagnosis method, and program
Zhang et al. Video anomaly detection based on locality sensitive hashing filters
Hodge et al. A survey of outlier detection methodologies
Takeuchi et al. A unifying framework for detecting outliers and change points from time series
US10402627B2 (en) Method and apparatus for determining identity identifier of face in face image, and terminal
US8504876B2 (en) Anomaly detection for database systems
Hayton et al. Static and dynamic novelty detection methods for jet engine health monitoring
US7395188B1 (en) System and method for equipment life estimation
EP2859418B1 (en) Generalized pattern recognition for fault diagnosis in machine condition monitoring
Bodik et al. Fingerprinting the datacenter: automated classification of performance crises
US20160219067A1 (en) Method of detecting anomalies suspected of attack, based on time series statistics
Clifton et al. Novelty detection with multivariate extreme value statistics
US8988236B2 (en) System and method for failure prediction for rod pump artificial lift systems
RU2573735C2 (en) Method and system for analysis of flight data recorded during aircraft flight
US7533070B2 (en) Automatic fault classification for model-based process monitoring
Qiu et al. Granger causality for time-series anomaly detection
US20180068233A1 (en) Self-learning log classification system
US9412067B2 (en) Anomaly detection in spatial and temporal memory system
JP4970531B2 (en) A method for automatically characterizing the behavior of one or more objects.
JP3209163B2 (en) Classifier
US8938404B2 (en) System and method for detecting abnormal audio events

Legal Events

Date Code Title Description
A201 Request for examination
E902 Notification of reason for refusal
E701 Decision to grant or registration of patent right
GRNT Written decision to grant