KR101843879B1 - System for detecting abnormality data, method thereof and computer recordable medium storing the method - Google Patents

Abstract

The present invention relates to an abnormal data detection system, a method for the same, and a computer-readable recording medium in which the method is recorded. The present invention collects data about emission materials and states through a measurement device installed in a fluid discharge facility including pollutants, determines the operating state of the facility discharging fluid by applying hierarchical statistical technique to the collected data, and determines abnormality of the collected data about each operation state. The abnormal data detection system, which collects and analyzes measurement value signals from a plurality of measurement devices outputting the measurement value signals by measuring the state of fluid and a plurality of materials included in the fluid, includes: a communication unit which receives the measurement value signals from the plurality of measurement devices; a database unit which stores the measurement value signals received through the communication unit as measurement data; a user interface unit which provides an interface for the selection of detection criteria and abnormal data detection objects and provides an interface displaying a detection result; and a detection unit which analyzes the measurement data stored in the database unit by using the detection criteria and the abnormal data detection objects selected through the user interface unit, determines the operation state of the facility in which the plurality of measurement devices is installed according to the analyzed result, selects a singular value depending on the determined operation state, and displays the determination result and the selection result through the user interface unit as the detection result. The present invention is able to precisely detect abnormal data.

Description

TECHNICAL FIELD The present invention relates to an abnormal data detection system, a method for the same, and a computer readable recording medium on which the method is recorded.

The present invention relates to an abnormal data detection system, a method for the same, and a computer-readable recording medium on which the method is recorded. More particularly, the present invention relates to a data detection system, An abnormal data detection system that collects data, identifies the operation state of a facility that discharges fluid by applying hierarchical statistical techniques to the collected data, and determines whether or not the collected data is abnormal for each operation state. And a computer-readable recording medium on which the method is recorded.

In order to prevent environmental pollution caused by a facility for discharging a liquid containing a pollutant or a gas, a continuous measurement of pollutants is carried out by various automatic measurement monitoring systems. For example, in order to prevent air pollution caused by pollutants contained in the exhaust gas of a workplace where a chimney is installed, the air pollutants emission sites (TSP, Total Suspended Particles), nitrogen oxides (NOx), sulfur oxides (CleanSYS), which can continuously monitor the seven pollutants such as SOx, HCl, HF, NH3, and CO during 24 hours, Measurements can be collected every 5 minutes from more than 1,500 chimneys nationwide.

However, since the conventional automatic measurement monitoring system accumulates a large amount of data through the operation of a system connected with a large number of measuring instruments, the reliability of the data can not be guaranteed, so that the accumulated data can be simply monitored in real- And to use it as administrative data for

That is, in managing the quality of the data collected in the conventional automatic measurement monitoring system, it is necessary to consider the measurement value of the measuring apparatus and the normal, Based on the 4 status information of faulty operation, power failure, and maintenance, and records of operation status of the facilities, it is possible to select the abnormal data among the accumulated data of experts who understand the target facilities. Experience, and competence, there is a problem in that the reliability of the data can not be guaranteed.

Patent Registration No. 10-1199924

An object of the present invention is to solve the problems of the prior art by introducing statistical processing techniques for quality control of collected data from pollutant discharge facilities and thereby eliminating variations in data quality due to differences in competence of skilled workers A method for the same, and a computer-readable recording medium on which the method is recorded.

It is also an object of the present invention to provide a method and apparatus for accurately detecting abnormal data by detecting an operation state of a discharge facility according to the distribution pattern of data collected from a pollutant discharge facility and applying an abnormal data selection filter corresponding to the operation state An abnormal data detection system, a method for the same, and a computer readable recording medium on which the method is recorded.

In order to accomplish the above object, the abnormal data detection system of the present invention includes an abnormality detection system for collecting and analyzing the measurement value signals from a plurality of measurement devices for measuring states of a plurality of substances contained in a fluid and a fluid, A data detection system comprising: a communication unit for receiving the measurement value signal from the plurality of measurement devices; A data base for storing the measurement value signal received through the communication unit as measurement data; A user interface unit for providing an interface for selection of an abnormal data detection target and a detection criterion and providing an interface for displaying a detection result; And analyzing measurement data stored in the database unit using the abnormal data detection object and the detection criterion selected through the user interface unit and determining an operation state of the facility equipped with the plurality of measuring instruments according to a result of the analysis, And a detection unit for selecting a specific value according to the determined operation state and displaying the determination result and the selection result through the user interface unit as the detection result.

Here, the detection unit may define a distribution range of measurement data for each operation state of a facility in which the plurality of measurement devices are installed, with respect to the measurement data stored in the database unit, and transmit the measured value signal input through the communication unit to a defined distribution range And a pattern analyzer for determining the operation state of the facility in which the plurality of measuring instruments are installed.

Also, the pattern analyzing unit may perform K average ++ clustering analysis for dividing the measurement data into a set corresponding to the number of operating states, and calculate a distribution of a plurality of sets derived from the K average ++ clustering analysis execution result Range can be defined as a distribution range of measurement data for each operating state.

Meanwhile, the detecting unit may select a plurality of main items by Principal Component Analysis among a plurality of items related to states of a plurality of substances and fluids measured by the measuring apparatus, Calculating a T²-statistic for the measurement data stored in the database unit, comparing the calculated T²-statistic with an upper-limit management value based on the T²-statistic, and selecting a specific value of the measurement data stored in the database unit And an element analysis unit.

The detecting unit may be configured to select a plurality of main items by Principal Component Analysis among a plurality of items related to states of a plurality of substances and fluids measured by the measuring apparatus, A Q-statistic is calculated for the measurement data stored in the database unit with respect to the Q-statistics, and the main component for selecting the specific value of the measurement data stored in the database unit by comparing the calculated Q- And an element analysis unit.

On the other hand, the detection unit may select two items out of a plurality of items related to the states of the plurality of substances and fluids measured by the measuring instrument, and determine T < 2 > for the measurement data stored in the database unit for the two selected items A correlation analyzer for calculating a statistic and comparing the calculated T < 2 > -statistic with an upper-limit management value based on the T < 2 > -statement to select a specific value of the measurement data stored in the database.

The detecting unit may be configured to select one item among a plurality of items relating to the states of the plurality of substances and fluids measured by the measuring device and to determine the one or more items Calculating a mean and a standard deviation for each operation state, comparing the calculated average and the standard deviation with a control threshold value of a predetermined confidence level according to the calculated average and the calculated standard deviation (?), And selecting a single value of the measurement data stored in the database unit And a variable analyzing unit.

On the other hand, the detection unit may calculate the pre-correction data on the basis of the correction value applied to the measurement data when the measurement apparatus is measured, compare the difference between the measurement data and the pre- And a correction conversion analysis unit for selecting a specific value of the data.

The detection unit may set the distribution range of measurement data for each operating state defined by the pattern analysis unit as a first rule set and compare the measured value signal inputted through the communication unit with the first rule set, And a rule set management unit for determining the operation status of the facility in which the measuring instrument is installed.

On the other hand, the detection unit may set an upper limit management value based on the T < 2 > -statistic amount of the principal component analysis unit as a second rule set, and calculate a T & And a rule set manager for comparing the measured value signal inputted through the communication unit with a specific value of the measured value signal.

The detection unit may set the upper limit management value based on the Q-statistic amount of the principal component analysis unit as a third rule set, and calculate a Q-statistic amount calculated for the measured value signal input through the third rule set and the communication unit as And a rule set manager for comparing the measured value signal inputted through the communication unit with a specific value of the measured value signal.

On the other hand, the detection unit sets the upper limit management value based on the T < 2 > -state of the correlation analyzer as a fourth rule, and calculates a T < 2 > -statement amount calculated for the measured value signal input through the fourth rule- And a rule set manager for comparing the measured value signal inputted through the communication unit with a specific value of the measured value signal.

The detecting unit may set the management limit value of the single variable analyzing unit as the fifth rule and compare the value of the measured value signal input through the communication unit with the fifth rule set, And a rule set manager for determining whether the signal has a singular value.

In order to achieve the above object, the abnormal data detection method of the present invention comprises the steps of: collecting and analyzing the measurement value signals from a plurality of measurement devices for measuring a state of a plurality of substances contained in a fluid and a fluid, A data detection method, comprising: a first step of receiving a measurement value signal from a plurality of measurement devices; A second step of storing the measured value signal received in the first step as measurement data; A third step of receiving an abnormal data detection object and a detection standard; And analyzing the measurement data stored in the second step using the abnormal data detection object and the detection criterion input through the third step, and determining the operation state of the facility equipped with the plurality of measuring instruments according to the analyzed result And a fourth step of selecting a specific value according to the determined operation state, and displaying the determination result and the selection result.

According to another aspect of the present invention, there is provided a computer-readable recording medium recording an abnormal data detection method.

The present invention applies a statistical filtering method for automatic sorting of abnormal data on data collected from a pollutant discharge facility to prevent deviation in the quality of confirmed data among personnel in charge and to improve the reliability of confirmed data have.

The present invention also relates to a method and apparatus for detecting pollutant discharge facility operation status according to the form of the collected data and determining a correlation between all data corresponding to the detected operation status, a correlation between a pair of variables, It is possible to detect more sophisticated abnormal data through multilevel filtering operation which comprehensively reflects data correction according to temperature, humidity and oxygen.

Meanwhile, the present invention has the effect of processing a large amount of data collected in 1,500 or more pollutant discharge facilities in real time by introducing Ruleset reflecting the multistage filtering operation according to the operating state.

1 is a diagram illustrating an abnormal data detection system according to an embodiment of the present invention.
2 is a view showing an example of a user interface screen provided by a user interface unit of the abnormal data detection system according to an embodiment of the present invention.
FIG. 3 is a diagram specifically illustrating a detection unit of the abnormal data detection system according to an embodiment of the present invention.
4A to 4D are graphs illustrating the operation of the pattern analyzer of the abnormal data detection system according to an embodiment of the present invention.
5A and 5B are graphs illustrating the operation of the correlation analyzer of the abnormal data detection system according to an embodiment of the present invention.
6 is a diagram illustrating an example of a screen for controlling the ruleset management unit provided by the user interface unit of the abnormal data detection system according to an embodiment of the present invention.
7 is a flowchart illustrating an abnormal data detection method according to an embodiment of the present invention.
8A to 8C are graphs illustrating a user interface unit of the abnormal data detection system according to an embodiment of the present invention.
9A and 9B are diagrams illustrating other graphs provided by the user interface unit of the abnormal data detection system according to the embodiment of the present invention.
10A to 10F are graphs showing the operation state classification operation of the abnormal data detection system according to an embodiment of the present invention.

The description of the disclosed technique is merely an example for structural or functional explanation and the scope of the disclosed technology should not be construed as being limited by the embodiments described in the text. That is, the embodiments are to be construed as being variously embodied and having various forms, so that the scope of the disclosed technology should be understood to include equivalents capable of realizing technical ideas.

Meanwhile, the meaning of the terms described in the present application should be understood as follows.

The terms " first ", " second ", and the like are used to distinguish one element from another and should not be limited by these terms. For example, the first component may be referred to as a second component, and similarly, the second component may also be referred to as a first component.

It is to be understood that when an element is referred to as being "connected" to another element, it may be directly connected to the other element, but there may be other elements in between. On the other hand, when an element is referred to as being "directly connected" to another element, it should be understood that there are no other elements in between. On the other hand, other expressions that describe the relationship between components, such as "between" and "between" or "neighboring to" and "directly adjacent to" should be interpreted as well.

It is to be understood that the singular " include " or "have" are to be construed as including the stated feature, number, step, operation, It is to be understood that the combination is intended to specify that it is present and not to preclude the presence or addition of one or more other features, numbers, steps, operations, components, parts or combinations thereof.

Each step may take place differently from the stated order unless explicitly stated in a specific order in the context. That is, each step may occur in the same order as described, may be performed substantially concurrently, or may be performed in reverse order.

All terms used herein have the same meaning as commonly understood by one of ordinary skill in the art to which the disclosed technology belongs, unless otherwise defined. Terms defined in commonly used dictionaries should be interpreted to be consistent with meaning in the context of the relevant art and can not be construed as having ideal or overly formal meaning unless expressly defined in the present application.

FIG. 1 and FIG. 3 are diagrams showing an abnormal data detection system according to an embodiment of the present invention, and FIG. 2 is an example of a user interface screen provided by the abnormal data detection system according to an embodiment of the present invention The abnormal data detection system according to an embodiment of the present invention includes a communication unit 100, a database unit 200, a user interface unit 300, and a detection unit 400.

The communication unit 100 receives measurement value signals from a plurality of measurement devices 10 and outputs the received signals to the data base unit 200 or the detection unit 400. [ Here, the communication unit 100 may be a wired or wireless communication. However, when using wireless communication, a wireless communication module, for example, a Code Division Multiple Access (CDMA) ) Module and receives information directly from the communication unit 100 or allows a terminal (not shown) having a wireless communication module to read the weighing values of the individual weighing equipment 10, .

The plurality of measuring instruments 10 measures the state of a plurality of substances contained in the fluid and the state of the fluid and is installed in a pollutant discharge facility, for example, a chimney of an oil refinery, , Flow rate, temperature, etc., in real time, and transmits a signal for the measured value to the communication unit 100. [ At this time, the pollutants measuring the concentration in the plurality of measuring instruments 10 may include seven pollutants such as dust, nitrogen oxides, sulfur oxides, hydrogen chloride, hydrogen fluoride, ammonia, and carbon monoxide.

In addition to the measured values of the concentrations of the above-mentioned pollutants, the plurality of measuring instruments 10 may further include, as transmission items for the correction, the exhaust gas flow rate, the exhaust gas temperature, the combustion chamber internal temperature, the oxygen concentration, For example.

The database unit 200 stores the measurement value signal received through the communication unit 100 as measurement data and outputs the measurement data stored under the control of the detection unit 400 to the detection unit 400. At this time, the database unit 200 can store data collected from a plurality of measuring instruments 10 for several years. At this time, the database unit 200 stores seven pollutant measurement items (dust, nitrogen oxides, sulfur oxides, hydrogen chloride, hydrogen fluoride, ammonia, and carbon monoxide) according to the types of values measured by the plurality of measurement devices 10. [ Concentration) and 5 transmission items (exhaust gas flow rate, exhaust gas temperature, combustion chamber internal temperature, oxygen concentration, humidity).

2, the user interface unit 300 provides an interface for the system user to select the abnormal data detection object and detection standard, and provides an interface for displaying the detection result calculated by the detection unit 400 do.

The user interface unit 300 not only provides various selection interfaces for allowing a system user to select an abnormal data detection object and detection standard but also allows the system user to easily recognize the abnormal data calculated by the detection unit 400 The detection result can be displayed in various ways, which will be described in detail with reference to FIG. 2.

First, the user interface unit 300 selects sorting by factory ID or sorting method according to the emission facility classification system through the type selection item 311, by classification. In addition, the user interface unit 300 may also provide a method (312) of searching for a business entity name by directly inputting a business entity name in the selection of an ideal data detection object. 2, the user interface unit 300 provides a list 313 of businesses to allow the user to select a target business site, for example, a GS Caltex Yeosu factory To be selected. The user interface unit 300 may also provide a list 314 of target outlets, for example, a chimney so that the user can select a target outlet, for example, a boiler in the Yeosu plant of GS Caltex Co., . Meanwhile, the user interface unit 300 may provide emission discrimination information 315 such as standard oxygen concentration and standard moisture concentration for correcting the detailed information of the selected target outlet and data.

In addition, the user interface unit 300 sets (320, Outlier Detect), rules (330, Ruleset Management) and evaluates (340, Ruleset) a rule set reflecting the multi- Evaluation), but is not limited thereto. For example, the user interface unit 300 may provide an item 321 for selecting a collection period of measurement data to be fetched from the database unit 200 for rule set setting.

The detection unit 400 analyzes the measurement data stored in the database unit 200 using the selected ideal data detection object and detection reference through the user interface unit 300, And displays the determination result and the selection result through the user interface unit 300 as the detection result. Here, anomalous data or singular values may mean data in a form that deviates from the range characteristics such as general emission concentration in the same operating state.

3, the detecting unit 400 includes a pattern analyzing unit 410, a principal component analyzing unit 420, a correlation analyzing unit 430, a single variable analyzing unit 440, (450) and rule set management unit (460).

The pattern analyzing unit 410 performs pattern analysis on the measurement data stored in the database unit 200 and configures a distribution range of measurement data for each operation state of a facility in which a plurality of measurement devices 10 are installed, The measured value signal inputted in real time through the controller 100 is compared with the configured distribution range to determine the operation state of the facility in which the plurality of measuring instruments 10 is installed. At this time, the user interface unit 300 may provide a pattern analysis item 322 so that a time window and a pattern when analyzing a pattern of measurement data stored in the database unit 200 of the pattern analysis unit 410 (Max Pattern Number) can be set. In addition, the pattern analysis can be defined as defining pollutant discharge concentration characteristics according to the operation mode of the pollutant discharge facility and clustering time zones showing similar emission concentration ranges.

4B), nitrogen oxide concentration (shown in FIG. 4C), exhaust gas temperature (also shown in FIG. 4B), and oxygen concentration 4d), it is possible to define four patterns of operation states as shown in Table 1 below, but the present invention is not limited to this, and various patterns such as operation interruption, preparation for operation, subtraction operation, Can be defined.

Pattern Definition Measurement data characteristics Standard combustion Intermediate flow rate, medium oxygen concentration, intermediate temperature (exhaust gas) Air supply excess Low flow rate, low temperature (exhaust gas), high concentration of nitrogen oxide High temperature combustion High flow rate, high temperature (exhaust gas), low oxygen concentration Downtime No flow

Here, the measurement data used in Table 1 are data collected from the combustion facility using solid fuel for one month, and through the user interface unit 300, as shown in FIGS. 4A to 4D, , That is, the standard combustion is red, the excess air supply is blue, the high temperature combustion is green, and the shutdown is yellow.

That is, since the correlation between the concentration distribution of pollutants, the distribution of temperature, the distribution of oxygen concentration, the distribution of flow rate, and the measured values differs according to the operation state of the facility, the distribution and interrelation of these measurement data It is possible to define the range of the characteristic that distinguishes the operating state. 4A to 4D, it can be seen that the operation state of the facility is not classified according to the distribution characteristic of a single measurement value but is divided according to the characteristics and correlation of a plurality of measurement values.

In order to distinguish the operation state, cluster analysis, for example, K average ++ (KMeanPlusPlus) clustering, Fuzzy-K mean clustering and DBSCAN (Density-Based Spatial Clustering of Applications with Noise) In particular, it is preferable to use a K average ++ clustering method in order to maximize the degree of similarity between divided patterns. In addition, an earth mover distance measurement method capable of quantifying a distribution form of a variable by a distance measurement method for non-inference measurement may be used, but is not limited thereto.

An example of a method of performing pattern analysis by the pattern analysis unit 410 will be described in detail as follows.

First, the pattern analyzing unit 410 calculates statistical values for each measurement item such as oxygen concentration, flow rate, nitrogen oxide concentration, and exhaust gas temperature, that is, average, standard deviation, and degree of distortion for pattern analysis. Thereafter, the pattern analyzing unit 410 performs normalization for each item of measurement data to prevent clustering from being influenced by the measurement data of items having a large value as the units between the measurement items differ. Next, the pattern analyzing unit 410 calculates a weight for processing the measurement data as a weighted distance function before clustering when there is an item such as an oxygen concentration, which is intended to have a larger weight intentionally The coefficient can be set. Next, the pattern analyzing unit 410 analyzes the number of the active states (Max Pattern Number) (see the pattern analysis item 322 shown in FIG. 2) with respect to the measurement data in the database unit 200, The K average ++ clustering analysis is performed by dividing into four sets of downtime, operation preparation, subtraction operation, and prevention facilities, and the distribution range of the plurality of sets derived from the K average ++ clustering analysis execution result It can be defined as the distribution range of measurement data. Here, K average clustering is an algorithm for grouping given data into k clusters, as known to those of ordinary skill in the art, and operates in a manner that minimizes dispersion of the distance difference with each cluster, and K average ++ clustering is a function of K average clustering The algorithm can select the initial value (the center point of the clustering) as the optimal value and apply the K average ++ clustering algorithm to prevent the degradation of the algorithm performance according to the initial value selection. In addition, the pattern analyzing unit 410 may use a time window (see the pattern analysis item 322 shown in FIG. 2) for controlling the time range of the measurement data in extracting the pattern from the measurement data .

If the distribution of the specific pattern for defining each operating state as a result of the pattern analysis is not clear, the pattern analyzing unit 410 may generate a detailed pattern by further subdividing the target pattern or integrate the target pattern into another similar pattern Function can be provided. For this, the pattern analyzer 410 may provide a function of defining a transitional pattern according to the variation of the operating state.

The principal component analyzing unit 420 analyzes a characteristic of the measurement data stored in the database unit 200 using the T < 2 > statistic or the Q-statistic based on a model constructed through a principal component analysis (PCA) Value, preferably a specific value for each operating state. Hereinafter, the operation of the principal component analysis unit 420 will be described in detail.

First, multivariate statistical analysis should be applied in order to detect abnormal data, ie, singular value, among multiple data of dependent variables. Such multivariate statistical analysis can be performed by using the data obtained from normal operating conditions And a second step of identifying a specific value that deviates from a normal process using a model. Here, the first step may be performed by principal component analysis in the statistical method, and the second step may be performed using the T²-statistic or the Q-statistic (SPE) of the hotel ring. In this case, in the second step, only one of the T²-statistic or the Q-statistic can be used, but preferably the T²-statistic and the Q-statistic can be used at the same time.

In the data in which the measurement values are added in real time, the principal component analysis model should be varied over time. The principal component analysis unit 420 uses the time window for controlling the time range of the measurement data, You can delete the data and perform principal component analysis through the new Covariance Function created by the new time window.

First, an arbitrary data matrix X with a degree n x j can be expressed as a linear combination of a scoring matrix T with order n × k and a loading matrix P with order j × k. Where n is the total number of measurements, k is the number of principal components, and j is the number of variables. The first component is the component that describes the data the most and the second component is the component that explains the data the second. Thus, we can interpret the main components such as the third and fourth. The number k of principal components has a number less than or equal to one of n or j. At this time, the data matrix X can be constructed as shown in the following equation.

Here, E denotes a residual matrix having a degree n x j.

)은 하기 수학식에 의해 계산될 수 있다.If the loading matrix calculated for the measurement data stored in the database unit 200 is P, a score value (?) For a new measurement value signal received through the communication unit 100

) Can be calculated by the following equation.

Here, the lower case means a vector, the upper case means a matrix, and t means vectorized data in which a new measured value signal received through the communication unit 100 is reflected.

The T < 2 > -statistics of a hotel ring means how similar the old measurement data is to the new measurement data and is calculated by calculating the T < 2 > statistic, i.e. Mahalanobis distance,

Here, S denotes a diagonal matrix of a covariance matrix for the score value of the principal component analysis for past measurement data, i.e., a diagonal matrix of eigenvalues of selected data among measurement data, and x denotes a center of gravity Quot; means a predicted score vector value.

In addition, the principal component analysis unit 420 can calculate the upper management limit value according to the following equation in order to select the specific value of the new measurement data with respect to the past measured data stored in the database unit 200 have.

는 α분위수에서 자유도가 p와 n-1인 F분포의 임계값(Critical Value)이다. 만약, 주성분 요소 분석부(420)는, 수학식 3에 의해 계산된 Tㅂ값이 수학식 4에 의해 계산된 임계값을 초과하게 되면 특이값으로 판단하게 된다.Where n is the number of measured values in the measurement data, p is the number of principal components used in the principal component analysis,

Is the critical value of the F distribution whose degrees of freedom are p and n-1 in the α-quantile. If the T value calculated by Equation (3) exceeds the threshold value calculated by Equation (4), the principal component analysis unit 420 determines that the T value is an outlier.

The Q-statistic is used to determine the abnormality while monitoring the difference between the actual measured value and the estimated value, and can be calculated by the following equation in consideration of the portion corresponding to the residual matrix in the principal component analysis.

Here, P is a loading matrix (covariance matrix) calculated for the measurement data stored in the database unit 200, x is a predictive score vector value centered as an average for new measurement data, and I is a unit matrix .

)은 하기 수학식을 이용하여 계산된다.Under normal conditions, the Q-statistics follow a multivariate normal distribution, so the threshold can be estimated through a weighted chi-square distribution,

) Is calculated using the following equation.

는 정규분포의 (1-α)분위수에 해당하는 값을 의미하며,

는 하기 수학식에 의해 산출될 수 있다.here,

Means a value corresponding to the (1-α) quantile of the normal distribution,

Can be calculated by the following equation.

는 하기 수학식에 의해 산출될 수 있다.Also,

Can be calculated by the following equation.

는 주성분 요소 분석에서 얻어진 고유값(eigenvalue)을 의미하며, a는 변수의 개수를 의미하고, d는 주성분 분석에서 사용된 주성분의 개수를 의미한다.here,

Denotes the eigenvalue obtained from principal component analysis, a denotes the number of variables, and d denotes the number of principal components used in the principal component analysis.

The principal component analysis unit 420 determines that the Q value calculated by Equation (5) exceeds the threshold value calculated by Equation (6) as an abnormal data, that is, a singular value.

The correlation analyzing unit 430 selects two items out of the plurality of items related to the states of the plurality of substances and fluids measured by the measuring instrument 10 and stores the two items in the database 200 related to the selected item A correlation analysis is performed on the measurement data by the Mahalanobis distance method, and the analysis results are used to select the specific values of the measurement data on the two items. That is, the correlation analyzing unit 430 calculates a correlation using measurement data on two items of the measurement data stored in the database unit 200 as two variables, for example, the concentration of nitrogen oxide and the concentration of oxygen And extracts the singular value by using the correlation as a variable.

The correlation analyzing unit 430 analyzes the measurement data stored in the measurement device 10 received through the communication unit 100 in response to a situation that the measurement data should be detected as a specific value in the mutual relationship between two items A covariance matrix is constructed using the measurement data on the two items, and the Mahalanobis distance is calculated based on the covariance matrix as shown in equation (3), and the calculated value is multiplied by p, which is the number of principal component elements in equation If the number of items for correlation analysis, 2, is exceeded, it is extracted as abnormal data. That is, the correlation analyzer 430 transforms the equation (4) into a mathematical expression for determining a threshold value for extracting a singular value between two variables as shown in the following equation.

는 100(1-α)% 신뢰 구간에 대한 F분포의 역함수를 의미한다.Here, m is a number of items of observation data, and when two items are used, 2 can be substituted as described above. n is the number of measured values in the measurement data, which may be the number of 5-minute data in case of the Tele-Monitoring System (TMS), which is a chimney remote monitoring system.

Is the inverse of the F distribution for the 100 (1-α)% confidence interval.

)와 산소 농도(

)의 좌표값(

,

)을 도 5a에 도시된 바와 같이 공분산 평면으로 사상(Mapping)하고 공분산 분포의 중심점으로부터 각 좌표의 거리를 원 형태의 분포로 환산한 거리, 즉, 마하라노비스 거리를 측정하게 된다.5A and 5B are graphs for explaining the operation of the correlation analyzing unit 430. The correlation analyzing unit 430 compares two selected items, for example, the concentration of nitrogen oxides (

) And oxygen concentration

) Coordinate value (

,

) Is mapped to the covariance plane as shown in FIG. 5A, and the distance obtained by converting the distance of each coordinate from the center point of the covariance distribution into the distribution of the circle shape, that is, Mahalanobis distance is measured.

First, assuming that the measurement data of two correlated items are distributed in the time zone as shown in the left side of FIG. 5A, the measurement data of the two items in the same time zone are ignored and the two- The converted result is the same as the coordinates shown on the right side of FIG. 5A.

In this case, as shown in FIG. 5B, when the transformed coordinate value (Multivariate outlier: our of control) is located at a position significantly different from other coordinate values, it can be judged as an abnormal data, that is, a singular value.

The correlation analyzer 430 transforms the elliptical distribution relationship into a circular distribution relationship as shown in Figs. 5A and 5B through the Mahalanobis distance methodology, and measures the distance between the center point and each measured value It is possible to quantitatively extract the singular value. In other words, it is difficult to quantitatively judge which coordinate value is a singular value by referring to the elliptic correlation distribution depending on the type of data mapped to the covariance plane. When the coordinates of two specific measured values are one Is located at the vertex of the ellipse and the other is located near the center point of the ellipse, the distance from the center point is the same, one is the normal value and the other is the singular value. That is, the correlation analyzer 430 uses the covariance matrix as shown in Equations (2) and (3) in converting the elliptic distribution relationship into the circular distribution.

The single variable analyzing unit 440 determines a normal distribution range of data for each item of the measurement data storing the measurement value signal of the measuring instrument 10 received through the communication unit 100, Extrapolate to an extrinsic value if out of range. In particular, the single variable analyzing unit 440 determines the normal distribution range of each item of the measurement data differently according to the operation state of the facility in which the plurality of measuring instruments 10 is installed.

That is, the single variable analyzing unit 440 selects one item among a plurality of items related to the states of a plurality of substances and fluids measured by the measuring instrument 10, Calculates a mean and a standard deviation for each of the operation states determined by the pattern analysis unit 410 with respect to the measurement data stored in the storage unit 410 and calculates a predetermined confidence level based on the calculated average and the calculated standard deviation? , 95, 99, 99.99% confidence level, and selects the specific value of the measurement data stored in the database unit (200).

The range for detecting the singular value determined by the single variable analysis unit 440 can be expressed by a management limit range having an upper limit and a lower limit. For example, a range of three times the standard deviation, that is, The measured values can be selected with a specific value. Applying a 3σ management limit value means that it is outside of the normal range for the test of the null hypothesis (corresponding to the singular value), that is, the 99.5% confidence level, using a significance level such as the following equation: In this case, the trust level applied to the single variable analyzer 440 can be arbitrarily set according to the management level of the user, and can be set to various values such as 90, 95, 99, 99.99% as well as the 99.5% confidence level.

는 측정 데이터의 각 측정값, μ는 데이터 베이스부(200) 내에 저장된 측정 데이터에 대해 패턴 분석부(410)에 의해 판정된 가동 상태 별로 계산된 평균을 의미한다.here,

Denotes an average calculated by the pattern analyzing unit 410 for each measurement value stored in the database unit 200, and?

That is, the management upper limit value UCL and the management lower limit value LCL of the range for detecting the singular value determined by the single variable analyzing unit 440 can be determined by the following equation.

Here, p denotes a probability value based on a normal distribution assumption.

The calibration conversion analysis unit 450 calculates the pre-correction data based on the correction value applied to the measurement data stored in the database unit 200 when the measurement apparatus 10 is measured, The difference is compared with a preset value to select a specific value of the measurement data. That is, the correction conversion analysis unit 450 analyzes the characteristics of the data distribution before and after the correction of the oxygen concentration, the temperature, and the flow rate, and extracts the singular value, that is, the abnormal data.

The plurality of measuring instruments 10 installed in the pollutant discharge facility transmits the correction value to the communication unit 100 as a measurement value, Even if no specific value is found in the measurement value received from the measuring instrument 10 in the case of distortion, the actual value should be treated as the ideal data. Therefore, the correction value conversion unit 450 selects the specific value of the pre-correction data . Further, as the flow rate and the temperature of the fluid discharged from the pollutant discharge facility suddenly change, the correction applied to the correction analysis section 450 may suddenly increase or decrease the value applied to the correction. (In fact, it is not an abnormal value, which is an abnormal data, but an error to be selected as an abnormal value, etc.).

In addition to the method of calculating the pre-correction data, the correction conversion analysis unit 450 may directly compare the pre-correction data with the measurement data stored in the data base unit 200, It does not.

A process of converting the data on the pollutant measurement concentration in the measurement data into the pre-correction data by the temperature, the oxygen concentration, and the flow rate in the correction conversion analysis unit 450 will be described below. At this time, temperature and pressure are converted into Standard Temperature and Pressure (STP) in the following formula, and pressure is assumed to be 1 atm.

)을 오염물질 농도의 전송용 측정값(

)으로 환산하는 수학식은 아래와 같다.The measured value of the concentration of the pollutant in the measuring device 10

) For the transfer of the pollutant concentration (

) Is as follows.

는 표준상태(0℃, 1기압)로 환산한 오염물질의 농도(mg/Smㅃ, ppm)이고,

는 실측 또는 임의의 설정 온도에서 측정한 오염물질의 농도(mg/mㅃ, ppm)이며,

는 실측 또는 임의의 설정 온도(℃)이다.here,

Is the concentration of contaminant (mg / Sm ㅃ, ppm) converted to the standard state (0 ℃, 1 atm)

Is the concentration (mg / m ㅃ, ppm) of the pollutant measured at the actual or arbitrary set temperature,

Is the actual temperature or an arbitrary set temperature (占 폚).

)을 오염물질 농도의 전송용 측정값(

)으로 환산하는 수학식은 아래와 같다.In addition, the measured value of the contaminant concentration in the measuring instrument 10

) For the transfer of the pollutant concentration (

) Is as follows.

는 표준산소 농도로서 배출시절의 연소 장비를 적정하게 운전하고 배출 기체 희석을 방지하기 위해 설정되는 농도(%)를 의미하며,

는 실측 산소 농도(%)를 의미한다.here,

Means the concentration (%) which is set as a standard oxygen concentration for proper operation of the combustion equipment at the time of discharge and prevention of exhaust gas dilution,

Means the actual oxygen concentration (%).

)을 배출 유량의 전송용 측정값(

)으로 환산하는 수학식은 아래와 같다.Further, in the measuring device 10, an actual measured value of the discharge flow rate (

) For the discharge flow rate measurement (

) Is as follows.

는 표준상태(0℃, 1기압)로 환산한 유량(㎥/30분)이고,

는 실측 또는 임의의 설정 온도에서 측정한 유량(㎥/30분)이다.here,

Is the flow rate (m3 / 30 min) converted to the standard state (0 DEG C, 1 atm)

(M3 / 30 minutes) measured at the actual or arbitrary set temperature.

)을 배출 유량의 전송용 측정값(

)으로 환산하는 수학식은 아래와 같다.On the other hand, the measured value of the discharge flow rate by the oxygen concentration correction in the measuring instrument 10

) For the discharge flow rate measurement (

) Is as follows.

는 표준상태(0℃, 1기압)로 환산한 유량(㎥/30분)이고,

는 실측 또는 임의의 설정 온도에서 측정한 유량(㎥/30분)이며,

는 실측 또는 임의의 설정 온도(℃)이다.here,

Is the flow rate (m3 / 30 min) converted to the standard state (0 DEG C, 1 atm)

(M < 3 > / 30 min) measured at the actual or arbitrary set temperature,

Is the actual temperature or an arbitrary set temperature (占 폚).

The correction conversion analysis unit 450 can apply a general confidence interval methodology to select singular values. In other words, the correction conversion analysis unit 450 accumulates the value obtained by calculating the difference between the measurement data and the pre-correction data, and uses the average and standard deviation of the accumulated data to compare the measurement data to be determined as the singular value, It is checked whether the difference of the entire data is within the range of the upper limit and the lower limit at the confidence level of 95% or 99.5%.

Here, the correction conversion analysis unit 450 may convert the difference between the measured data (corrected data) and the unmodified data into a percentage by using the following equation, but is not limited thereto.

The ruleset management unit 460 analyzes the results of the operational state classification, the principal component analysis, the correlation analysis, and the single variable analysis result, which are obtained through the preliminary analysis of the measurement data in the database unit 200, The principal component analyzing unit 420, the correlation analyzing unit 430 and the single variable analyzing unit 440 and stores the input method and results in the form of a rule set. In the case of analyzing the inputted measured value signal in real time, a specific value, that is, an ideal data is extracted only by applying the rule set without performing calculation operation on the measurement data in the separate database unit 200. In other words, the singular value selection method by the pattern analyzing unit 410, the principal component analyzing unit 420, the correlation analyzing unit 430, and the single variable analyzing unit 440, as described above, The statistical analysis and the data calculation are required. Therefore, the rule set management unit 460 compensates for these disadvantages because it is not suitable for real-time abnormal data extraction.

6 is a diagram illustrating an example of a screen for controlling the ruleset management unit 460 provided by the user interface unit 300. The operation of the ruleset management unit 460 will be described with reference to FIGS. same.

The rule set management unit 460 sets the distribution range of the measurement data for each operating state defined by the pattern analysis unit 410 as a first rule set, Value signals are compared with each other to determine the operation state of the facility in which a plurality of measuring instruments 10 are installed. For example, the ruleset management unit 460 divides the operation state into an excess air, a high temperature, and a downtime state, as shown in FIG. 6, and the user interface unit 300 controls the ruleset management unit 460 And a selection interface tool 331 for displaying information on the distribution range of the measurement data set for each operation state.

The rule-set management unit 460 sets the upper limit management value based on the T2-statistic calculated by the principal component analysis unit 420 using Equation (4) as the second rule set, and sets the second rule set to the communication unit 100, The measured value signal inputted through the communication unit 100 can be compared with the T < 2 > -state calculated using Equation (3) to determine whether the measured value signal inputted in real time is a singular value. For example, the ruleset manager 460 sets the upper limit management value based on the T < 2 > -state calculated using Equation 4 as a threshold value, and the user interface unit 300 sets the upper limit management value to the rule set manager 460 (For example, 0.000 to 24.428 as shown in FIG. 6), as well as to select a confidence level and an enabled state of the corresponding function (refer to 332 ).

In addition, the rule-set management unit 460 sets the upper limit management value based on the Q-statistic calculated using the equation (6) by the principal component analysis unit 420 to the third rule set, and sets the third rule set to the communication unit 100, the Q-statistic calculated using Equation (5) may be compared with each other to determine whether the measured value signal inputted in real time through the communication unit 100 has a specific value.

On the other hand, the rule-set management unit 460 sets the upper limit management value based on the T²-statistic calculated by the equation (9) by the correlation analysis unit 430 to the fourth rule set, 100 to compare the T < 2 > -statistics calculated using Equation (3) to determine whether the measured value signal inputted in real time through the communication unit 100 has a specific value.

Further, the ruleset management unit 460 sets the management threshold value calculated using Equation 10 or the arbitrarily determined confidence level and Equation 11 by the single variable analyzing unit 440 to the fifth rule set, It is possible to compare the value of the measured value signal input through the communication unit 100 with the rule set and determine whether the measured value signal inputted in real time through the communication unit 100 has a specific value.

At this time, the detection unit 400 determines whether the singular values selected by the ruleset management unit 460 are included in the respective analysis results, such as the principal component number analysis unit 420, the correlation analysis unit 430, and the single variable analysis unit 440 The result of the abnormal data extraction in which the first rule set to the fifth rule set are all reflected can be derived and the grade can be set for the result of the extracted abnormal data, It is not limited. That is, the detecting unit 400 distinguishes abnormality of abnormal data, for example, 5 classes (NORMAL / WARNING / MINOR / MAJOR / CRITICAL) .

FIG. 7 is a flowchart illustrating an abnormal data detecting method according to an embodiment of the present invention. Referring to FIGS. 1 to 7, the abnormal data detecting method of the present invention will be described as follows.

First, a measurement value signal is received from a plurality of measuring instruments 10 through the communication unit 100 at predetermined intervals, for example, every five minutes (S100), and the received measurement value signal is transmitted to the data base 200) (S200). At this time, the measurement data stored in the database unit 200 may be data accumulated for several months to several years according to each measuring instrument 10, that is, the outlet.

Next, the abnormal data detection object and detection standard are inputted through the user interface unit 300 as shown in FIG. 2 (S300). At this time, the detection object selection through the user interface unit 300 can be performed before receiving the measurement value signal from the plurality of measuring instruments 10 through the communication unit 100, and accordingly, , GS Caltex Co., Ltd. You can set up to receive the measured value signal from the No. 1 chimney installed at the boiler and heating facility of the Yeosu plant. In addition, the user interface unit 300 provides a selection interface such as an analysis period setting in the selection of an abnormal data detection reference, a multi-step filtering operation according to the operation state for setting a ruleset after data analysis according to the present invention is performed, (330), and the evaluation (340).

Next, the detection unit 400 analyzes the measurement data stored in the database unit 200 using the abnormal data detection object and detection standard input through the user interface unit 300, and according to the result of analysis, The operation state of the facility in which the measuring instrument 10 is installed is judged, a specific value is selected for each determined operation state, and the determination result and the selection result are displayed through the user interface unit 300 (S400). The operation of the detector 400 will now be described in detail.

The detection unit 400 detects the measurement data (hereinafter, referred to as 5-minute data) received and stored at intervals of 5 minutes based on the correlation between a plurality of items related to the states of the plurality of substances and fluids measured by the plurality of measurement devices 10, Quot;). ≪ / RTI > For example, the detection unit 400 calculates statistical values, i.e., average, standard deviation, and distortion, for each measurement item such as oxygen concentration, flow rate, nitrogen oxide concentration, and exhaust gas temperature for pattern analysis, In order to prevent the clustering from being influenced by the measurement data of the item having a large value as the unit between the items is different, the measurement data is normalized according to each item, and an item intentionally puts a larger weight, for example, The weighting coefficient for processing the measurement data as a function of the distance weight before the clustering is set, and the number of the operating states (the pattern shown in FIG. 2 (See analysis item (322)), for example, K average ++ clustering analysis is performed by dividing into four sets of downtime, operation preparation, subtraction operation, The distribution range of the plurality of sets derived according to the clustering analysis execution result can be defined as the distribution range of the measurement data for each operating state. At this time, the detector 400 preferably provides the capability of analyzing measurement data including a measurement value of 100,000 or more per year within 10 minutes.

At this time, the detection unit 400 can display a screen visualized by color-categorizing the same state in the time series graph and the data distribution between items through the user interface unit 300. The user can utilize such visualization data to classify the operation state of the emission sorting, to track the long-term pattern change and generation period, and to select the target period for the rule set setting.

For example, FIGS. 8A to 8C show graphs provided by the detecting unit 400 through the user interface unit 200, which will be described below.

8A is a time-series graph of the flow rate (FL1) of 2015 received from the selected measuring instrument 10, showing an intermittent Pitot tube clogging phenomenon before the measurement of the flow velocity, Indicating that a persistent ideal eigenvalue occurs. 8B is a time series graph relating to the 2016 flow rate FL1 received from the selected measuring instrument 10 and shows a continuous error due to clogging of the pitot tube after a period in which normal measurement is impossible due to sample line leakage and clogging And a specific amount is generated.

FIG. 8C is a graph showing the distribution of data between items. FIG. 8C is a graph showing the distribution of data between items. FIG. 8C is a graph showing the distribution of data between items. FIG. - Oxygen concentration, carbon monoxide concentration (CO) - Oxygen concentration, exhaust gas flow rate (FL1) - oxygen concentration, exhaust gas temperature (TMP) - oxygen concentration, combustion chamber internal temperature (TM1) - oxygen concentration, fine dust concentration - And the like. At this time, it can be confirmed that the portion 323 corresponding to the flow rate shown in FIG. 8B is included in the correlation between the exhaust gas flow rate and the oxygen concentration, and it can be understood that the pitot tube plugging phenomenon has occurred.

9A is a time-series graph showing a specific value 324 determined as an abnormally high load section through data analysis for six months with respect to the discharged gas flow rate data, that is, determined as the upper management upper limit value. As described above, by using the principal component analysis, the T²-statistic, the Q-statistic, and the management limit value of the predetermined confidence level, it is possible to detect the singular value for each operation state and display the detection result as shown in FIG. 9A.

9B is a correlation plot showing two items, that is, a specific value (red) relating to the correlation between the flow rate and the oxygen concentration. The detection unit 400 detects a correlation The abnormal data according to the time series graph can be displayed.

The detection unit 400 sets the upper limit management value by the T²-statistic, the upper limit management value by the Q-statistic, the management limit of the preset trust level, etc. as a threshold value by applying the analysis results of various patterns, The threshold value can be applied as a ruleset that is a real-time detection level. For example, when a setting command 333 for applying a ruleset is input from the user interface unit 300 as shown in FIG. 6, the detecting unit 400 detects a T²- The upper limit management value based on the statistic is set as the threshold value, the threshold level applied to the real-time detection determined by referring to the threshold value, for example, 0.000 ~ 24.428 is displayed, the trust level of the function is displayed, Command can be input.

Thereafter, the detection unit 400 can determine the operation state based on a preset rule set for the real-time measurement value input through the communication unit 100, and select a specific value. For example, in the case where the oxygen concentration of 21% is set as the lower limit value indicating the shutdown state by data analysis, the detection unit 400 displays the portion exceeding the oxygen concentration of 21% in the time series graph in a separate color, It is possible to inform the user that the nitrogen oxide concentration is in a stopped state and when a nitrogen oxide concentration of 40% is set as a lower limit value by the data analysis, the portion exceeding 40% of the nitrogen oxide concentration in the time series graph is displayed in a separate color, It is possible to inform the user that the number of freezing prevention facilities is decreased.

10A to 10F are graphs showing the results of classification of the operating state of the detecting unit 400. The detecting unit 400 applies a threshold value set as a rule set by pattern analysis to real-time data (May 17, 2017) Usually, it can be divided into four operating states: excess air, high temperature, and shutdown. That is, the detection unit 400 classifies data in which the nitrogen oxide concentration and oxygen concentration are high in the real-time transmission data into the excess air condition, the data in which the flow rate and the carbon monoxide concentration are in the high concentration condition, Can be classified into a normal state. According to the results shown in Figs. 10A to 10F, although the generation of carbon monoxide is reduced by the complete combustion when the excess air is introduced, the concentration of nitrogen oxide increases with the increase of oxygen, and the carbon monoxide is increased by incomplete combustion at a high temperature It can be confirmed that the results consistent with the results of combustion engineering analysis are derived.

The abnormal data detection method according to the present invention can be implemented by a program and stored in a computer-readable recording medium (CD-ROM, RAM, ROM, floppy disk, hard disk, magneto-optical disk, etc.).

Although the disclosed method and apparatus have been described with reference to the embodiments shown in the drawings for illustrative purposes, those skilled in the art will appreciate that various modifications and equivalent embodiments are possible without departing from the scope of the present invention. I will understand that. Accordingly, the true scope of protection of the disclosed technology should be determined by the appended claims.

10: Measuring equipment
100:
200:
300: user interface section
400:
410: pattern analysis unit
420: Principal component analysis unit
430: correlation analysis section
440: Single variable analysis unit
450: Calibration conversion analysis section
460: Ruleset manager

Claims (15)

1. An abnormal data detection system for collecting and analyzing a measurement value signal from a plurality of measuring instruments for measuring a state of a plurality of materials and fluids contained in a fluid and outputting a measurement value signal,
A communication unit for receiving the measurement value signal from the plurality of measurement devices;
A data base for storing the measurement value signal received through the communication unit as measurement data;
A user interface unit for providing an interface for selection of an abnormal data detection target and a detection criterion and providing an interface for displaying a detection result; And
Wherein the control unit analyzes the measurement data stored in the database unit using the abnormal data detection object and the detection criterion selected through the user interface unit and determines the operation state of the facility equipped with the plurality of measurement apparatuses according to the result of the analysis, And displaying the determination result and the selection result as the detection result through the user interface unit,
Wherein:
A distribution range of measurement data for each operation state of the facility in which the plurality of measurement devices are installed with respect to the measurement data stored in the database unit and comparing the measured value signal inputted through the communication unit with a defined distribution range, And a pattern analyzer for determining the operation state of the facility in which the measuring instrument is installed,
The pattern analyzing unit,
A K average ++ clustering analysis is performed for dividing the measurement data into a set corresponding to the number of operating states, and the distribution range of a plurality of sets derived according to the K average ++ clustering analysis execution result is measured by the above- An abnormal data detection system that defines the range of data distribution.
delete delete The method according to claim 1,
Wherein:
A plurality of main items are selected by Principal Component Analysis among a plurality of items related to states of a plurality of substances and fluids measured by the measuring device, And a principal component analysis unit for calculating a T < 2 > -state for the stored measurement data and comparing the calculated T < 2 > -state with an upper limit management value based on the T & Data detection system.
The method according to claim 1,
Wherein:
A plurality of main items are selected by Principal Component Analysis among a plurality of items related to states of a plurality of substances and fluids measured by the measuring device, And a principal component analysis unit for calculating a Q-statistic for the stored measurement data and comparing the calculated Q-statistic with an upper limit management value based on the Q-statistic to select a specific value of the measurement data stored in the database unit Data detection system.
The method according to claim 1,
Wherein:
Selecting two items out of a plurality of items relating to the states of the plurality of substances and fluids measured by the measuring instrument, calculating a T < 2 > -state for the measurement data stored in the database portion concerning the selected two items, And a correlation analyzer for comparing the calculated T < 2 > - statistic with an upper limit management value based on the T < 2 > - statistic to select a specific value of the measurement data stored in the database.
The method according to claim 1,
Wherein:
And a control unit for selecting one item from among a plurality of items relating to the states of the plurality of substances and fluids measured by the measuring device and for comparing the measured data stored in the database unit with the average and standard And a single variable analyzing unit for calculating the deviation and selecting a specific value of the measurement data stored in the database unit by comparing the calculated average value with a management threshold value of a predetermined confidence level according to the calculated standard deviation Data detection system.
The method according to claim 1,
Wherein:
Calculating a pre-correction data based on a correction value applied to the measurement data in the measurement of the measurement device, comparing the difference between the measurement data and the pre-correction data with a preset value, And a correction-conversion-analysis unit for correcting the abnormal data.
The method according to claim 1,
Wherein:
And a control unit for setting a distribution range of measurement data for each operating state defined by the pattern analyzer as a first rule set and comparing the first rule set and a measured value signal inputted through the communication unit, And a rule set manager for determining a state of the abnormal data.
The method of claim 4,
Wherein:
A second rule set based on the T²-statistic of the principal component analysis unit as a second rule set, and a T²-statistic value calculated for the measured value signal inputted through the communication unit is compared with the second rule set, And a rule set management unit for determining whether the inputted measured value signal is a singular value or not.
The method of claim 5,
Wherein:
A third rule set is set as an upper limit management value based on the Q-statistic of the principal component analysis unit, and the Q-statistic calculated for the measured value signal inputted through the communication unit is compared with the third rule set, And a rule set management unit for determining whether the inputted measured value signal is a singular value or not.
The method of claim 6,
Wherein:
The upper limit management value based on the T < 2 > -state of the correlation analyzer is set to a fourth rule, and the T < 2 > -state calculated for the measured value signal inputted through the communication unit is compared with the fourth rule, And a rule set management unit for determining whether the inputted measured value signal is a singular value or not.
The method of claim 7,
Wherein:
Wherein the control unit sets the management limit value of the single variable analyzer as a fifth rule and compares the value of the measured value signal inputted through the communication unit with the fifth rule to determine whether the measured value signal inputted through the communication unit is a specific value And a rule set management unit that determines whether or not the abnormal data is detected.
An abnormal data detection method for collecting and analyzing a measurement value signal from a plurality of measuring instruments for measuring a state of a plurality of materials and fluids contained in a fluid and outputting a measurement value signal,
A first step of receiving the measurement value signal from the plurality of measurement devices;
A second step of storing the measured value signal received in the first step as measurement data;
A third step of receiving an abnormal data detection object and a detection standard; And
Analyzing the measurement data stored in the second step using the abnormal data detection object and the detection criterion input through the third step, and determining an operation state of the facility equipped with the plurality of measuring instruments according to the analyzed result A fourth step of selecting a specific value according to the determined operation state, and displaying a determination result and a selection result,
In the fourth step,
A K average ++ clustering analysis is performed for dividing the measurement data into a set corresponding to the number of operation states with respect to the measurement data while defining a distribution range of measurement data for each operation state of the facility in which the plurality of measurement devices are installed , K average ++ defines a distribution range of a plurality of sets derived according to a result of execution of the clustering analysis as a distribution range of the measurement data for each of the operating states and compares the received measurement value signal with the defined distribution range, An abnormal data detection method for determining the operation state of a facility in which two measuring devices are installed.
A computer-readable recording medium on which a program for executing the abnormal data detection method according to claim 14 is recorded.

Images